Table of contents
General Introduction ..................................................................................... 15
Objectives of the research ............................................................................. 35
Registration of third molar development: Influence of measurements versus
stages on age estimation ............................................................................... 41
Staged third molar development registration: Influence of the number of
stages on age estimation. .............................................................................. 53
Statistical modeling of third molar development data: Influence of regression
analyses versus Bayesian approach on age estimation ................................. 65
Collection and comparison of 13 country-specific third molar development
databases. ...................................................................................................... 79
Comparison of age estimation based on 13 country-specific third molar
development databases. ................................................................................ 93
Influence of tooth morphological age predictors on age estimation based on
third molar development. ............................................................................ 111
Influence of skeletal age predictors on age estimation based on third molar
development. ............................................................................................... 121
General discussion and conclusion ............................................................. 135
5
Jury
Promoter
Prof. Guy Willems
Co-promoters
Prof. Tore Solheim
Prof. Wim Van de Voorde
Chair
Prof. Dominique Declerck
Jury members
Prof. Werner Jacobs
Prof. Sigrid Kvaal
Prof. Helen Liversidge
Prof. Maria-Helena Smet
Prof. Sabine Tejpar
Prof. Dirk Vandermeulen
7
Dankwoord
Dank aan:
Prof. Dr. Guy Willems voor alle advies, steun, hulp, medeleven en vooral om de
ontstane vriendschap.
Prof. Dr. Tore Solheim en Prof. Dr. Wim Van de Voorde voor hun aandeel als
copromotor.
Prof. Dr. Dominique Declerck, Prof. Dr. Werner Jacobs, Prof. Dr. Sigrid Kvaal,
Prof. Dr. Helen Liversidge, Prof. Dr. Maria-Helena Smet, Prof. Dr. Sabine
Tejpar, Prof. Dr. Dirk Vandermeulen, voor hun werk als jurylid en de kritische
evaluatie van dit manuscript.
Dr. Steffen Fieuws voor zijn onmisbare begeleiding, die mij het gevoel gaf over
een extra copromotor te beschikken.
Iedereen, die wereldwijd, meehielp aan het verzamelen van data.
Alle “Master in forensic odontology”-studenten die ik begeleidde. Ieder van hen
werkte op haar of zijn manier inspirerend.
Ali om mij telkens op de hoogte te houden van nodige administratieve
regelingen.
De Katholieke Universiteit Leuven, omdat ik deel mocht uitmaken van haar
onderzoeksgroep.
Opgedragen aan:
Benoit, Els en Vincent voor alle tijd die ik niet aan hen kon spenderen.
Patrick Thevissen
Juli 2013
9
Abbreviations
BA
Be
Cervical vertebrae development registration technique
according Baccetti et al. (2005)
Belgium
Br
Brazil
CA
CEJW
Tooth development registration technique according
Cameriere et al. (2008)
Cervical vertebrae development registration technique
according Caldas et al. (2007, 2010)
Tooth length from occlusal plane till cement enamel
junction
Crown width at cement enamel junction
Ch
China
CT
Computed tomography
CW
Maximal crown width
DM
DS
Tooth development registration technique according
Demirjian et al. (1973)
Developmental score
DTMD
Degree of third molars development
F
Female
GK
In
Tooth development registration technique according
Gustafson and Koch (1974)
Tooth development registration technique according
Haavikko (1974)
Tooth development registration technique according
Harris and Nortje (1984)
South-India
It
Italy
Ja
Japan
KO
Tooth development registration technique according Köhler
et al. (1994)
Korea
CAL
CEJL
HA
HN
Ko
KU
L
Tooth development registration technique according
Kullman et al. (1992)
Mean value of length ratios according Kvaal et al. (1994)
11
Abbreviations
LL
Lower left
LR
Lower right
M
Male
M
K
Mean value of all ratios according Kvaal et al. (1994)
Ma
Malaysia
MO
Tooth development registration technique according
Moorrees et al. (1963)
Magnetic resonance imaging
MRI
OPL
PC
PHL
Tooth length from occlusal plane till most apical calcified
tooth point
The score on the first principal component
Po
Tooth length from occlusal plane till most occlusal pulp
horn point
Poland
R1
TTL48/TTL47
R2
OTL48/OTL47
R3
R348/R347
R347
R3
48
R447
48
TTL/CEJW on second molar
TTL/CEJW on third molar
R548/R547
R5
R547
R5
TTL/CW on third molar
R448/R447
R4
R4
TTL/CW on second molar
48
TTL/PHL on second molar
TTL/PHL on third molar
R6
R648/R647
R647
TTL/CEJL on second molar
R648
TTL/CEJL on third molar
R7
48
RA
TTL48/OPL48
Sa
Tooth development registration technique according
Raungpaka (1988)
Mandibular development registration technique according
Rai et al. (2008)
Saudi-Arabia
SE
Cervical vertebrae development registration technique
RAI
12
Abbreviations
according Seedat and Forsberg (2005)
Th
Thailand
TTL
Tu
Tooth length from most occlusal till most apical calcified
tooth point
Turkey
Ua
United Arab Emirates
UL
Upper left
UR
Upper right
W
Mean value of width ratios according Kvaal et al. (1994)
13
Chapter 1
General Introduction
15
General introduction
AGE
The knowledge and the proof of age are indispensable and invaluable for all
human beings. The human age is a measure, mostly expressed in years, of
the period lived since birth, so the indisputable registration of the date of
birth is essential to determine an individual’s chronological age. Age can
also be a measure of the biological, psychological and social changes that
occur in the course of a lifetime.
Age, together with the name, nationality and gender, enables to
one’s identity to be established and verified. Attaining specific ages gives
access to certain rights or interests by law and authorizes the administration
of sanctions and penalties for the violation of laws and the breaking of
agreements. Although the precise ages at which these thresholds are set may
vary from country to country, they set the legal rhythms of the individual’s
life the world over.
During a lifetime, age thresholds giving access to certain privileges
or incurring specific obligations need to be respected for legal,
socioeconomic, religious, cultural and professional reasons. Forensic ageestimations are requested in relation to age thresholds in criminal
investigations, during immigration procedures, and for civil purposes. As
regards crime, an individual is considered capable of committing a crime
when the legally set age threshold for criminal responsibility is attained,
generally in childhood. The age of majority is the legally defined moment
when a person is legally no longer considered a minor. The protected status
regarding care, health and education provided to a child changes for adult
status in which full control over one’s own person, decisions and actions is
assumed (Massoglia and Uggen, 2010). The loss of protected child status at
the age of majority has consequences for the migration and asylum policies
applied by the countries to which refugees flee (Kalt et al., 2013). Children
and the elderly are the most vulnerable age groups in a society. Certain civil
rights depend on age, such as children being protected from early
employment and forced marriages and the elderly being guaranteed care
(Hoogeveen et al., 2005).
The death of a person is officially registered and a death certificate
is issued that provides information about the identity of the deceased and the
circumstances and the cause of death (NCHS, 2003). The inclusion of the
date and time of death (and thus the age of the deceased) is required for the
validity of the document. The certificate enables family members to
commence closure and settle the estate.
17
General introduction
INCIDENCE FOR AGE ESTIMATION
The legal consequences related to an individual’s particular age make her or
him a subject for forensic age estimations when no or dubious age
documentation is available.
Of concern here are individuals for whom no registered age
documentation is available and also unidentified human remains. In Western
societies, the indirect registration of birth was introduced in the 16th century
with the recording of the date of baptism in parish registers. Articles 7 and 8
of ‘The Convention on the Rights of the Child’ (Resolution 44/25, 1989)
stipulate that every child has the right to be registered at birth by the state
within whose jurisdiction the child is born. The registration of the place, date
and time of birth makes a birth certificate universally accepted proof of an
individual’s age. Despite these regulations and rights, approximately1/3 of
all births are still not registered worldwide (Cody and Plan, 2009).
Moreover, in areas of armed conflict, registrations are usually temporary,
often destroyed and frequently suspended (UNICEF, 2007). According to the
United States Bureau of Justice Statistics, in an average year in the United
States, 4,400 unidentified human remains are reported (Hickman et al.,
2007). These numbers increase exponentially in areas where disasters have
occurred. In fact, worldwide more than 30,000 people were killed in
disasters in 2011 (Guha-Sapir et al., 2012).
When the age of juveniles is unknown or no legally supported
documentation or incorrect documentation is suspected, age estimation
examinations are requested: to classify the suspects by the age of criminal
responsibility in criminal procedures (Schmeling et al., 2001), to determine
the age of unaccompanied young applicants in migration and asylum
procedures, to place the participant in the correct age group in sports
(Engebretsen et al., 2010), to provide children with a birth date in adoption
procedures (Crossner and Mansfeld, 1983), to protect children who are
treated as adults and even exploited by adults (Smith and Brownlees, 2011),
and to determine retirement benefits and pensions for the elderly (RitzTimme et al., 2002). In 2010, the Member States of the European Union
received 30,200 asylum applications of which 12,230 concerned
unaccompanied minors (Bulletin, 2012).
The high number of individuals requiring forensic-age estimations
makes it one of the major specialties in forensic odontology.
AGE ESTIMATION METHODS
The age of an individual is estimated on the basis of the conversion of agerelated markers to chronological age. The age-related markers are variables
in biological, psychological and social appearance that are present during or
at distinct periods in human life. Samples allowing one to observe, in a
specific age interval, the age-related variables of concern are collected from
18
General introduction
human populations. In these samples, the observed status of the age-related
variables are quantified and registered. The extracted data are analysed
statistically to identify significant population-dependent and age-related
features. The resulting information enables one to construct atlases, tables
and models that permit age predictions for subjects of unknown age of the
same population. The status of the corresponding age-related variable
observed in the subject is matched with the information on the table or atlas
or integrated into the constructed model. The estimated age value consists of
an age outcome and a measure of the prediction uncertainty. To validate the
accuracy of age estimates, the difference between the estimated and the true
age needs to be quantified by means of a test sample.
Human biological age-related variables are defined as human body
parts that change in function of age. The age predicting value of the
variables considered depends on its correlation with age. In a human body,
the optimal age-related variables have been detected in the skeleton and
classified in a bone and a dental group.
DENTAL AGE ESTIMATION
A Medline/Pubmed search for dental age estimation studies was performed
using the search string: ‘(age determination by teeth) OR (dental age
estimation methods) OR (dental age estimation assessment) OR (dental age
estimation calculation) OR (radiological dental age estimation) OR (forensic
dental age estimation) OR (timing tooth formation)’. The search resulted,
before further manual sorting, in 1,517 (02 March 2013) published studies
mainly on dental age estimation methods, modifications to the methods and
their validation results on specific populations. The search revealed that
many dental age estimation methods have been developed. In forensic age
estimation assessments, the choice of a particular estimation method depends
upon the age-related dental variables available in the material at hand.
Accordingly, classifications of the estimation methods are based on the
technique used to observe the dental variables or on the dental feature
examined in the observed variable. The age-related dental variables are
observed clinically by means of medical imaging techniques, after tooth
extraction or after tooth destruction. The method used must be ethical and
legal. For example, to estimate the age of living persons, it is ethically
unacceptable to extract teeth, and the law may prohibit using ionizing
techniques.
The principal age-related features observed in the dental variables
are based on changes in the morphology or the development of the tooth and
changes in the biochemistry of tooth material [Fig.1].
19
General introduction
Figure 1: Classification of age estimation methods based on age-related dental
variables
Classifying age estimation methods according to biochemical, morphological, or
developmental characteristics permits a parallel classification related to the
proper age category of the individual. It also indicates the tooth types or each
specific method. On a time line of a human lifetime, starting at conception and
ending with death, the figure gives the applicability of the dental age estimation
methods. During life, age estimation methods based on tooth biochemistry are
appropriate for all teeth, and methods based on tooth morphology are suitable for
all permanent teeth in adults. Methods based on tooth development are divided
into two groups: deciduous and all permanent teeth are considered in children,
and the late third molar development is assessed in sub-adults.
Tooth biochemistry
With ageing, an increasing transition from levorotatory to dextrorotatory
enantiomers of aspartic acid was detected in tooth materials. In particular,
this is found in dentin (Helfman and Bada, 1976), enamel (Griffin et al.,
2008b), cementum (Ohtani, 1995b) as well as whole tooth samples (Sakuma
et al., 2012) in both, permanent and deciduous teeth (Ohtani, 1994; Ohtani et
al., 2005). This phenomenon has been described as aspartic acid
racemization, and it enables one to estimate the age at death of an individual
(Waite et al., 1999). In adults, the method provides highly accurate age
estimates (standard deviation 3 years) (Ohtani, 1995a). Although the method
is applicable to living individuals, the inherent separation of tooth material
remains an ethical obstacle. The position of enamel enables one to collect
tooth material with very little damage to the tooth. An etching technique
20
General introduction
developed to extract amino acids from enamel surfaces is minimally tooth
destructive and promises to be applicable to the living (Griffin et al., 2008a).
The detection of radioactive carbon concentrations in teeth has been
proposed as age estimation method. The radiocarbon age estimation method
is based on the comparison of the increases over time of global carbon-14
levels due to the aboveground testing of nuclear weapons between 1955 and
1963 with the measured proportion of carbon-14 incorporated into the
enamel separated from the teeth of the examined individual. The comparison
provides an estimate of the date of birth (Spalding et al., 2005). The method
allows one to measure the carbon-14 concentration in the incisal and cervical
enamel of a tooth. From these measurements, one can derive the formation
period of the crowns of different tooth types. The resulting information
permits one to choose the correct side of the peak on the bomb fallout curve
(Kondo-Nakamura et al., 2011). The individual’s birth date is estimated with
an average absolute date of birth prediction error between 1.3 and 1.9 years
(Alkass et al., 2010a, b).
Tooth morphology
Dental age estimation methods based on changes in tooth morphology are
based mainly on observed transformations in the anatomy of dental
structures due to attrition, the formation of secondary dentin, root resorption,
root translucency, cementum apposition and the attachment of the
periodontal ligament [Fig.2]. The gradual morphological transformations
observed in the variables are classified and registered according to staging
systems (Gustafson, 1950; Dalitz, 1962; Johanson, 1971). Because the
attrition and the attachment of the periodontal ligament are clinically visible
and because secondary dentine apposition can be observed by means of
Figure 2: Principal age-related variables, based on tooth morphology.
The
The gradual
gradual transformations
transformations in
in tooth
tooth anatomy
anatomy as
as aa function
function of
of age
age are
are illustrated
illustrated
for each of the main morphological variables. The parameters increase with
advancing age except for the periodontal ligament attachment position, which
moves towards the root apex [figure
fromfrom
(Johanson,
1971)]. 1971)].
[figureadapted
adapted
(Johanson,
21
General introduction
medical imaging, these variables can be used to estimate the age of living
persons. The other variables are observed after tooth extraction and,
occasionally, tooth sectioning. Therefore, they are legally and ethically used
to estimate the age only of deceased individuals.
Tooth sectioning is mainly carried out on a mid-tooth level in the
long axis and in bucco-lingual direction (Solheim, 1984). Morphological
dental age estimation methods use all of these variables (Gustafson, 1950;
Johanson, 1971; Maples, 1978), some of them (Dalitz, 1962; Maples, 1978;
Lamendin et al., 1992) or only one (Bang and Ramm, 1970; Lorentsen and
Solheim, 1989; Solheim, 1990; Solheim, 1992a, b; Kim et al., 2000). Certain
variables need to be observed on intact and/or sectioned teeth.
In addition to stage registrations, specific measurements enable the
status of the observed variable to be registered (Lamendin et al., 1992).
Staged and measured registrations of the described parameters are combined
with supplementary variables such as tooth colour (Solheim, 1988b) and
dental root surface structure (Solheim, 1993). In the cementum formed at the
mid-third of the root, alternating dark and translucent layers are visible
microscopically. Each pair of lines corresponds to 1 year of life, so counting
these cementum annulations provided an age estimation (Kvaal and Solheim,
1995; Kagerer and Grupe, 2001), but specialized tooth cutting techniques are
required (Maat et al., 2006).
Due to the apposition of secondary dentine, the volume within the
pulp chamber decreases with ageing, which can be observed by means of
medical imaging. Ratios from the length and width measurements of pulp
and tooth observed on periapical radiographs (Kvaal et al., 1995) allow one
to register the changes in the pulp volume and to estimate age with
regression formulas. The technique was validated for panoramic radiographs
(Bosmans et al., 2005; Paewinsky et al., 2005; Landa et al.,2009; Erbudak et
al., 2012) and modified with tooth and pulp area information (Cameriere et
al., 2004). Direct registration of tooth and pulp volumes can be done with
microfocus computed tomography (CT), CT and cone beam CT images,
which are at the basis of each of these techniques. Specific age estimation
methods based on pulp volume changes have been reported (Vandevoort et
al., 2004; Yang et al., 2006; Someda et al., 2009 Star et al., 2011). The agerelated value of the variable depends on the tooth type and the tooth position.
Therefore, age estimation methods specify whether they are applicable for a
specific tooth, for all teeth, or for a particular group of teeth. The
morphological age estimation methods allow one to estimate the age of
adults with a precision expressed in the standard error of the estimate
between 4 and 12 years (Ritz-Timme et al., 2000).
An important morphological change in the tooth enamel appears 3 to
4 days after birth as a hypo-mineralized step-like rupture in the enamel
matrix. This is the ‘neonatal line’ (Rushton, 1933) that separates the smaller
enamel prisms in the postnatal enamel from the prenatal prisms. This line is
22
General introduction
observed microscopically (Sabel et al., 2008) in thin section preparations
(70-150µm thick) of teeth that had crowns maturing around the time of birth
(deciduous teeth and the first permanent molars) (Zanolli et al., 2011). The
neonatal line provides forensic evidence to determine whether a child was
born dead or had lived at least a few days. The neonatal line is actually a
modified incremental enamel growth line (Boyde, 1997) and is called the
‘line of Retzius’ (Risnes, 1987). In humans, the formation of enamel starts at
the dentin and moves 2 to 8 µ a day, so counting these lines and measuring
the enamel thickness can be used to estimate human age (Dean, 2010).
Tooth development
The development of teeth involves the formation of an organic matrix and its
subsequent calcification or mineralization (Smith, 1991). The formation
sequence follows a chronological pattern and starts on the sixth to seventh
intra-uterine week with the development of buds in the dental lamina (Kraus
and Jordan, 1965; Tonge, 1969). The buds differentiate in caps and bells and
form crypts in the jaw bone. In the crypt, calcification and mineralization of
the future crown cusp tips or incisal tooth edge proceed continuously layer
by layer until the apical root ends of the developing teeth are closed (Massler
and Schour, 1946; Calonius et al., 1970). Medical imaging can specify the
mineralization process in the living. To quantify the timing of this
maturation sequence, the chronological age at which a tooth developed into
an arbitrarily chosen stage is registered. The threshold between stages are
reflected in specific tooth traits (such as calcification commencement, crown
completion, root completion), or the root and tooth fractions used to divide
the maturing process in consecutive steps [Fig.4, Chap. 3]. A particular stage
during the tooth maturation process is the eruption of the tooth through the
alveolar bone and gums into the mouth. The maturation quantification is
used for age estimations based on a certain tooth type, a specific tooth
position or a group of teeth. The age estimation methods based on tooth
development are generally classified as methods based on third molar
development and methods based on the development of all the other teeth
[Fig.1].
Development of all teeth except third molars
Since all teeth except the third molars mature on average from the sixth
week of intra-uterine life (Kraus and Jordan, 1965) until the age of 14
(Liversidge and Marsden, 2010) to 16.5 years (AlQahtani et al., 2010), their
age-related development is used to estimate age in children. For this purpose,
two methods have been described.
First, in atlases, pictorial charts and tables, a specific stage
representing a degree of development of a single tooth, a part of the dentition
or the whole dentition is illustrated for specific ages. The developmental
stage of the corresponding tooth or teeth is compared and matched with the
23
General introduction
nearest best-fit diagram and related to age. The atlases are constructed based
on dissection data (Logan and Kronfeld, 1933), radiographic imaging studies
(Schour and Massler, 1941; Kahl and Schwarze, 1988) or archaeological
material (Ubelaker, 1978). In addition, specific tooth traits are correlated
with the mean age of appearance for each tooth related, and a measure of
their variability is given in tables (Gustafson and Koch, 1974). Particular
interest is assigned to the timing of tooth eruption. Evaluating tooth eruption
considers the tooth position in relation to the surrounding structures. The
alveolar crest (Haavikko, 1970), the oral mucosa (Ando et al., 1965), the
inferior border of the mandible (Shumaker and El Hadary, 1960), other teeth
(Schopf, 1970), and the occlusal plane (Bengston, 1935) are examples of the
reference points used. They have been evaluated clinically and radiologically
(Mattila and Haavikko, 1969). However, the different definitions of tooth
eruption used rules out comparison (Hurme, 1949). The wide variations in
registered mean eruption times (Leroy et al, 2008) are due to the varied
definitions of tooth eruption and the lack of registration of the time between
the time of tooth eruption and the time of observation.
Second, the tooth maturing process is arbitrarily divided into
successive developmental stages, the number and the duration of which
differ in function of the specific staging technique used. The thresholds
between the stages need to be well described and refer to observable
anatomical tooth traits (Raungpaka, 1988), the proportions of the future
crown and root lengths (Moorrees et al., 1963) or the size of already
developed parts of the tooth (Demirjian et al., 1973). Referral age standards
are collected in tables for each of the stages and related to single tooth types
or tooth positions (Moorrees et al., 1963; Prahl-Anderson and van der
Linder, 1972; Anderson et al,, 1976).
The degree of dental maturation of a child can be related to
chronological age and used for age estimations. To do this, the
developmental stage of a particular tooth or a group of teeth is assessed.
Considering a specific tooth, the observed developmental tooth stage is
related to a tabulated corresponding age. The tables are generally established
by averaging the chronological age of sampled subjects sorted by stage
(Gleiser and Hunt, 1955). For a group of teeth, a sex-specific weighted
maturity score is given to each observed tooth. The technique was adapted
from skeletal maturity assessments in which the stages of hand-wrist bone
development were used (Tanner et al., 1962). The sum of the weighted
maturity scores is related to chronological age in percentile curves or score
tables. The seven left mandibular permanent teeth (without the third molar)
are considered (Demirjian et al., 1973), and the method has been adapted for
use of four teeth (first and second left mandibular premolar and molar)
(Demirjian and Goldstein, 1976) and eight teeth (Acharya, 2011b). To obtain
age estimates as a function of dental maturity, the method was revisited
using polynomial functions (Chaillet et al., 2004a, b). With the use of a
24
General introduction
weighted analysis of variance, the seven-tooth technique has been modified
to provide sex-specific tables to convert the observed developmental stage of
each tooth immediately into an age value. The estimated chronological age is
the sum of the seven age values (Willems et al., 2001).The reliability of ageestimation methods based on the development of permanent teeth except the
third molars is reported with a 95% confidence interval from ±0.65 for early
tooth stages to ±2.59 years for late tooth stages (Liversidge et al., 2010).
Development of third molars
The third molars are the only maturing teeth remaining in the age range
between 16 and 23 years. On average, the crypt formation of the third molar
is observed on panoramic radiographs during the 8th year (Liversidge,
2008b), so late third-molar development is used to estimate age in this subadult age range [Fig.1]. In analogy with age estimation methods based on the
development of all of the permanent teeth without the third molars, age
estimates based on the third molars are divided into methodologies using
atlases or tables and those based on the observed succeeding stages in the
third-molar development process. Most atlases constructed in function of
tooth development include third-molar development and its relation to age
(Schour and Massler, 1941; Ubelaker, 1978; AlQahtani et al., 2010; Blenkin
and Taylor, 2012). Third-molar eruption is a particular developmental stage
studied for age-estimation purposes (Schmeling et al., 2010; Wedl and
Friedrich, 2005; Caldas et al., 2012). Of all of the teeth, the third molars
have the widest variation in eruption time and, when validated for age
estimation purposes, it was found to provide overestimations of as much as
seven years (Scheurer et al., 2011).
The arbitrarily constructed tooth development staging techniques are
applicable to the third-molar development process. The technique is applied
to longitudinal (Moorrees et al., 1963; Anderson et al., 1976) and mostly
retrospective and cross-sectional collected reference samples to establish
tables or models to estimate age. The origin, registration and sampling of the
reference data indicate the applicability of the related tables and models for a
specific case. Therefore, the specific third-molar position considered (Nortjé,
1983), the third molar jaw position (Moorrees et al., 1963) or the third molar
mouth side (Kullman, 1992) all have to match. Furthermore, the number of
sampled subjects, the geographical, biological and ethnic origin of the
subjects, their considered age range and age distribution, the integration of
sex differences, the third-molar development registration technique used,
and the time of sampling need to be taken into account in choosing the most
appropriate approach [Table1].
Forensic application
Most of the forensic dental-age estimations need to be performed in the
context of migration and asylum procedures. Based on the children's rights
25
General introduction
Table 1: Dental age estimation studies based on third molar development, part 1/6
26
General introduction
Table 1: Dental age estimation studies based on third molar development, part 2/6
27
General introduction
Table 1: Dental age estimation studies based on third molar development, part 3/6
28
General introduction
Table 1: Dental age estimation studies based on third molar development, part 4/6
29
General introduction
Table 1: Dental age estimation studies based on third molar development, part 5/6
30
General introduction
Table 1: Dental age estimation studies based on third molar development, part 6/6
31
General introduction
(UN Resolution 44/25, 1989), unaccompanied immigrating children are
protected. Legally a person is considered to be a child as long as the age of
maturity is not reached. For immigrants, the age of onset of maturity as
defined in the country of arrival has to be considered. The authorities of the
countries in which immigration is requested are entitled to check the age of
the applicant for which medical-age estimation tests are used. Different
examination protocols for these tests (Schmeling et al., 2006; Olze et al.,
2006a; Solheim and Vonen, 2006; Schmeling et al., 2007; Nuzzolese and Di
Vella, 2008; Nelki and Bailey, 2010; Aynley–Green, 2011), and guidelines
for quality control have been described (IOFOS, 2008). No consensus on a
common practice for age-estimation examination has been reached
internationally and often not even on the national level. As an example, the
age-examination protocol for unaccompanied young refugees developed at
the Katholieke Universiteit Leuven (KULeuven) and applied in Belgium is
described. Because the protocol integrates at least three different tests, it is
called the ‘Triple Test’.
In Belgium, any governmental agency or person can report young
unaccompanied refugees without a residence permit to the Guardianship
Service in order to grant them protection. The Guardianship Service, which
is part the Federal Public Service for Justice, has the mission to ensure
judicial protection of all unescorted minors (asylum seeker or not) staying or
arriving in Belgium. The initial task of the Guardianship Service is to
identify if an applicant can qualify as a young unaccompanied refugee: the
service verifies if the applicant is less than 18 years of age, not accompanied
by a person who has parental authority, originates from a country outside the
European Community and resides in Belgium as an asylum seeker or without
a residence permit. If the age of the applicant is unknown or if there is doubt
about the alleged age, Belgian law prescribes (Article 7 of the Guardianship
Act (Wetgeving, 2002)) that the Guardianship Service orders a medical test
to verify whether or not the person is younger than 18 years old. Doubt about
the age given by the refugee rises from possibly interpreted conversations
with the applicant, analysis of his or her identification documents and
opinions of social workers, the centres they temporarily stay in, or his or her
guardian. The Guardian Service provides all applicants identified as minors
with proper reception and housing, social protection, legal assistance and a
Belgium guardian.
The Triple Test combines at least three medical tests and is mainly
based on dental-age estimation. Because each test considers other biological
variables, different age estimates with their associated levels of uncertainty
are obtained. Multiple test results increase the accuracy of the estimated age,
expand the age range possible, and can confirm the test results. The
biological variance between persons gives rise to scientifically unexplainable
discrepancies between the test results. When there is doubt about the
estimated age, Belgian law prescribes that the medical test delivering the
32
General introduction
youngest age result prevails (Wetgeving, 2002). The Triple Test is
performed after obtaining informed consent of the applicant.
To exclude diseases or syndromes that could affect tooth and
skeletal development, a clinical dental examination is performed. The
objective is to provide a clinical impression of the dental age of the
applicant. The number of teeth, the amount of decay, stain, and restorations,
the positions of the periodontal attachments, the degree of attrition,
especially in molars, and the dental occlusion are evaluated. The clinical
impression obtained by the dentist provides a reasonably good estimation of
whether the applicant is younger or older than 18 years old (Solheim and
Vonen, 2006). The examiner who registers the clinical impression may be
biased by seeing and clinically examining the applicant. Therefore, the other
parts of the Triple Test are also performed by another, independent
examiner. If the results of the two examiners disagree, the tests are
reconsidered until a consensus is reached.
A dental panoramic radiograph is then taken and evaluated. If
developing permanent teeth (except third molars) are observed, the age is
estimated in function of the registered developmental stages of the
mandibular left permanent teeth using the method of Willems et al. (2001).
If all the permanent teeth (except the third molars) are mature, the age is
estimated based on the registered developmental stages of the available third
Figure 3: Chart relating the KULeuven Triple Test outcomes and the age of
maturity threshold of 18 years.
For each of the radiographic tests (*), the mean age providing detect age-related
information in the observed variables is listed for females and males separately.
In females, complete ossification of all hand-wrist bones provides no information
on whether or not the subject is 18 years. When the wisdom teeth are mature, only
the medial ossification centers of the clavicles can provide age-related
information (up to a mean age of 26.7 years for both sexes).
F: female, M: Male, y: years
33
General introduction
molars with the use of the method described by Gunst et al. (2003). This
method also allows one to calculate the probability of an applicant being
older or younger than 18 years old in the event of full third-molar
development. In addition to the panoramic radiograph, a hand-wrist
radiograph of the non-handedness side is used to verify the dental test result.
The ossification of the hand-wrist bones, in particular the ossification of the
radius and ulna, is determined by means of the atlas of Greulich and Pyle
(1959).
If the hand-wrist bones are mature, sterno-clavicular radiographs
(frontal and oblique) are taken to determine the ossification of the medial
part of the clavicles. Accordingly, the age is estimated with the method of
Schmeling et al. (2004). The evaluation of the clavicles allows one to
estimate an age even when all the available third molars are completely
mature [Fig.3].
The Triple Test Protocol was adapted in function of the research
outcomes obtained in the current thesis. The final version is described in
Chapter 10 below: ‘General Discussion and Conclusion’
34
Chapter 2
Objectives of the research
35
Research objectives
GENERAL RESEARCH AIM
Increasing global human migration raises management concerns in the
countries where immigrants arrive. A special protective status is given to
immigrating unaccompanied children (Abbing, 2011). Therefore, most states
require specialized medical investigations to obtain proof of the age of
unaccompanied youngsters who have no official identification documents
and claim to be minors (Solheim and Vonen, 2006). Dental age estimation in
this sub-adult age group relies on the only dental age predictor available,
namely the developing third molars (Gunst et al., 2003; AlQahtani et al.,
2010; Liversidge and Marsden, 2010). However, adequately validated age
estimation methods and suitable population-specific reference databases are
lacking. Accordingly, there is a practical inability to perform scientifically
correct dental age estimations in sub-adults, especially those from distant
countries and of diverse ethnicities. The general research aim is to optimize
dental age estimation based on third molar development.
To achieve this, the problems in age estimations based on third
molar development were defined and related hypotheses formulated and
tested in specially designed research setups.
Panoramic radiographs were sampled retrospectively and crosssectionally and data registering third molar development were collected.
Two registration techniques were used. First, the sequence of third molar
development was divided into succeeding stages, and the observed third
molar development was classified in the corresponding stage and registered
accordingly (Gleiser and Hunt, 1955). Second, during its maturation, the
dimensions of the third molar increase and the third molar sizes were
registered (Israel and Lewis, 1971). The former technique provided ordinal,
the latter continuous data. To obtain optimal age estimations, the third molar
development registration technique providing the best age prediction
performances has to be determined.
Research Hypothesis 1:
The third molar development registration technique measuring third molar
development provides better age estimation than does the staging technique.
Multiple tooth development staging techniques were examined in
function of described borderlines between the succeeding stages.
Consequently, the number of stages covering the third molar development
process differs between techniques. Therefore, it has to be determined if the
number of stages used in a staging technique affects the age prediction.
37
Research objectives
Research Hypothesis 2:
The number of stages used in the third molar development staging technique
influences the age predictions.
The classic approach for age estimation uses regression analysis to
model the collected data. Several drawbacks of this technique are examined:
the age distribution of the residuals, the high correlation between the
independent variables, often observed missing values of the independent
variables, and systematic biases in the age predictions. Therefore, in the
current study, a Bayesian approach of age estimation is established on third
molar development. The age prediction performances of both approaches are
compared.
Research Hypothesis 3:
A Bayesian approach using third molar development provides more accurate
age predictions than does the classic regression analyses in sub-adults.
In forensics, sub-adult age estimations are requested primarily to
discriminate a child from an adult during migration and asylum procedures.
Due to the migration aspect, the age of an applicant with a particular
geographical and biologic origin was frequently estimated using methods or
models developed from a reference sample that includes subjects of varying
origin. Since dental age estimates in the sub-adult group are based on third
molar development, it has to be determined if there are differences in third
molar development between populations of different geographic and
biological origin. Therefore, third molar development in uniformly and
country-specific collected samples are analysed and compared.
Research Hypothesis 4:
There are differences in third molar development between country-specific
sub-adult populations.
Age estimation models developed from a particular reference sample were
validated for their age prediction performances using a specific validation
sample. The impact on the age prediction using a validation sample from a
different geographic and biological origin as the reference sample was not
examined for age estimation models based on third molar development.
Research Hypothesis 5:
The statistical model established on a Belgian reference sample is the most
appropriate for dental age estimation in unaccompanied minors.
38
Research objectives
Research Hypothesis 6:
The statistical model established on pooled country-specific reference
samples renders, in the absence of a model constructed on a country-specific
reference sample, the most accurate dental age estimation in sub-adults.
The age prediction performance of age estimation models constructed on a
single age-related variable may be improved by adding age-related
information of one or more variables present in the considered period of life.
Therefore, reference samples registering a specific moment third molar
development as well as other age-related variables were compiled, modelled
and analysed.
Research Hypothesis 7:
In sub-adults, the accuracy of age estimations based on third molar
development is improved by adding age-related information from tooth
morphological age predictors.
Research Hypothesis 8:
In sub-adults, the accuracy of age estimations based on third molar
development is improved by adding age-related information from cervical
vertebrae maturation.
TESTING RESEARCH HYPOTHESES
To test the research hypotheses, panoramic radiographs from an equal
number of female (F) and male (M) subjects were examined retrospectively
to determine the presence of at least one third molar and the absence of third
molar pathologies. The age range selected for each sample was chosen in
function of the ages of interest in the specific hypothesis evaluated and
truncated to provide minimal bias in the study outcomes. The subjects were
homogeneously distributed in age. In particular, within the collected age
range, an equal number of subjects were randomly selected for each age
category of one year. Only one radiograph per individual was included to
avoid integration of similar information in the samples. The subjects’ birth
date, gender, nationality, and biological group were verified and registered
together with the date of the radiographic exposure. The image quality of the
radiographs permitted clear observation of the teeth present and, in
particular, of the third molars. The images were obtained in digital format or
digitized and analysed using photo ameliorating software. For testing
Hypothesis 8 lateral cephalograms were also collected.
The collected samples were placed in a reference group in order to
construct age prediction models and a test group in order to prove specific
features. The majority of the samples were reference samples collected from
39
Research objectives
country-specific populations. From these samples, databases were assembled
of information on the developing third molars. The extracted data were
statistically analysed and modelled.
The test samples were used, first, to select the data registration
technique and the data modelling procedure providing the most accurate age
predictions (Research Hypotheses 1-3); second, to validate the model
outcomes and to compare especially the country-specific model outcomes
(Research Hypotheses 4-6); third, to evaluate the effect on age prediction
when combining age-related information from the developing third molars
with age-related information from other variables (Research Hypotheses 78).
In Chapter 5 was detected that a Bayesian approach was the optimal
methodology to model the third molars development. Because this model
assumed dependence between the variables, it reached a high level of
complexity. As such the number of integrated variables was limited to the
four third molars. Hence the constructed model could not be used to test
research hypotheses in which variables were added to the third molars
information (Research Hypotheses 1, 7 and 8). Moreover, the complexity of
the constructed model was also increased using a higher number of third
molar stages. For that reason, it was not applied to test Research Hypothesis
2.
40
Chapter 3:
Registration of third molar
development: Influence of
measurements versus stages
on age estimation
THIS CHAPTER IS BASED ON THE FOLLOWING MANUSCRIPT.
Human third molar development: measurements versus scores as age predictor
Thevissen PW, Fieuws S, Willems G
Published in Archives of Oral Biology
2011 56(10):1035-40
Oral presentation at the annual scientific meeting of the American Academy of
Forensic Sciences, Chicago, 2011
TESTING RESEARCH HYPOTHESIS 1:
The third molar development registration technique measuring third molar
development provides better age estimation performances than does the staging
technique
41
Measurements versus stages
INTRODUCTION
To observe third molar development, panoramic radiographs were
retrospectively and cross-sectional selected. From these radiographs, data
registering the observed third molar development were collected. A major
issue in this data collection is the choice of registration technique used. On
the one hand, the sequence of third molar development is divided into
succeeding stages, and the third molar development is classified in the
Figure 4: Tooth development staging technique constructed by Gleiser and
Hunt (1955), modified by Köhler et al. (1994) (KO).
The degree of development reached at each stage threshold is schematically
illustrated. In the Köhler et al. (1994) tooth staging technique (KO), the tooth
development sequence is divided into 10 stages. ‘Crown complete’, ‘Root initial’
and ‘Apex complete’are based on objective anatomical descriptions. The other
stages depend on subjective predictions of unknown tooth dimensions. Each
developmental stage is related to a corresponding score from 1 to 10, starting
with ‘Crown ½ formed’ and ending at ‘Apex complete’. A tooth in Stage 5 passed
the developmental threshold of ¼ of the root length formed but not the threshold
of ½ of the root length.
43
Measurements versus stages
corresponding stage and registered accordingly. On the other hand, during its
maturation, the dimensions of the third molars increase, and measures of the
observed third molar sizes are registered. For optimal data collection, the
registration technique affording the best age prediction performances has to
be determined.
Third molar growth is a chronological sequence starting with the
formation of an organic matrix followed by its subsequent calcification or
mineralization (Smith, 1991). The process starts with the growth of caps and
bells forming crypts in the jaw bone. It ends when the tooth roots are
completely calcified and is established with the closure of the root apices
(Calonius et al., 1970). The intermediate tooth development can be assessed
in succeeding and arbitrarily chosen stages of growth. Accordingly, multiple
tooth staging and related scoring techniques have been developed (Gleiser
and Hunt, 1955; Moorrees et al., 1963; Demirjian et al., 1973; Häävikko,
1974; Gustafson and Koch, 1974; Harris and Nortjé, 1984; Kullman, 1992).
These techniques provide ordinal data. Anatomic tooth features or
predictions of future tooth-part dimensions are used to identify the
thresholds between the stages. The ‘complete calcification of the tooth
crown’ is an example of an anatomical borderline between two stages, while
‘root half completed’ specifies a reference point without the final root length
once the tooth has stopped growing being known [Fig.4].The subjective
approach in this case is seen as a drawback of this staging technique for age
estimation. Furthermore, the degree of third molar development between
equally staged subjects can differ. The difference is the most between
subjects with features that classify them with a third molar development just
passing the lowest threshold of a specific stage and subjects with a degree of
third molar development classified immediately before the highest threshold
of the same stage. These differences remain, regardless of the number of
stages described in the technique. Both disadvantages could be avoided by
measuring the lengths of the developing third molar on the radiographs.
These measurements provide continuous data and have been reported to
afford an objective, precise and accordingly highly reproducible tool of
registration (Israel and Lewis, 1971; Liversidge and Molleson, 1999;
Cardoso, 2007; Santoro et al., 2008; Smith and Buschang, 2010). Moreover,
it was observed that these measurements correct certain deformations
inherent to the radiographic set-ups. Geometric image deformations can be
circumvented by calculating tooth measurement ratios (Kvaal et al., 1995;
Cameriere et al., 2006) and deformations due to a tilted cheek position of the
measured tooth can be detected and revised. Taking into account dimensions
of the second molar could diminish the variability in tooth size between
individuals.
The aim of this study was to measure the dimensions of third and
preceding second molars on panoramic radiographs and to check the
significance of possible relations between these measurements and age.
44
Measurements versus stages
Figure 5: Illustration of performed tooth dimension measurements
Op=occlusal plane, 1=Total Tooth Length (TTL), 2=Occlusal Plane Length
(OPL), 3=Pulp Horn Length, 4=Cement-Enamel Junction Length (CEJL),
5=Crown Width, 6=Cement-Enamel Junction Width [Table 2]. On the left panel
the four length measures of tooth # 48 are illustrated. In cases where the cement
enamel junction on the mesial and distal side was not at the same horizontal level,
the mean height between the two points was considered. On the right panel the
two width measures of tooth # 48 are indicated (5,6).
Whether or not these measurements add information to age prediction once
the staging and scoring of third molar development is performed will be
checked. In line with the advantages of measurements reported in the
literature, the research hypothesis stating that the registration technique
measuring third molar development provides better age estimation
performances compared to the staging technique will be investigated.
MATERIALS AND METHODS
In the age range between seven and 24 years of age, 340 (170 F, 170 M)
panoramic radiographs, digitally captured with a Veraviewepocs 2D unit (J.
Morita Inc., Irvine California, USA) were retrospectively selected. More
specific in each age category of 0.1 year, starting at seven years of age, 1 F
and 1 M subject were randomly picked from the dental clinic files of the
Katholieke Universiteit Leuven (Belgium). To collect indices on all the
present lower right third (Fédération dentaire international (FDI) #48) and
second molars (FDI #47), the radiographs were imported in Adobe®
Photoshop® (Adobe Systems Incorporated, San José California, United
States America).
45
Measurements versus stages
Table 2: Overview of abbreviations and descriptions of collected indices
Indices group
Abbreviation
Tooth length
TTL*
Tooth length from the most occlusal to the most apical
calcified tooth point
OPL*
Tooth length from the occlusal plane tothe most apical
calcified tooth point
PHL*
Tooth length from the occlusal plane to the most occlusal pulp
horn point
CEJL*
Tooth length from the occlusal plane to the cement enamel
junction
Tooth width
CW*
CEJW*
Ratio
Crown width at the cement enamel junction
TTL48/TTL47
R2
OTL48/OTL47
TTL/CW on the third molar
47
TTL/CW on the second molar
R448
TTL/CEJW on the third molar
R447
TTL/CEJW on the second molar
R3
R5
48
TTL/PHL on the third molar
R5
47
TTL/PHL on the second molar
R6
48
TTL/CEJL on the third molar
R647
R7
Score
Maximal crown width
R1
R348
Ratio of ratios
Description
48
TTL/CEJL on the second molar
TTL48/OPL48
R3
R348/R347
R4
R448/R447
R5
R548/R547
R6
R648/R647
KO*
PC
Developmental score following Köhler et al. (1994)
The score on the first principal component
*To specify the measured or scored tooth, the indices were given an additional
indication of the corresponding tooth number (e.g., TTL measured on lower right third
molar: TTL48). 48: indices on lower right third molar, 47: indices on the lower right
second molar
First, both molars were scored following the 10-point staging
technique developed by Gleiser and Hunt (1955) and modified by Köhler et
46
Measurements versus stages
al. (1994) (KO) [Fig.4].
Second, four tooth lengths were measured: total tooth length (TTL),
occlusal plane length (OPL), pulp horn length (PHL) and cement enamel
junction length (CEJL); and 2 tooth widths: crown width (CW) and cement
enamel junction width (CEJW) [Fig.5, Table 2]. For optimal measurements,
the radiographs were zoomed at 300%, the screen canvas was arbitrarily
rotated parallel to the occlusal plane of the examined tooth, guides were
dragged at the selected tooth marks, and the measurements were made with
the measurement tool snapped to the guides. The occlusal plane of a tooth
was defined as the line connecting the tips of a mesial and distal cusp
radiologically projected on other tooth materials. These settings were
installed separately for the length and the width measurements of each tooth
(FDI # 47, # 48).
Third, the ratios of these measurements and, fourth, the ratios of
these ratios were calculated [Table 2]. The ratios of tooth lengths and/or
tooth widths from the same tooth (R348, R347, R448, R447, R548, R547, R648,
R647, and R748) were considered in order to correct for radiographical
deformation. The ratios of the corresponding tooth lengths obtained on the
third and second molar (R1, R2) and the ratios of ratios obtained on the third
and second molar (R3, R4, R5, R6) were calculated in order to diminish the
effect of variability in tooth size. Specifically for the evaluation of this
effect, the original sample was divided into individuals having a fully
developed second molar (KO47 = 10) and individuals with a calcifying
second molar (KO47 < 10). The ratio between TTL48 and OPL48 (R748 ) gives
an indication of the degree of bucco-palatal inclination of the third molar
(Ratio = 1 is no inclination).
To quantify differences in the amounts of information between
various age-related indices, coefficients of determination (R²) and root mean
squared errors (RMSE) is derived from regression models with age as the
response. A model was used for each index separately. Non-linearity in the
relation between the index and the age was allowed using restricted cubic
splines (Harrell, 2001). All indices were treated equivalently as continuous
variable, meaning that KO was not treated as an ordinal or categorical
predictor but with the same number of degrees of freedom as all indices. The
developmental status of the second molar (fully developed versus not fully
developed) was included as a binary factor, and the relation between the
index and age was allowed to differ as a function of this status (by including
the interaction between index and status). Multivariable regression models
were used to check if the other indices added information to age prediction
once KO48 was used to determine if combining indices reduced the RMSE.
A principal component analysis was performed on all the length and
width measurements and ratios. The scores of the subjects on the first
principal component (explaining 79.1% of the variability) can be interpreted
47
Measurements versus stages
as an index of development. This score (PC) is a weighted average of all the
indices and was used as an alternative predictor for age estimation.
Since there is no crown information yet at younger ages, the absence
of information of KO, PHL, CEJL, CW, and CEJW is related to age.
Therefore, a factor with two levels (0 = no information missing, 1 =
information missing) was added to the regression models using these indices.
Exploring the regression models revealed that the variance of age
was not constant. To handle this, the variance was allowed to be specific for
3 KO48 categories, namely for the KO48 less than 5, KO48 between 5 and 9,
and KO48 equal to 10. The models were fitted separately for F and M. All
analyses were performed using SAS software, Version 9.2, of the SAS
System for Windows (© 2002 SAS Institute Inc.). SAS and all other SAS
Institute Inc. product or service names are registered trademarks or
trademarks of SAS Institute Inc., Cary, NC, USA. The procedure PROC
MIXED was used to fit models with non-constant variance.
RESULTS
55.6% (189/340) and 17.77% (60/340) of the second and the third molars,
respectively, were fully developed. For 53.9% (151/280) of the third molars
that were not completely developed, the corresponding second molar had not
reached the final developmental stage. For M and F, the latter percentage
equalled 56.1% (78/139) and 51.8% (73/141), respectively.
The univariate regression models for each index separately revealed
that using KO48 yielded the most accurate age predictions of all of the
indices. Indeed, the R² for the model using KO48 was the highest and the
RMSE was the lowest at each of the variance specific levels. The
performance of KO was best (higher R², lower RMSE) for M compared to F.
There was no evidence that indices based on ratios would yield better age
predictions than indices based on full-length measurements (TTL, OPL)
[Table 3].
None of the other indices added significant information to the age
prediction once KO48 was used, independent of the calcification status of the
second molar (results not shown). An exception was found for F, where
adding PC or R648 provided a statistically significant but clinically small gain
of information. The increase in R² was maximally 2% and, in both cases, the
RMSE, calculated for the three variance specific levels, changed hardly.
Multivariable regression models revealed that a combination of the
best-performing length index (OPL) with indices based on ratios does not
yield a better age prediction than does the use of only KO48. For example, for
M the highest R² was obtained with the combination of OPL48, R448 (R² =
0.76) and OPL, R4 (R² = 0.76). Neither combination contains more
information than KO48 (R² = 0.86) (results not shown).
48
Measurements versus stages
Table 3: List of coefficients of determination (R²) and root mean squared errors
(RMSE) calculated from index-specific regression models with age as response.
Females (N = 170)
Indices
48
KO
TTL48
OPL48
PHL48
CEJL48
CW48
CEJW48
R1
R2
R348
R3
R448
R4
R548
R5
R648
R6
R748
PC
Males (N = 170)
N
R²
RMSE RMSE
KO48<5 5≤KO48<10
RMSE
KO48 = 10
N
R²
RMSE RMSE
KO48<5 5≤KO48<10
RMSE
KO48 = 10
132
133
133
116
111
131
110
0.78
0.72
0.73
0.51
0.56
0.46
0.56
0.70
0.70
0.74
0.69
0.73
0.70
0.70
0.62
0.69
0.62
0.58
0.76
1.65
1.58
1.59
2.04
1.67
2.16
1.75
1.73
1.73
1.50
1.67
1.45
1.51
1.71
1.76
1.84
1.79
1.82
1.47
1.20
1.91
1.97
4.24
4.28
4.50
4.29
1.35
1.54
2.01
1.69
1.60
1.62
2.17
2.25
2.50
2.75
3.55
1.29
130
135
135
113
113
134
111
0.86
0.72
0.75
0.59
0.59
0.54
0.58
0.72
0.74
0.74
0.69
0.76
0.74
0.67
0.64
0.65
0.64
0.62
0.73
1.20
1.48
1.31
1.70
1.63
1.61
1.52
1.57
1.38
1.37
1.59
1.12
1.19
1.39
1.56
1.50
1.60
1.62
1.28
1.39
1.82
2.00
3.46
3.07
3.87
4.02
1.82
2.06
1.77
1.85
2.05
2.03
3.23
3.40
3.16
3.09
2.86
2.41
2.12
2.20
2.13
2.36
2.41
2.42
2.33
2.67
2.55
2.14
2.62
2.60
2.76
2.16
2.56
1.96
2.44
2.36
2.42
1.47
2.37
2.17
2.40
2.59
2.64
2.42
2.31
2.12
2.40
2.52
2.36
2.38
2.22
2.20
2.28
2.25
2.47
2.21
N: number of subjects with information on the index. For each model, three RMSEs are
reported since a model with non-constant variance was needed. Note that the regression
models are always based on N = 170 (see statistical methodology).
DISCUSSION
The age range of the subjects included in a study will bias the age
predictions as soon as the age distribution conditional on predictors is
truncated. A straightforward example of such bias would occur if the KO 48
score is used for age prediction, and the maximal age of the subjects is
restricted. In this situation, using the results of such a study for age
predictions will underestimate the age of subjects with a fully developed
third molar. Similarly, if length measurements are used for prediction, the
age might be overestimated for subjects with lower values if the minimal age
to enter the study is chosen inappropriately. In the current study, 10 M and
10 F were included within each age range of 1 year. The maximal age was
set at the age range in which all 10 included subjects had a fully developed
third molar (i.e., 24 years old) during the random selection, thus avoiding
right truncation of the age distribution. The minimal age was set at 7 years
old. For this age category, all of the 10 M subjects had no calcifying third
molars during the random selection. For the girls, this was the case for all of
the subjects in the age categories less than 10 years old. For two boys in the
9-10 year-old range, a KO48 was available, and in the 8-9 year-old range
length measurements (TTL48, OPL48) were obtained for only one boy. As
49
Measurements versus stages
such, left truncation of the age distribution is unlikely for the subjects in the
earliest stages of third molar development.
The results related to TTL48 concur with the findings reported by
Liversidge and Molleson (1999). These authors found an S-shaped relation
between tooth length and age, and, in current study, nonlinear terms were
needed to describe the relation between TTL48 and age (results not shown).
They also reported an RMSE value (1.478 years) comparable with our
results for third molars. Note that the composition of the sample they used
consisted of the younger children, so their recommendation to prefer
information from other teeth for age prediction holds only for these young
ages (<5 year).
Of all the indices, KO48 yields the most accurate age predictions.
More specifically, the continuous data from the raw total third molar length
measurements (TTL48, OPL48) did not provide extra age-related information
beyond the categorical data from the ordinal KO48 stages (10 levels). Equal
results were obtained for all the ratios between the tooth lengths and the
tooth widths from the same tooth (R348, R347, R448, R447, R548, R547, R648
and R647, R748) used to eliminate radiographic distortions. Ratios
normalizing the raw third molar length measurements on corresponding
second molar length measurements (R1, R2, R5, R6) were used in an attempt
to reduce the influence of tooth size, particularly for individuals with a fully
developed second molar (KO47 = 10). But these ratios did not yield better
age predictions than did the raw third molar length measurements. Even the
PC score, which reflects information from all of the included indices, did not
outperform the KO48, human variability in tooth size probably being the
major cause of this. In this study, the variability of third molar size can be
derived from the TTL48 and OPL48 measurements on all the fully developed
third molars (KO48 = 10, n = 60), which range between 2.3 cm and 3.4 cm
and between 2.0 cm and 3.2 cm, respectively. On all fully developed second
molars (KO47 = 10, n = 189) the ranges for both measurements were
respectively 2.4 cm-4.1 cm and 2.3 cm-3.9 cm. Moreover, human variability
in difference between the third and the second molar size has to be taken into
account. The ranges for the difference in length between the second and the
third molar were -0.02 cm to 1.01 cm and -0.08 cm to 1.33 cm for TTL and
OPL measurements, respectively. Note that, for TTL and OPL, larger
measurements were regularly obtained for the second molar. Using length
indices of developing teeth as information for age estimation embodies
previous variability and results in an extra loss of age-related information.
Scoring third molar development is independent of tooth size variability if
the observed third molar calcification information is used as the standard.
Based on this standard, predictions of future third molar lengths can be
made, and these predictions allow one to categorize the developing wisdom
tooth regardless of tooth size variability. This implies that scoring has to rely
on the highest intra- and inter-observer reliability to rule out subjective
50
Measurements versus stages
operator influences. Moreover, predictions of third molar lengths should not
be based on (or compared with) the dimensions of neighbouring teeth.
Using combined length measurements of the second and third molar
(R1, R2, R3, R4, R5, R6) does not result in a gain of age-related information
over that provided by the raw third molar length measurements (TTL48,
OPL48). This is most likely due to the simultaneous development of the
second and third molars for a long period of time. In the data studied, all of
the second molars achieved complete development at the age of 19. More
specific 73 F and 78 M had KO47 less than 10, which means that only 55.6%
of all of the second molars were fully developed (KO47 = 10) and not in
developmental overlap with the corresponding third molars. An alternative
explanation would be that the measurement error in the measurements is too
high, so the relations with age are attenuated. To obtain an indication of the
amount of measurement error, intra-observer reliability was evaluated.
Therefore, the measurements of 10% of the M individuals (chosen at
random) were measured again by the same observer. The results on the
standard error of measurement (SEM), also expressed relative to the mean
value (within-subject coefficient of variation = WSCV), revealed a high
level of intra-observer agreement for all the measurements (SEM [0.006 0.022], WSCV [0.3% - 1.5%]). This indicates that the measurement errors
cannot be deemed a cause of the lack of gain in age-related information.
In further research, models with combinations of indices other than
those evaluated in this study could be explored. They are expected to be less
informative about age than stages and scorings, and care has to be taken not
to over fit the data.
CONCLUSIONS
Third molar stages (categorical data) were the best related to age and
provided the most accurate age predictions than did all the compiled tooth
measurements and ratios of tooth measurements (continuous data).
Moreover, combining the scored third molar stages with tooth measurements
or ratios contributed no clinically relevant information gain for age
prediction. Therefore, Research Hypothesis 1 was not accepted. The
technique of third molar staging and related scoring has to be recommended
over complicated dimension measurements or ratio calculations of second
and/or third molars for age estimates.
51
Chapter 4
Staged third molar
development registration:
Influence of the number of
stages on age estimation
THIS CHAPTER IS BASED ON THE FOLLOWING MANUSCRIPT.
Third molar development: Evaluation of nine tooth development registration
techniques for age estimations
Thevissen PW, Fieuws S, Willems G
Published in Journal Forensic Sciences
2013 10.1111/1556-4029.12063
Oral presentation at the annual scientific meeting of the American Academy of
Forensic Sciences, Chicago, 2011
TESTING RESEARCH HYPOTHESIS 2:
The number of stages used in the third molar development staging technique
influences the age predictions
53
Number of stages
INTRODUCTION
Forensic dental age estimation assessments in living sub-adult individuals
consider at most the methods based on third molar development (Melsen et
al., 1986; Schmeling et al., 2001; Solheim and Vonen, 2006; Schmeling et
al., 2006; Olze et al., 2006a). The radiologically observed third molar
development is detectable starting from the uncalcified crypt formation until
the apical closure of the tooth roots (Massler and Schour, 1946). The degree
of tooth development can be registered as a measure of the observed tooth
length (Israel and Lewis, 1971; Liversidge and Molleson, 1999) as a ratio of
perceived tooth dimensions (Cameriere et al., 2006; Cardoso, 2007;
Thevissen et al., 2011), or it can be classified in different stages (Gleiser and
Hunt, 1955). In the staging techniques, the criteria used to describe the
borderlines of the specific stages are based on a single or a combination of
distinct transformations observed on the tooth germ or on the emerged tooth.
For this purpose, anatomically detectable transitions such as: crypt
formation, initial calcification, formation of the crown cusps, eruption,
differentiation of the cement enamel junction, appearance of the interradicular cleft and apical root closure of the developing tooth were
considered. In addition, the length proportions of already formed tooth parts
were taken into account by Demirjian et al., 1973; Gustafson and Koch,
1974; Raungpaka, 1988, and estimates of parts of the future crown or root
length by Gleiser and Hunt, 1955; Moorrees et al., 1963; Häävikko, 1974;
Harris and Nortjé, 1984; Kullman et al., 1992. Hence, tooth development
registration was described in at least 4 and at most 15 stages. To estimate
age, the registered degree of third molar development was compared with
age standards assembled in tables (Liversidge, 2008b) and atlases
(AlQahtani et al., 2010) or implemented in age predicting models (Gunst et
al., 2003). An important clinical and forensic issue is the influence that the
techniques have on age prediction. In particular, the influence that the
arbitrary number of stages applied in the third molar development technique
has on the age estimation, has to be studied.
Table 4: Gender specific age distribution of research sample expressed in years.
Gender n
591
F
608
M
mean
std
min Q1
median Q3
max
18.22
18.24
4.84
5.04
4.31
5.27
18.12
18.59
30.10
33.91
14.84
14.64
21.72
21.83
n: number of subjects, std: standard deviation, min: minimal age, Q1: first quartile, Q3:
third quartile, max: maximal age, F: females, M: males
55
Number of stages
Table 5: For each tooth development registration technique the number of stages used
to register the development of the different tooth parts
Tooth development registration technique
Crypt
Crown
Root
Apex
Number stages
3
5
2
10
4
3
1
8
1
5
5
1
12
KU
MO HN
RA
GK
5
2
7
6
6
2
14
2
3
2
2
5
4
4
1
5
CA
Continuous data
KO DM HA
KO: Köhler et al. (1994), DM: Demirjian et al. (1973), HA: Haavikko (1974), KU:
Kullman et al. (1992), MO: Moorrees et al. (1963), HN: Harris and Nortje (1984), RA:
Raungpaka (1988), GK: Gustafson and Koch (1974), CA: Cameriere et al. (2008)
The present concern is to evaluate which third molar development
registration technique is the most suitable tool for sub-adult age estimation.
Therefore, the correlations between the nine third molar development
registration techniques and between these techniques and age were studied.
Corresponding regression models were calculated, and the best-performing
techniques were verified on a second sample. The research hypothesis
studied was that the number of stages used in the third molar development
registration technique influences the age predictions.
MATERIALS AND METHODS
From the Bhagwan Dental Clinic in Haryana and the Jain Diagnostic Centre
in Delhi in India, 1199 analogue panoramic radiographs of 591 F and 608 M
subjects were collected retrospectively. The age of the individuals ranged
between 4 and 34 years old [Table 4]. All of the radiographs were taken
from subjects of Indian nationality who had lived all their lives in the North
Indian states of Haryana or Delhi. The age of the individuals at the moment
of radiological exposure was calculated from their authentic birth certificate
and the registered exposure date. The subjects had no history of medical
diseases or interventions affecting the presence and/or development of teeth.
On the panoramic radiographs, the degree of third molar
development of 4147 present third molars was registered by means of nine
tooth development classification techniques. Each technique was described
by, and named after, one of the following nine authors: Gleiser and Hunt
(1955) modified by Köhler et al. (1994) (KO), Haavikko (1974) (HA),
Demirjian et al. (1973) (DM), Raungpaka (1988) (RA), Gustafson and Koch
(1974) (GK), Harris and Nortje (1984) (HN), Kullman et al. (1992) (KU),
Moorrees et al. (1963) (MO), Cameriere et al. (2008) (CA) [Table 5]. In the
first eight techniques, the degree of third molar development was staged and
56
Number of stages
Table 6: Gender specific listing of Spearman correlation coefficients between each
tooth development registration technique and age (ρ), determination coefficient (R²)
and root mean squared error (RMSE)
KO
DM
Tooth development registration technique
HA
KU
MO
HN
RA
GK
Females
CA
ρ
R²
RMSE
0.700
0.506
3.429
0.705
0.501
3.437
0.689
0.486
3.495
0.701
0.493
3.465
0.702 0.686
0.511 0.475
3.418 3.521
Males
0.697
0.488
3.482
0.677
0.471
3.532
-0.650
0.453
3.598
ρ
R²
RMSE
0.680
0.446
3.700
0.668
0.433
3.731
0.657
0.419
3.784
0.669
0.432
3.738
0.686
0.454
3.678
0.655
0.404
3.819
0.666
0.419
3.773
-0.636
0.380
3.903
0.659
0.403
3.825
KO: Köhler et al. (1994), DM: Demirjian et al. (1973), HA: Haavikko (1974), KU:
Kullman et al. (1992), MO: Moorrees et al. (1963), HN: Harris and Nortje (1984), RA:
Raungpaka (1988), GK: Gustafson and Koch (1974), CA: Cameriere et al. (2008),
RMSE expressed in years
scored. In the last technique, the normalizing ratios between the sum of the
distances between the inner sides of the third molar roots and measurements
of the corresponding third molar length were evaluated.
For the nine different third molar development registration
techniques applied on the lower-left third molars, Spearman's correlations
were calculated to explore associations among the third molar development
registration techniques and between each technique and age. In the absence
of a lower-left molar, the registration of the lower-right molar was used.
On the set of subjects having a third molar development registration
in the lower jaw from all nine techniques (n = 1121), a regression model
(allowing non-linearity) was fitted, for each registration technique
separately, with age as the response and the registered third molar
development as an categorical predictor except for the CA technique, which
was entered as a continuous predictor. To allow a nonlinear relation for this
technique, restricted cubic splines were used on the log-transformed
registrations. From each sex-specific model, the determination coefficient
(R²) indicating the proportion of variance in age explained by the tooth
development registration technique and the root mean squared error (RMSE)
were calculated. The RMSE is a standard deviation that reflects the
variability of the predicted age around the true age.
Using non-parametric bootstrapping (based on 5000 samples), 95%
confidence intervals were constructed for all pair wise differences in R 2
between the various registration techniques, which permits detection of the
significant differences at the 5% level.
57
Number of stages
Additionally, a supplementary validation sample, including 239
panoramic radiographs of 131 F and 108 M subjects between 16 and 23
years old was collected from the same population to validate the models of
the two most age-related tooth development registration techniques using
stages and the CA. technique. Furthermore, it was determined if added
information from an upper third molar improved the age prediction if the
score of the lower molar had already been used.
The probabilities of being older than 18 and 21 years of age given
the maximal score for lower and/or upper molars were obtained from the
regression model.
All the analyses were performed using SAS software, Version 9.2 of
the SAS System for Windows. (SAS Institute, Cary NC, USA)
RESULTS
In F, the Spearman correlation coefficients among the third molar
development registration techniques, except for the relations with CA, varied
between 0.901 (relation between KO and RA) and 0.995 (relation between
MO and KO). In M, these values were 0.903 and 0.993 for the same
relations, respectively. The Spearman correlation coefficients between CA
and each other tooth development registration technique ranged in F between
-0.884 (relation with KO) and -0.846 (relation with RA), in M between 0.910 (relation with KO) and -0.887 (relation with HA).The Spearman
correlation coefficients between each tooth development registration
technique and age was listed for F and M separately [Table 6]. The best age
predicting model is MO with, for both F and M, the highest R² (F 51%, M
45%) and accordingly the lowest RMSE (F 3.42 year; M 3.67 year) values
[Table 6].
Based on non-parametric bootstrapping on a total of 36 sex-specific
pairwise tooth development registration technique combinations, the
differences in mean R2 were significant at the 5% level for only 6 F (KOHN, KO-GK, DM-GK, MO-HN, MO-RA, MO-GK) and 13 M combinations
(KO-HA, KO-HN, KO-RA, KO-GK, KO-CA, DM-MO, DM-RA, HA-MO,
KU-HN, MO-HN, MO-RA, MO-GK, MO-CA). All the calculated mean R²
differences between the pair third molar development registration technique
combinations were small. The maximal mean difference in R2 was detected
in F as well as M between MO and CA (F 6.2%, M 8.0%). For all the tooth
development scoring and measuring techniques, information in the upper
jaw added significantly to the accuracy of the age prediction if the score of
the lower third molar had already been used (results not shown). Therefore,
regression models that take into account the third molar jaw position were
developed for the two best (KO, MO) and the least (CA) performing third
molar registration techniques. Figures 6 a, b and c present the results from
58
Number of stages
the validation on 239 subjects of the regression models for KO, MO and CA,
with upper and lower third molar information. For each model, the relation
between the true age and the difference between the predicted and the true
age was plotted to visualize the magnitude and the direction of the errors in
the age estimation. For F, a RMSE of 3.36, 3.37 and 3.85 years was obtained
for KO, MO and CA, respectively. In M, these values were 3.69, 3.75 and
4.00 years respectively. Expressed as mean absolute error, the values were
2.70, 2.72 and 3.07 years in the total validation sample for KO, MO and CA,
respectively. Note that there is on average a slight underestimation of the
true age (mean errors are positive for all methods). Typically, when applying
regression models for age estimation, this underestimation is stronger for
older subjects while overestimation is more likely for younger subjects.
The probability of being older than 18 years of age given the
maximal MO, KO and CA third molar development registration on the left
lower and/or upper third molar(s) ranged for F between 84% and 98% and
for M between 85% and 96%. For the age threshold set at 21 years of age,
these probabilities decreased to a range from 59% to 89% for F and from
60% to 83% for M [Table 7]. Of the 144 adults (older than18 years of age) in
the validation sample, 80.0%, 80.0% and 66.0% were correctly identified by
KO, MO and CA, respectively. Of the 95 juveniles (younger than 18 years of
age), 54.7%, 55.8% and 67.4% were correctly identified by KO, MO and
CA, respectively.
DISCUSSION
The comparison of the various third molar registration techniques indicates
that MO is the most promising for age estimation. Although a modest
Table 7: Probabilities to be older than 18 and 21 years for KO, MO and CA given a
maximal third molar development registration for lower and/or upper left third
molar(s).
Female
Tooth position
Upper
KO Lower
Upper and Lower
Upper
MO Lower
Upper and Lower
Upper
CA Lower
Upper and Lower
P(age>18yrs)
0.98
0.97
0.98
0.97
0.96
0.98
0.86
0.84
0.88
P(age>21yrs)
0.88
0.86
0.89
0.85
0.83
0.86
0.62
0.59
0.65
Male
P(age>18yrs)
0.95
0.96
0.96
0.94
0.95
0.96
0.85
0.85
0.86
P(age>21yrs)
0.79
0.81
0.83
0.78
0.80
0.81
0.61
0.60
0.62
KO: Köhler et al. (1994), MO: Moorrees et al. (1963), CA: Cameriere et al. (2008),
P(age>18yrs): probability being older than 18 years of age, P(age>21yrs): probability
being older than 21 years of age
59
Number of stages
number of the differences with other techniques were statistically significant,
clinically and practically they can be considered negligible. The difference in
R² with the worst performing technique was 5.8% and 7.4% for F and M,
respectively. The RMSE increased only by 0.18 years (66 days) for F and
0.23 years (82 days) for M using the worst method (CA) instead of MO.
Note that statistical significance was easily reached due to the relatively
large number of subjects. The non-parametric bootstrapping analysis allowed
for analyses that provided high power. They confirmed the small R2
differences for all the pair-wise combinations of the tooth development
registration techniques. The evaluation of the MO, KO and CA registration
techniques on a validation sample revealed the same pattern, i.e., similar
performances for the various techniques. Observation of the number of
correctly identified individuals older or younger than 18 years of age in the
validation sample confirmed these similar performances.
Although the regression model based on MO third molar registration
promises the best age predictions, the regression models based on the other
registration systems perform comparably. The maximum difference in the
proportion of variance explained by the models (R²) was 5.8% for F and
7.4% for M both detected between MO and CA. Between these models, the
variability of the predicted age around the true age (RMSE) increased for F
0.18 year (66 days) and for M 0.23 year (82 days). If one considers only the
registration techniques based on staging, the maximal difference between R²
values reduces to 4% for F and 5.1% for M and the RMSE values increase
maximally to a value of 0.11 year (40days) for F and 0.15 year (55 days) for
M [Table 6]. Although a few of these differences were statistically
significant (5% level), clinically and practically they can be considered
negligible.
The difference between the true and the predicted age as a function
of the true age was evaluated at the validation sample for the MO, KO and
CA registration techniques. The three graphs and the reported quantification
of the error [Fig. 6a, b, c] confirm similar performances of all the validated
techniques. Of the three techniques, the RMSE are larger than the MAE,
which indicates that the magnitude of the variance in the individual errors is
not uniform within the samples. The graphs with the difference between the
true and the predicted age as a function of the true age show that young
individuals are systematically overestimated (attraction of the middle) (Lucy
et al., 2002; Prince et al., 2008; Thevissen et al., 2010b). Moreover, the
analogous performances of especially the third molar registration techniques
based on staging were also observed in the probabilities to be older than 18
years or 21 years of age given the maximal MO, KO and CA [Table 7]. This
finding was confirmed by identifying the number of correctly identified
individuals older or younger than 18 years in the validation sample.
60
Number of stages
Figure 6a: Difference between the true and the predicted age as a function of
the true age for the validation sample using Köhler’s registration technique.
The predictions were obtained from the regression models for age using KO:
Köhler et al. (1994) (allowing a nonlinear relation). RMSE : root mean squared
error, y: years. A negative value refers to an overestimation of the true age and a
positive value to an underestimation.
The model based on the CA registration technique yields the least
accurate age predictions. This is because the CA technique functions at
different levels than do all the other evaluated tooth development registration
techniques: the CA technique registers continuous data and the CA data are
based on ratios between measurements of apical pulp widths and tooth
lengths. Since these tooth length measurements were integrated by
Cameriere et al. (2006) to neutralize possible distortions and magnifications
inherent to the panoramic imaging technique and unit used, the CA
technique essentially measures the changing apical pulp widths of
developing third molars. This is in contrast to all the other techniques, which
take into account the stages depending on the changing lengths of the
developing third molars. In the CA technique, tooth length measures are
used as the denominator in its ratios since the measures increased in a similar
way as the stages used in all other techniques. Thus, the CA values decrease
when the values of all the other techniques increase. As a result, the
Spearman correlation coefficients between the staging techniques and CA
were negative. Thevissen et al. (2011) contend that staging and scoring third
molar development (categorical data) were best related to age and promised
the most accurate age predictions relative to tooth measurements and ratios
of tooth measurements from third and second molars (continuous data). This
61
Number of stages
Figure 6b: Difference between the true and the predicted age as a function of
the true age for the validation sample using Moorrees’ registration technique
The predictions are obtained from the regression models for age using MO:
Moorrees et al. (1963). RMSE: root mean squared error, y: years.
finding was explained as being due to the measures and related ratios used to
register molar development incorporating the variance in tooth size between
individuals. This finding, in addition to the previously described finding
about apical tooth width measurements, could explain the lowest R² and
highest RMSE values for CA compared to all the other third molar
development registration techniques, which are based on stages.
The present study demonstrates that the MO scoring technique is the
best predictor of tooth development, followed by KO, DM and KU in turn.
These findings are in agreement with the conclusions of Olze et al. (2005),
who considered DM as the best, followed by KO and KU. In their study, five
tooth development registration techniques based on staging were evaluated,
but MO was not included. Furthermore, in the Olze study, the precision
(reproducibility) of the registrations was evaluated and integrated into the
conclusions. Indeed, DM classifies the different tooth developmental stages
on the basis of objective observations and so avoids predictions of lengths of
tooth parts and results in greater precision.
The validated regression formulas provide a tool for age estimation
of North Indian individuals depending on the tooth development scoring
technique used (MO, KO or CA) and taking into account the relative
position in the mouth of the evaluated third molar as well as gender.
62
Number of stages
Figure 6c: Difference between the true and the predicted age as a function of
the true age for the validation sample using Cameriere’s registration technique.
The predictions were obtained from the regression models for age using CA:
Cameriere et al. (2006). RMSE: root mean squared error, y: years.
Considering the regression model for KO on this North–Indian sample, tooth
development is slower than in other countries, e.g., for a lower and upper
score 7 median age is 20.4 years old for F and 20.5 years old for M
compared to Thailand (Thevissen et al., 2009) and Belgium (Gunst et al.,
2003) for the same score median ages being 18.9 and 18.5 years old for F
and 18.6 and 17.8 years old for M, respectively.
In practice, the choice of the staging technique should depend
largely on the number of stages available in the developmental period of
interest. For early third molar development, third molar development
registration techniques without (HN, KU) or with few (KO) initial stages
(crypt and crown formation) should be excluded. Similarly, considering late
third molar development (root apex formation), registration techniques
without (RA, GK) final stages should be omitted [Table 5]. Further on, the
chosen staging technique should include stages that allow the entire third
molar maturation sequence to be covered. As such, extreme outliers can be
registered and included even when a specific developmental period is of
interest. Some techniques can be considered refinements of another
technique, i.e., a given stage is split into two sub-stages, which provides for
more differentiation. Due to this split, the initial technique is partially
integrated, or nested, in the refined technique. These nested techniques
63
Number of stages
include a high number of equal stages and perform most similarly (e.g.,
MO/KO). When there are nested techniques available for the assessed
period, the very small gain of information provided by the technique with the
highest number of stages should not compromise the feasibility of correctly
registering all the stages described (Corradi et al., 2013). Indeed, the more
stages a technique involves, the less precise the classification. Certainly, this
is the case when the thresholds between stages are very close, unclearly
described, hard to distinguish, or dependent on predictions of future lengths
of tooth parts. Since nested techniques perform similarly, the error made by
a misclassification will weigh much more in the age prediction than the
ignorable gain that can be obtained by choosing a technique with more
stages, which promises slightly better age predictions. The KO technique
provides well-described stages covering the entire tooth maturation sequence
[Table 5]. Moreover, it is a technique that permits the registration of the late
third molar development. No other technique, except KU, is nested in KO.
Because KU is only suitable after crown formation, the KO staging was
chosen as the third molar development staging technique in the further
research.
CONCLUSION
Similar age predicting performances were detected by comparing regression
models using nine third molar development registration techniques.
Although some of the differences between the examined third molar
development registration techniques were statistically significant, these
differences were clinically unimportant. The number of stages used in the
third molar registration technique slightly influenced the age predictions, so
Research Hypothesis 2 has to be rejected. The choice of the third molar
development registration technique has to depend on the stages described for
the developmental period of interest and should not compromise the
feasibility of correctly registering all of these stages. Accordingly, the KO
technique was chosen as the third molar development staging technique for
the further research.
64
Chapter 5
Statistical modeling of third
molar development data:
Influence of regression
analyses versus Bayesian
approach on age estimation
THIS CHAPTER IS BASED ON THE FOLLOWING MANUSCRIPT.
Human dental age estimation using third molar developmental stages: does a
Bayesian approach outperform regression models to discriminate between
juveniles and adults?
Thevissen PW, Fieuws S, Willems G
Published in International Journal of Legal Medicine
2010 Jan;124(1):35-42
Oral presentation at Vierzehnten Treffen der Arbeitsgemeinschaft für Forensische
Altersdiagnostik (AGFAD), Berlin 23 03 2011
TESTING RESEARCH HYPOTHESIS 3:
A Bayesian approach using third molar development provides more accurate
age predictions than does classic regression analyses in sub-adults
65
Regression analyses versus Bayesian approach
INTRODUCTION
Legal systems around the world, based either on civil jurisdiction or on
common law, have an interest in the age of unaccompanied young refugees.
In particular, their status as juvenile or adult is of concern (Schmeling et al.,
2001; Solheim and Vonen, 2006; Olze et al., 2006a; Benomran, 2009). To
determine the age of living sub-adults, dental age estimation methods based
on the radiologically observed stages of third molar development are used
(Willems, 2001) [Fig. 1, Chap. 1]. Therefore, in retrospective studies,
reference samples of panoramic radiographs from individuals with known
chronological age at the time of the radiologic exposure, gender and origin
were collected [Table 1, Chap. 1]. Individuals with no medical history, no
visible dental pathology on the radiographs and at least one third molar
present, were included. The observed degree of third molar development of
all the third molars present is classified using one of the scoring systems
proposed by several authors (Moorrees et al., 1963; Demirjian et al., 1973;
Gustafson and Koch, 1974; Häävikko, 1974; Harris and Nortjé, 1984;
Kullman et al., 1992; Köhler et al., 1994) [Table 5, Chap. 4]. As such the
reference samples provide categorical data and allow tables or prediction
models to be devised that give age estimates and a quantification of the
uncertainty of the prediction.
The classic approach for age estimation models based on third molar
developmental stages is the use of a regression model with which the age of
the ith subject is predicted using the information of one or more
developmental stages:
Age i  h(x i1 ,..., x i4 )  ε i ,
(A)
where xi1… xi4 denote the developmental stages of the four third molars.
When only one third molar is used, (A) reduces to a simple regression
model. Typically, h(.) is a linear function, i.e.,
Age i  α  βx i  ε i ,
and the number of used third molars is restricted due to the high correlation
between third molars as proposed by, for example, Mesotten et al. (2002),
Gunst et al.(2003) and Mesotten et al. (2003). Within the linear regression
modelling framework, the assumption of linearity can be relaxed by using a
spline or a polynomial function like h(.). Chaillet and Dermirjian (2004) and
Chaillet et al. (2004b) considered regression models with a cubic function
for one or two stages. For example, with only one third molar equation (A)
then becomes
Agei  α β1x i β 2 x i2 β 3x 3
i εi .
Note that these extensions still fit within the linear regression framework,
since age is still modelled as a linear function of a set of terms (e.g., linear,
67
Regression analyses versus Bayesian approach
quadratic, cubic). A crucial feature of the discussed regression model is that
the residuals εi (i.e., the difference between an observed value of the
response variable and the value predicted by the model: Moore and McCabe,
1993) are assumed to be normally distributed around the regression line with
a constant variance. As such, the use of the regression approach implies a
strict assumption about the shape (normal) and the variance (constant) of the
age distribution. It should be emphasized that it is exactly this distribution
that is used to quantify the uncertainty about the predicted age (i.e., by
constructing a 95% prediction interval) and to calculate specific
probabilities, such as the probability of being a juvenile [Fig.7].This reveals
the first drawback of the linear regression model for age, since the
assumption about the age distribution might often be too restrictive in
practice.
Other practical limitations concern the high correlation between the
independent variables in the regression model and the presence of missing
values for them. To circumvent this so-called multi-collinearity problem,
regression models are restricted in practice to a limited set of the available
third molars. To handle the missing values, separate models are constructed
for the various patterns of observed information. As such, the regressionmodel approach results in an extended set of regression equations, each of
them designed for a specific situation (Mesotten et al., 2002; Gunst et al.,
2003; Mesotten et al., 2003).
A more serious concern, discussed in depth by Aykroyd et al. (1997,
1999) is the systematic bias (attraction of the middle) in age estimation when
Figure 7: Illustration of the assumption of constant variance and normality in
the regression model.
This is a hypothetical example where the age is modeled as a linear function of
one developmental stage. The shaded part pertains to the probability of being
mature given a specific stage.
68
Regression analyses versus Bayesian approach
the classic regression approach is used: the weaker the relation between the
stages and age, the more the residuals εi in (A) will be related to age. As
such, the estimated ages are too old for young individuals and too young for
old individuals. The direction of this bias is exactly what is not tolerable in
the current legal context unlike a bias pattern where the age of juveniles
would be underestimated.
To remove the bias induced by the use of a regression model for age,
Lucy et al. (2002) proposed a Bayesian approach. Recent examples can be
found in Prince et al. (2008) and Prince and Konigsberg (2008). Following
the notation in (A), the age distribution given a specific pattern of stages
would here be obtained as:
f(x i1 ,..., x i4 | agei )f(agei )
f(agei | x i1 ,..., x i4 ) 
.
(x i1 ,..., x i4 | agei )f(agei )
i
(B)
Equation (B) consists of three parts. The left-hand side of the equation, i.e.,
the age distribution given an observed pattern of stages, is referred to as the
posterior distribution. Note that its equivalent in the regression approach is
the normal distribution with constant variance. However, the age distribution
conditional on the observed stages is not assumed a priori to have a specific
shape and variability in the Bayesian framework. Indeed, both features will
depend on the likelihood
f(x i1 ,..., x i4 | age i )
and the prior distribution
f(age i ) .

The likelihood function reflects the probability of the observed
pattern of stages given that a subject has a specific age. The denominator in
(B), which represents the probability of the observed pattern of
developmental stages, is used only for normalization purposes such that the
total surface of the posterior distribution equals one and surfaces under the
posterior distribution can be interpreted as probabilities. Indices of location
(e.g., median, modus) of the posterior distribution can be used as point
predictions for age, but more informative is the prediction interval that can
be obtained from its percentiles (e.g., 2.5th and 97.5th percentiles to represent
a 95% prediction interval). Moreover, the calculation of a specific surface
under the posterior distribution, i.e., the probability that the age exceeds 18
years old, will be crucial in the current context. The prior distribution of age
f(y) will often be a uniform distribution over a particular age range and can
be changed if there is specific prior knowledge about the age.
The aim of this chapter is to compare the human age estimation
based on third molar information using classic regression models and a
Bayesian framework. In particular, the aim is to verify if a Bayesian
approach discriminates more accurately between adults and juveniles, and
69
Regression analyses versus Bayesian approach
removes the bias, which disadvantages the latter group. The corresponding
research hypothesis is as follows: Based on third molar development, a
Bayesian approach provides more accurate and precise age predictions than
does the classic regression analyses in sub-adults.
MATERIALS AND METHODS.
Sample and measurements
A collection of 2,513 panoramic radiographs of Belgian Caucasian
individuals with ages between 15.7 and 23.3 years old collected by Gunst et
al. (2003) was studied. In conformity with the conclusion of the previous
chapter, the development of all the available third molars was staged and
scored with the technique of Gleiser and Hunt (1955) as modified by Köhler
et al. (1994) (KO). Since the results of an approach might be related to age, a
subset of patients between 16 and 22 years of age was used for comparing
the two approaches. This subset has a uniform age distribution (Smith, 1991)
and contains for each one-year interval between 16 and 22 years of age, the
largest overall number of women (n = 75) and men (n = 55) out of the main
collection. The subset was split at random into reference and validation sets
of comparable size. The models were developed on the reference set, and the
performance of both approaches is assessed using the validation set.
Classical regression model
Mesotten et al. (2002), Gunst et al. (2003) and Mesotten et al. (2003)
calculated 30 linear regression and multiple linear regression formulae (for
F and M 17 and 13, respectively) to assess age based on the developmental
stages of the third molars. This assortment of formulae was proposed
because many subjects lacked one or more third molars, and, due to the high
correlation between the different third molar stages, two stages at most could
be used as predictors. Here, a more parsimonious strategy is proposed.
Thevissen et al. (2009) detected no evidence for a left-right asymmetry in
third molar development in a selected Thai sample. Therefore, the stages of
the left molars, unless they were missing, were arbitrarily chosen as
predictors. Three regression models were designed for M and F separately.
One model pertains to the situation where information is available in both
jaws, and the two other models apply when there is information on only one
jaw (upper or lower). In the current study, a side symmetry in third molar
development was confirmed, so the number of regression models is likely
restricted. Following Chaillet and Demirjian (2004) and Chaillet et
al.(2004a,b), the inappropriate assumption of a linear relationship between
stages and age is relaxed using a cubic function for the upper and/or the
lower stage. The models are fitted using the PROC REG procedure in the
statistical package SAS, Version 9.1 (SAS Institute Inc., Cary NC, USA).
70
Regression analyses versus Bayesian approach
Bayesian approach
In the classic regression model, the molar stages serve as predictors to model
the variability in age. The distribution and variability of the molar stages and
the correlation between them are not modelled. The distribution of the molar
stages is even irrelevant to the present concern, except for the correlation
that, as has been indicated, can induce multicollinearity problems. As such,
the approach is straightforward from a computational point of view since the
response in the model (i.e., age) remains univariate, irrespective of the
number of stages used as predictors. In the Bayesian approach, the most
important factors are the likelihoods
f(x i1 ,..., x i4 | age i )
that, combined with prior information about age, yield a posterior age
distribution. To obtain these likelihoods, a model is needed for
( x i1,..., x i4 ) ,
and we are faced with a multivariate instead of a univariate response. The
challenge of the Bayesian approach in this setting is the construction of a
model for the multivariate distribution of the stages conditional on age, i.e.,
the likelihood function. Note that each stage represents an ordinal variable.
Therefore, we propose the use of a multivariate ordinal regression model
with a random subject effect (Hedeker and Gibbons, 1994) to incorporate the
correlation between the third molar developmental stages of the subject.
Fitting the multivariate ordinal regression model such that the likelihoods
can be obtained is computationally intensive. Details on this model and on
the calculation of the posterior probabilities are given in Appendix A.
Various quantifications have been used to compare the performance
of the classic regression and the Bayesian approach. First, the difference
between the predicted and the true age is calculated. The smaller this
difference is, the more accurate the model. Second, the precision of the
prediction is reflected by the width of the 95% confidence intervals (CI).
Obviously, the precision should be realistic so, when using 95%CI, only 5%
of the observed ages should fall outside the CI. If this percentage is higher,
this might reflect either a systematic bias in the point prediction or a
prediction interval that is too optimistic (too narrow). This property is
referred to as coverage. Third, the Pearson correlation between age and the
difference between the true and the predicted age is used to quantify the
degree of bias. The stronger the correlation is, the more the age that young
individuals will be overestimated. Finally, the posterior probabilities to be a
major and their corresponding diagnostic indices (sensitivity, specificity) are
compared.
In the classic approach, a regression model with only linear terms
and a model with a third degree polynomial function are used. In the
remainder, the latter regression model will be called the polynomial model.
71
Regression analyses versus Bayesian approach
RESULTS
Unless otherwise stated, the results from the classic approach are based on
the polynomial model.
Figure 8 shows the age distribution observed at different
developmental stages. Clearly, the shape of the age distribution differs for
the various stages. While the variability (e.g., the height of the boxes) does
not strongly differ, the skewness does: for early stages, the age distribution is
right-skewed (i.e., a tail towards higher age values). For late stages the
opposite holds. The line representing the trend in the relationship between
the stages and age illustrates the inappropriate assumption of linearity
[Fig.8].
In more than 98.5% of the sample, the difference between the
developmental stage of the left and the corresponding right third molar were
less than or equal to 1. A paired Wilcoxon test did not reveal any systematic
differences between the stages on the left or the right side (p = 0.22).
The polynomial model as well as the multivariate ordinal model
offer evidence for a difference between M and F (p<0.0001 for both
models). Note that the gender difference is assessed by comparison of a
model with all parameters gender-specific and a model with all parameters
shared between M and F (p<0.0001 for both models). In the polynomial
model, an interval of likely ages given the third molar developmental
stage(s) of new subjects is given by the percentiles of the normal distribution
Figure 8: Relation between the stage and age for male subjects.
Boxplots (whiskers pertain to minimum and maximum value) for the age
distribution observed at each of the possible stages. Stages in the upper jaw (right
stage if the left one is not available) from 991 male subjects are chosen for
illustrative purposes. Stages ≤ 5 are considered as one category. The trend line
illustrates the inappropriate assumption of linearity for the relation between the
stage and the age.
72
Regression analyses versus Bayesian approach
with the mean being a function of the stage(s) and the variance independent
of the stages. In the Bayesian approach, the interval of likely ages is obtained
from the percentiles of the posterior distribution where the shape can vary as
a function of the pattern stages. Figure 9 presents some posterior
distributions for specific stage patterns, clearly having different shapes,
obtained for M subjects. The probability of being mature corresponds to the
surface to the right of 18 years old under the posterior distribution. For
example, if an M subject has a Stage 6 for all 4 molars, the 95% range of
likely ages is (≤15; 19.8) and the probability of being mature 16.5%. Note
that the polynomial model yields a 95% prediction interval of (14.5; 20.1),
which is similar to the Bayesian approach. However, the probability of being
mature is increased to 30.3%. The following paragraphs give the results of
the systematic comparison of both approaches based on the subset of patients
between 16 and 22 years of age with a uniform age distribution.
Using the polynomial model the mean absolute difference between
the observed and the predicted age equals 1.13 years (median (Me) = 0.97,
interquartile range (IQR): 0.49-1.57). Using the median of the posterior
distribution as a point prediction for the age, the Bayesian approach yields a
comparable distribution for the mean absolute differences (Mann-Whitney U
test, p = 0.40) of 1.13 years (Me = 0.89, IQR: 0.44-1.62).
If the 95% range of likely age values obtained with both approaches
is realistic, then ideally 95% of the ages in the test dataset should fall within
this range. The same should hold for other ranges of likely values (e.g., 90%
range). With the polynomial model, 93.2% and 97.2% of the observed ages
fall within the 90% and 95% prediction intervals, respectively. Also with the
Bayesian approach the obtained ranges of likely age values are slightly too
wide: 96% and 98.6% of the observed ages fall in the 90% and 95%
prediction intervals. However, within the Bayesian approach, the width of
the range of likely values will depend not only on the amount of information
used (the more stages available, the narrower the prediction interval) but also
on the degree of agreement between the various stages. More specifically, it
is expected that, with the Bayesian approach, the range of likely ages will be
narrower if the stages within a jaw correspond with each other. To illustrate
this, the mean width of the 95% prediction interval is 6.34 years (Me = 6.60,
IQR: 6.0-7.0) if the left and right stages are not equal in either the lower or
the upper jaw (N = 169). The width of the 95% prediction interval is
significantly (p = 0.0002) reduced when the left and right stages agree in
both jaws: the mean width of the 95% prediction interval equals 5.95 years
(Me = 6.0, IQR: 5.3-6.8).
73
Regression analyses versus Bayesian approach
The bias as a function of age is clearly present when using the
polynomial model. There is a strong (Pearson r = 0.66) positive correlation
between age and the difference between the predicted and the observed age,
which implies that the age of younger subjects is systematically
overestimated. This bias is reduced with the Bayesian framework using the
Figure 9: Posterior distribution of male subjects for different stage patterns.
The upper panel represents all possible homogeneous stage patterns (four times
the same stage). The right-skewed distribution of age for teeth in the lowest
developmental stage with all four stages equal to five or lower (5555) smoothly
evolves to a left-skewed age distribution when all of the third molars are fully
developed (10101010). The lower panel shows the subtle differences in age
distribution when the stage of one third molar changes one unit.
74
Regression analyses versus Bayesian approach
median (r = 0.38) and the modus (r = 0.01) from the posterior distribution as
the point prediction.
Using the posterior probabilities to be mature (P(m)) to discriminate
between juveniles and mature subjects, we find that both approaches yield a
similar overall performance: the area under the curve (the receiver-operating
characteristic (ROC) curve) equals 0.847 for the polynomial model and
0.853 for the Bayesian approach. However, for the 260 subjects in the
verification dataset who are younger than 18 years old, the median P(m) is
0.51 (IQR: 0.40-0.72) with the polynomial model and 0.31 (IQR: 0.17-0.61)
with the Bayesian approach, which results in a stronger tendency for the
polynomial model to classify younger subjects too soon as mature.
DISCUSSION
Using linear regression models is the classic approach to estimate the age
and to discriminate between juveniles and adults using third molar
developmental stages. These models are easy to apply and can be adapted
such that non-linear relations between a stage and age are allowed. However,
the use of these models for age estimation has met with criticism. The first
shortcoming of the approach is the unrealistic assumption that, at every
combination of stages, age has the same distribution with respect to shape
and variance, which could yield inappropriate prediction intervals. Second,
arbitrary strategies are needed to handle the correlations between the stages.
Typically, two stages at most will be used so there is some loss of
information. Finally, the age of juveniles will systematically be
overestimated, which is unacceptable for young asylum seekers. To
overcome these disadvantages, the use of a Bayesian framework has been
advocated (see, for example, Prince and Konigsberg (2008); Prince et al.
(2008)). In the present study, Bayes’ rule was used to derive the distribution
of age given the four third molar stages. For the conditional distribution of
the molar stages, i.e., the challenging part of the rule, the use of a
generalized linear mixed model for ordinal data was proposed. A clearly
higher degree of flexibility was obtained with the Bayesian approach. The
posterior distributions varied in shape and variability as a function of the
various stage patterns, so the assumption of one common normal distribution
was clearly inappropriate. Also, the presence of multicolinearity is dealt with
in a natural way since the stages are not considered as predictors but as a set
of repeated responses. A random subject effect is used in the generalized
linear mixed model to capture the correlation between these responses (i.e.,
stages). An additional advantage of considering the stages as responses is
that the presence of missing stages does not generate any problems. Note
that, in the classic approach, a separate regression model needs to be built for
each possible pattern of missing information. Further, the Bayesian approach
yields confidence intervals whose width varies as a function of the amount
75
Regression analyses versus Bayesian approach
of available information and as a function of the degree of agreement
between the information.
However, the Bayesian approach comes at the cost of greater
computational complexity and it does not greatly outperform the classic
approach in general. Indeed there is neither a strong reduction of the
differences between the observed and predicted age nor any increase in
precision, and the prediction intervals do not cover the observed age
distributions more appropriately. However, the Bayesian approach does
reduce the bias typically present in the regression model approach. The age
of juveniles is less overestimated, yielding a better discrimination between
subjects older and younger than 18 years of age such that those younger than
18 will be classified correctly more often.
Aykroyd et al. (1999) avoided the use of the computationally
intensive mixed model for multivariate ordinal data. They assumed that the
observed correlation between the stages is accounted for by the age of a
subject, meaning that, given the age, the four stages are conditionally
independent. The consequence of this assumption is that the conditional
multivariate density f(x/age) can be written as a product of univariate
densities, thus avoiding the computational complexity of fitting a
multivariate ordinal model. Moreover a non-parametric approach for the
conditional distribution is applied since the observed distribution of third
molar stages as a function of pre-specified age categories is used. In a further
extension, they relaxed this strong conditional independence by using a
weaker partial conditional independence (Lucy et al., 2002). In further
research, both the generalized linear mixed model proposed in this study and
the multivariate model proposed by Lucy et al. (2002) should be evaluated to
verify if the model with higher computational burden outperforms the model
with the partially conditional independence assumptions.
Age groups less than 16 years of age and accordingly third molar
developmental stages under 5 are not included in the original data and so not
in the verification dataset. Collecting and importing a dataset including these
subjects into the Bayesian model could improve the prediction of the
probability of a subject to be older than 18 years of age.
Regardless of the approach, the resulting knowledge of the third
molar developmental stages of a subject does not strongly reduce the
uncertainty about the age. Furthermore, alarmingly high prediction intervals
(approximately 6 years, 95% prediction interval) and far from optimal
discrimination of maturity is obtained. Therefore, modern forensic age
estimation protocols for unaccompanied asylum seekers use, in addition to
the clinical findings, the evaluation of third molar developmental stages, the
ossification stages of the medial clavicle epiphysis, and the comparison of a
radiograph from the subject’s left hand with standard radiographs classified
by age (Schmeling et al., 2008) [Fig.3, Chap. 1]. Incorporation of these
additional sources of information into the Bayesian framework could be
76
Regression analyses versus Bayesian approach
considered. Because of the expected correlation, this new information should
be gathered simultaneously on each reference individual. In practice, it will
be difficult to establish a large data set of subjects with which these three
age-related variables can be examined simultaneously because ionising
medical imaging techniques need to be used in the living. Ethically,
submitting test individuals to the necessary quantity of ionization is not
justified (ICRP, 2007), and, in some countries, it is even illegal to use
ionising techniques for age-estimation purposes. A retrospective collection
of post-mortem, full body computed tomography (CT) images in the age
group of interest or the application of magnetic resonance imaging (MRI)
techniques in living sub-adults are other ways to compile ethically
acceptable collections of the necessary data. The gentlest way to incorporate
other age-related variables into a Bayesian framework would be taking into
account their partial correlation and implementing them in the model
presented by Lucy et al. (2002). It allows the implementation of ten dental
developmental stages (Köhler et al., 1994) together with a registration of the
absent third molars, five clavicular ossification stages (Schmeling et al.,
2004; Schulz et al., 2005; Schulz et al., 2008a,b), and comparisons with all
standard hand radiographs (Greulich and Pyle, 1959). One could also
consider the introduction of non-destructive dental age estimation methods
based on clinically observable variables such as the attrition (Gustafson,
1950; Solheim, 1988a; Kim et al., 2000; Prince et al., 2008) and the position
of the periodontal ligament attachment (Gustafson, 1950; Solheim, 1992b)
into the model and so to evaluate the enhanced age prediction performance.
CONCLUSION
On the basis of staged third molar development, the linear and polynomial
statistical regression analysis and a newly constructed Bayesian model were
verified and compared. Research Hypothesis 3 has to be rejected because
both models provide similar accuracy, precision and coverage in age
estimation outcomes. However, the Bayesian approach reduces the bias that
is typically present in the regression models. The age of juveniles is less
overestimated, so it yields better discrimination between subjects older or
younger than 18 years of age and implies that subjects younger than 18 years
of age are classified correctly more often. Moreover, the Bayesian model
integrates all the third molar information available. Indeed, no choices of
which third molar position or regression model to apply need to be made.
Accordingly, all the applicants for dental age examination are considered
equally.
77
Chapter 6
Collection and comparison of
13 country-specific third
molar development databases
THIS CHAPTER IS BASED ON THE FOLLOWING MANUSCRIPTS.
Human third molars development: Comparison of 9 country-specific
populations
Thevissen PW, Fieuws S, Willems G
Published in Forensic Science International
2010 Sep 10;201(1-3):102-5
Oral presentation at the annual scientific meeting of the American Academy of
Forensic Sciences, Seattle 2010
Estimating age of majority on third molars developmental stages in young
adults from Thailand using a modified scoring technique
Thevissen PW, Pittayapat P, Fieuws S, Willems G
Published in Journal Forensic Sciences
2009 Mar;54(2):428-32
Oral presentation at the annual scientific meeting of the American Academy of
Forensic Sciences, Washington 2009
TESTING RESEARCH HYPOTHESIS 4:
Differences in third molar development between country-specific sub-adult
populations exist
79
Country-specific third molar development
INTRODUCTION
Forensic dental age estimations in living individuals are requested primarily
to advise legal authorities about the age of unaccompanied young refugees
entering their country (Olze et al., 2006a; Solheim and Vonen, 2006;
Nuzzolese and Di Vella, 2008). At present, worldwide migration is a
common phenomenon, so age estimation examinations are solicited for
individuals of different geographical and biological origins. The age of
interest is the legal age of majority in the country of arrival. As such, the age
category of concern here is that of sub-adults. In this age category, dental
age estimations are mostly based on the development of third molars [Fig.1,
Chap. 1]. To generate correct and legally indisputable dental age estimates, it
has to be determined if there are differences in third molar maturation
between individuals from different populations.
Although most of the studies on third molar development and subadult dental age estimation are based on samples of populations with welldescribed and precisely defined origin, the study outcomes cannot be fairly
compared. Indeed, the incomparability originates in the differences in the
research protocols, the number, age and gender distribution of the sampled
subjects considered, the third molar positions examined, the third molar
registration technique used, the statistics generated, and the study outcomes
quantified [Table 1, Chap. 1].
The aim of the present study is to compare the degree of third molar
development (DTMD) between country-specific populations by means of
standardized collection and analysis of radiologically obtained third molar
data. The research hypothesis to be tested is that there are, indeed,
differences in third molar development between country-specific sub-adult
populations.
MATERIALS AND METHODS
In on-going research, archived panoramic radiographs, taken for diagnostic
and treatment-planning purposes, were collected from country-specific
populations.
In the first research phase, third molar developmental data were
uniformly registered and collected from nine country-specific populations:
Belgium (Be), China (Ch), Japan (Ja), Korea (Ko), Poland (Po), Saudi
Arabia (Sa), South India (In), Thailand (Th), and Turkey (Tu). For each
radiograph, the nationality, birth date and gender of the related individual
were verified by means of the official birth certificate and/or identity card. It
was also determined if all of the subjects had lived their entire lives in their
native country and had originated, within each population, from the same
biological group. In each population, F and M subjects in the age range
between 16 and 22 years were collected [Table 11a, b, Chap. 7]. Therefore
81
Country-specific third molar development
the date of radiographic exposure was also registered. The choice for
country-specific data collection allowed grouping the samples according to
biological or additional socio-geographical criteria. On all of the
radiographs, at least one third molar was present, and subjects with a history
of third molar extraction were excluded. As for the findings in Chapters 3
and 4, every available third molar was staged and scored following the 10point scoring system described by Gleiser et al.(1955) and modified by
Kohler et al. (1994) (KO). Missing third molars were registered without a
score value. The four third molar scores were registered as a four-digit
number so that the left to right order of the digits corresponded with the
position of the upper right, upper left, lower left and lower right third molar,
respectively. For each country, the staging and scoring of all third molars
were done by a different investigator. Additionally, in each country 10% of
the subjects were randomly selected and re-scored by an extra observer and
the country-specific investigator, separately one month later. A detailed
description of the data collection was published for Thailand (Thevissen et
al., 2009) and accordingly applied for each included country.
In the second research phase, four country-specific samples from
Brazil (Br), Italy (It), Malaysia (Ma) and the United Arab Emirates (Ua)
were collected and also studied [Table 11b, Chap. 7].
From all of the subjects, information of the four wisdom teeth was
compiled that preserved the ordinal character of their developmental stages
and handled the presence of missing values. An index quantifying the degree
of third molar development of each subject in the group of all of the subjects
was established. Therefore, a generalized linear mixed model for
multivariate ordinal data was fitted on the data (Thevissen et al., 2010b)
[Chap. 6]. The model contains a fixed effect for third molar position (i.e.,
upper versus lower) and a random subject effect. Fitting this model is similar
in spirit as performing a confirmatory factor analysis. For each subject, the
empirical Bayesian estimate of the random effect can be interpreted as its
score on a latent factor underlying and summarizing the developmental
stages of the four third molars, i.e., quantifying the overall degree of third
molar development (DTMD). In what follows, we will refer to this factor
score as the developmental score (DS). The DS is a normally distributed
variable (z-score) with mean and standard deviation equal to zero and one,
respectively. A DS equal to zero corresponds to a subject with an average
DTMD in the group of subjects from the nine studied countries. Due to the
exclusion of age, gender and countries as fixed effects in the model,
differences in DS between the subjects reflect differences in DTMD between
the age categories, the gender and the countries.
For M and F separately, a linear regression model with the DS as
dependent variable and age and countries as predictors were used to evaluate
differences in DTMD between countries. Tukey’s adjustments were used for
multiple comparisons between countries. An interaction between age and
82
Country-specific third molar development
countries was included in the model that allowed the differences between
countries to depend on age. Inclusion of a quadratic term for age allowed for
deviations from linearity.
In the first research phase, the nine initial country-specific
population samples were accordingly analysed. In the second phase, the
analysis was repeated on all 13 country-specific population samples, and the
four samples collected last were used to verify the findings obtained with the
nine initial samples.
RESULTS
Kappa statistics revealed no significant intra- or inter observer effects.
Table 8 presents some examples of patterns of observed third molar
scores and their resulting DS.
Pairwise average differences in DS between the countries were
maximally 0.43 and 0.52 (z-score) for F and M subjects, respectively
[Fig.10]. 55% (20/36) of the differences between countries were significant
for F and 86% (31/36) for M, at the 5% level [Table 9]. F subjects developed
significantly the most rapidly in Belgium compared to all the other countries.
Table 8: Developmental scores for some patterns of observed third molar scores
#
1
2
3
4
5
6
7
8
9
10
11
12
13
14
Observed third molar scores
DS
UR
3
3
4
5
6
7
8
9
9
10
*
*
*
9
-2.54
-2.11
-1.84
-1.34
-0.86
-0.46
-0.11
0.25
0.34
1.06
1.05
1.03
0.91
0.004
UL
3
3
4
5
6
7
8
9
9
10
10
*
*
8
LL
3
4
4
5
6
7
8
9
9
10
10
10
*
*
LR
3
4
4
5
6
7
8
9
10
10
10
10
10
*
The factor analysis provided for all possible combinations of four observed third molar
scores a corresponding developmental score (DS) representing the degree of third
molar development (DTMD). Rows 11 till 13 illustrate that the DS differed according to
the appearance and position of missing third molars. The pattern of four observed third
molar scores with a DS nearest to zero, thus corresponding to a subject with an average
DTMD in the total dataset, is listed in row 14.
#: row number, UR: upper right, UL: upper left, LL: lower left, LR: lower right, DS:
Developmental score, *: missing third molar
83
Country-specific third molar development
Figure 10: Least-squares means and 95% confidence intervals obtained from
the linear regression model assuming that differences between countries do not
depend on age
Be: Belgium, Ch: China, In: South-India, Ja: Japan, Ko: Korea, Po: Poland, Sa:
Saudi-Arabia, Th: Thailand, Tu: Turkey
In Japan, the development was the slowest, but it was not significantly
different from Poland or South India. M subjects in South India had a
significantly lower DTMD than did all the other countries. The DTMD was
84
Country-specific third molar development
Table 9: Comparison of Developmental Score between pairs of countries, irrespective
the age
Be
Ch
Sa
Th
Tu
Ko
Po
Ja
In
Male
Be
<0.0001
<0.0001
0.22
0.002
<0.0001
<0.0001
<0.0001
<0.0001
0.013
<0.0001
0.002
<0.0001
0.55
0.005
0.019
<0.0001
0.033
0.44
0.069
<0.0001
<0.0001
0.007
<0.0001
<0.0001
<0.0001
<0.0001
0.002
0.16
<0.0001
<0.0001
0.001
<0.0001
<0.0001
0.007
0.001
Ch
<0.0001
Sa
<0.0001
0.68
Th
<0.0001
0.058
0.043
Tu
<0.0001
0.035
0.002
0.11
Ko
<0.0001
0.35
0.044
0.17
0.34
Po
<0.0001
0.25
0.25
0.004
0.001
0.033
Ja
<0.0001
0.004
0.008
<0.0001
<0.0001
<0.0001
0.3
0.33
0.43
0.2
0.056
0.09
0.43
In
0.003
0.1
Female
Be: Belgium, Ch: China, In: South-India, Ja: Japan, Ko: Korea, Po: Poland, Sa: SaudiArabia, Th: Thailand, Tu: Turkey. Lower and upper off-diagonal part, refer to female
and male respectively. P-values are obtained after applying a correction for multiple
testing (Tukey’s adjustment)
significantly higher in Thailand than in all the other countries except for
Belgium. Within each country, the DS was lower for F than M subjects.
The model indicated that the differences between the countries
depend on age (p = 0.004 for F and p<0.0001 for M). Thus, the lines
depicting the change of the DS over age for each country were not parallel
[Fig.11]. No clear patterns of differences in DS could be distinguished
between them. Indeed, over the different ages, the DS differences between
the countries changed irregularly. For F, the shape of the regression lines
representing the relation between DS and age was similar for all of the
countries except for Turkey. Thus, the highest DS were detected in the
youngest and oldest ages in Turkey. In all the other age ranges, the DS was
the highest for Belgium and the lowest for Japan. For M, a constant finding
over the different ages was the almost persistently lowest DS value for India.
Further on, the differences between countries tended to be the smallest or
even non-existent (intersections) around 18 years of age. The pairwise
differences in average DS between the countries calculated over all full years
in the range between 16 and 22 years of age were significant (p<0.05) in18%
(45/252) and 32% (80/252) of the F and M subjects, respectively (results not
shown). In Table 11, for the total age range and at each full year separately,
85
Country-specific third molar development
Figure 11: Country-specific relations between age and DS obtained from the
regression model with an interaction between the 9 countries and age, for
female and male subjects separately.
In females and males no common trend in the course of the regression curves is
detected.
86
Country-specific third molar development
Figure 12: Country-specific relations between age and DS obtained from the
regression model with an interaction between 13 countries and age, for female
and male subjects separately
The regression lines representing the countries from Phase 1 are marked in grey.
For reference, the Belgian regression line is illustrated as a dotted line. The
regression lines representing the four countries from Phase 2 are shown in black.
87
Country-specific third molar development
the countries were ranked in function of their average DS. The ranking
fluctuated strongly between the age categories, again illustrating that
differences between countries did not follow a clear trend [Table 10]. Note
also that there was a lack of correspondence between the rankings observed
for F and M for the ranking patterns within a country as well as for the
difference patterns between countries.
The statistically significant differences in DS between the countries
turned out to be minor observed clinical differences. First, the difference in
DS can be expressed as a difference in age. In the slowest developing
country (Ja, F), the largest difference in average DS corresponding to a oneyear increase equals 0.46 (z-score). Hence, the maximal DS difference
between countries observed at any age (0.52 (z-score)) would correspond to
a difference of 14 months [Fig. 11]. Second, the difference in DS can be
expressed as a difference in the pattern of third molar scores. For this
purpose, the DS was compared between subjects with succeeding patterns of
equally repeated third molar scores [Table 8] (e.g., the DS of a subject with
all four KO scores equal to 4 (4444) compared with the DS of a subject with
all KO scores equal to 5 (5555)). The maximal and minimal difference in DS
was 0.81 (pattern 9999 compared to pattern 10101010) and 0.35 (pattern
7777 compared to pattern 8888), respectively. As such, the maximal
Table 10: Mean ranking of 9 countries, based on third molar development in each age
category of 1 year
Age category
Female
Be
Ch
Ja
Ko
Po
Th
Tu
Sa
In
Male
16
17
18
19
20
21
22
T
16
17
18
19
20
21
22
T
3
1
1
1
1
1
3
1
3
2
1
2
3
5
6
2
6
7
6
5
3
4
4
5
6
8
8
7
7
7
7
7
7
9
9
9
9
9
9
9
2
4
7
8
8
8
9
8
4
4
3
3
2
2
2
2
9
6
3
1
1
1
4
3
8
8
8
8
8
7
6
8
5
5
6
6
6
6
5
6
2
2
2
4
5
5
5
4
1
1
2
3
2
2
2
1
1
3
7
7
6
3
1
3
4
3
4
5
5
3
1
4
9
7
4
2
4
6
7
6
8
7
5
4
4
4
3
5
5
5
5
6
7
8
8
7
7
9
9
9
9
9
8
9
Be: Belgium, Ch: China, Ja: Japan, Ko: Korea, Po: Poland, Th: Thailand, Tu: Turkey,
Sa: Saudi-Arabia, In: South-India, 16, 17...22: corresponding age range of 1 year, T:
age range between 16-22 year. Countries are ranked from highest (1) to lowest (9)
mean developmental score (DS)
88
Country-specific third molar development
difference between countries on average DS (0.43 (z-score) for F and 0.52
(z-score) for M) is in line with differences between the aforementioned
succeeding patterns.
In the second research phase, the results of the repeated analyses of
13 country-specific samples revealed similar findings as in the first research
phase (on nine country-specific samples) [Fig. 12]. Again, a high
heterogeneity in differences in DS between the countries was detected. The
third molar development of the four new countries fit within the frame and
ranges of third molar development detected in the nine countries of the first
research phase.
DISCUSSION
The data included information about the subjects’ third molar development
registered on each available third molar according to the KO technique.
Information about missing third molars and the position of the third molars
was also compiled. The factor score analysis, compressed all this
information into a single numbered DS that established the DTMD of each
subject. In contrast to the methodology used by Mesotten et al. (2002), Gunst
et al. (2003), and Mesotten et al. (2003), a single age-related regression
model for each gender was developed. In fact, previous authors were obliged
to develop regression models depending on the occurrence of
multicollinearity between the observed developmental KO scores of
different third molars within a subject and also related to the number of third
molars available per subject.
All considerations of the DTMD in the present study concern the
maturation stage of these teeth in subjects aged between 16 and 22 years of
age. Accordingly third molar KO scores less than or equal to three had a
very low prevalence, which means that mainly third molar root development
was evaluated.
The quantification of the maximal pair wise average difference in
DS between countries expressed in age reveals that individuals with an equal
DTMD vary at most by 14 months over the various countries. The impact of
this finding on the age prediction outcomes between countries has to be
related to the variability inherent to the physiological age indicators used
(Kasper et al., 2009). Because high levels of variability in third molar
development between subjects of the same geographic and biological origin
were detected (Liversidge, 2008b), minor differences in age estimation
outcomes between individuals from different countries are to be expected. It
was also observed that third molar development, especially in M, is most
clustered for countries in the chronological age category between 17 and 19
years of age. It can be expected that the detected minimal differences in DS
will diminish the differences in age predictions between countries in the age
89
Country-specific third molar development
zone around 18 years old. The adult-juvenile discrimination based on the 18year-old threshold will be influenced accordingly.
The tests for the interaction between country and age on the pair
wise average differences in DS indicate that the DTMD among countries
varies as a function of age. These differences depend on changes in slope
between country-specific linear models and cannot be considered a shift in
DTMD value between countries. This indicates that comparisons of the
DTMD between the countries have to be assessed at well-defined times of
life.
An overall higher DTMD for M compared to F is observed within
each country at different ages. This implies that the quantity of difference in
DTMD between genders depends on the age of the subjects. In the literature,
this finding was described by several authors who evaluated populations of
countries not included (Gleiser and Hunt, 1955; Prieto et al., 2005; Meinl et
al., 2007; Martin-de las Heras et al., 2008) as well as countries integrated in
the current study: Be (Mesotten et al., 2002; Gunst et al., 2003; Mesotten et
al., 2003), Ja (Arany et al., 2004; Olze et al., 2004b), Tu (Orhan et al.,2006;
Sisman et al., 2007) and Ch (Zeng et al., 2010).
The DTMD between ethnically or biologically related individuals
could be studied, and the countries classified into groups. Since no common
developmental trends between countries were detected, the same conclusion
can be drawn considering the evaluation of associated country groups
containing information from the Caucasian (Be, Po, Tu) and the Mongoloid
(Ch, Ja, Ko, Th) populations. Interest in the future will focus on further
country-specific data collection, especially from the Negroid and Australian
groups. Olze et al.(2004a) compared Caucasian, Mongoloid and (black)
African samples and found that, at the same DTMD, the Caucasians
occupied a middle position with the Mongoloids being slower on average
and the Africans faster. Harris (2007) reported on mandibular third molar
evaluation that American blacks have a higher DTMD than do American
whites. This finding was in agreement (except for M in the latest developing
stages) with the conclusions drawn by Blankenship et al., (2007). These
results fit into the country-specific DTMD described in the current study
with the assumption that third molar timing is faster in Negroid populations
than in all the other country-specific populations. Further standardized
country-specific data collection of subjects of Negroid origin has to examine
this assumption further. It would enable one to verify the gender-specific
conclusion of Liversidge (2008b) that black girls are on average timing
earlier than black boys, which does not concur with the related general
findings of the current study.
90
Country-specific third molar development
CONCLUSION
The DTMD is summarized for each subject using a factor score. Analyses of
these factor scores reveal many significant differences between countries.
Although Research Hypothesis 4 has to be accepted, the differences in
DTMD between countries are not constant over age and vary irregularly.
Moreover, the magnitude of the differences turns out to be small. As such,
there is no evidence for important differences in DTMD between the
countries. An overall lower DTMD in F compared to M subjects was
detected in the different country-specific population samples.
91
Chapter 7
Comparison of age estimation
based on 13 country-specific
third molar development
databases
THIS CHAPTER IS BASED ON THE FOLLOWING MANUSCRIPT.
Human dental age estimation using third molar developmental stages:
Accuracy of age predictions not using country-specific information.
Thevissen PW, Alqerban A, Asaumi J, Kahveci F, Kaur J, Kim YK, Pittayapat P,
Van Vlierberghe M, Zhang Y, Fieuws S, Willems G.
Published in Forensic Science International
2010 Sep 10; 201(1-3):106-11.
Oral presentation at the International Symposium on Forensic Odontology,
International Organisation for Forensic Odonto-Stomatology (IOFOS), Leuven
2010
TESTING RESEARCH HYPOTHESIS 5:
The statistical model established on a Belgian reference sample is the most
appropriate for dental age estimation in unaccompanied minors
TESTING RESEARCH HYPOTHESIS 6:
The statistical model established on pooled country-specific reference samples
renders, in the absence of a model constructed on a country-specific reference
sample, the most accurate dental age estimation in sub-adults
93
Country-specific age predictions
INTRODUCTION
Age verification is established with a combination of methodologies based
on scientifically accepted research and legal requirements (Schmeling et al.,
2008). The law requires the age estimation examiner to apply the legal
regulations and to take into account all of the requirements for an
indisputable conclusion. As such, age estimation procedures related to
unaccompanied young refugees consider the attainment of the age of
majority set by law in the receiving country. An important element in
obtaining undisputable age assessment requires forensic investigation using
age estimation methods based on a reference sample of the same
geographical and biological origin as the examined individual. If not,
scientifically based information about the consequences of using dissimilar
information on the validity of the reported age prediction should be
provided.
In Chapter 6, it was determined that the differences in degree of
third molar development (DTMD) between countries are small and vary
irregularly. The impact of these findings on the country-specific age
prediction outcomes based on third molar development needs to be
examined. The aim of this study was to quantify the age prediction
performances based on third molar development reference data from
Belgium or pooled from all other countries and to compare them with the
quantified age performances based on country-specific reference data.
Because, in a legal context,t he benefit of the doubt must be given to the
applicant, the scientific outcomes need to be interpreted accordingly.
Therefore, two research hypotheses were tested. First, was determined if the
statistical model established on Belgian reference data was the most suited
for the dental age estimation of unaccompanied minors. Second, was
determined if the statistical model established on all the pooled countryspecific data, renders, in the absence of a model constructed on countryspecific reference data corresponding to the nationality of the examined
individual, more accurate dental age estimations in sub-adults.
MATERIALS AND METHODS
The data collected from the 13 country-specific population samples in the
on-going research described in Phases 1 and 2 of Chapter 6 were studied:
Belgium (Be), Brazil (Br), China (Ch), Italy (It), Japan (Ja), Korea (Ko),
Malaysia (Ma), Poland (Po), Saudi Arabia (Sa), South India (In), Thailand
(Th), Turkey (Tu), and United Arab Emirates (Ua). In each population
sample, subjects in the age range between 16 and 22 years old were retained
for analysis and divided at random but stratified by age in a reference and a
validation dataset. The validation dataset was used to evaluate the
95
Country-specific age predictions
Table 11a: The number of female and male subjects used in the first 7 of 13 country-specific
analyses of population-specific samples and their partition in reference and validation
datasets specified for each year interval in the range between 16 and 22 years old.
Age distribution
Female
Male
16* 17* 18* 19* 20* 21* 16* 17* 18* 19* 20* 21* Total
Be po
159
162
191
211
249
307
125
103
146
158
174
218
2203
re
80
81
96
106
125
154
63
52
74
79
87
109
1106
va
79
81
95
105
124
153
62
51
72
79
87
109
1097
Ch po
51
48
47
50
46
53
46
51
37
42
40
32
543
re
26
25
24
25
24
27
24
26
19
22
21
17
280
va
25
23
23
25
22
26
22
25
18
20
19
15
263
Sa po
52
59
44
62
59
54
54
51
48
58
58
51
650
re
26
30
23
31
30
27
28
26
25
30
30
26
332
va
26
29
21
31
29
27
26
25
23
28
28
25
318
Th po
68
62
82
68
70
69
65
64
66
76
68
63
821
re
34
31
42
34
36
35
33
32
34
38
35
32
416
va
34
31
40
34
34
34
32
32
32
38
33
31
405
Tu po
50
49
50
50
51
49
50
48
49
51
50
50
597
re
25
25
25
26
26
25
25
24
25
26
25
25
302
va
25
24
25
24
25
24
25
24
24
25
25
25
295
Ko po
54
54
55
57
53
57
52
53
63
57
55
50
660
re
28
28
28
29
27
29
27
27
32
29
28
25
337
va
26
26
27
28
26
28
25
26
31
28
27
25
323
Po po
50
46
48
44
70
84
36
45
35
25
39
30
552
re
26
24
25
23
36
42
18
23
18
13
20
16
284
va
24
22
23
21
34
42
18
22
17
12
19
14
268
Be: Belgium, Ch: China, Sa: Saudi-Arabia, Th: Thailand, Tu: Turkey, Ko: Korea, Po: Poland,
po: analyzed population sample, re: reference dataset, va: validation dataset, 16* includes all
individuals aged 16.00 to 16.99 years etc.
96
Country-specific age predictions
Table 11b: The number of female and male subjects used following 6 of 13 countryspecific analyses of population specific samples and their partition into reference and
validation datasets specified for each year interval in the range between 16 and 22
years of age
Age distribution
Female
Male
16* 17* 18* 19* 20* 21* 16* 17* 18* 19* 20* 21* Total
Ja po
52
39
48
41
37
44
39
52
49
46
40
39
526
re
27
20
25
21
19
23
20
27
25
24
20
20
271
va
25
19
23
20
18
21
19
25
24
22
20
19
255
po
44
50
53
48
27
16
37
39
32
43
25
16
430
re
22
26
27
25
14
8
19
20
17
22
13
9
222
va
22
40
24
33
26
41
23
31
13
25
8
12
18
39
19
21
15
16
21
15
12
11
7
11
208
re
20
17
21
16
13
7
20
11
9
8
6
6
va
20
16
20
15
12
5
19
10
7
7
5
51
po
95
98
92
119
131
115
85
77
84
85
95
69
re
48
50
46
60
66
58
43
39
42
43
48
35
va
47
48
46
59
65
57
42
38
42
42
47
34
Ma po
32
42
43
47
44
37
31
29
31
42
56
40
re
16
22
22
24
23
19
16
15
16
22
29
20
va
16
20
21
23
21
18
15
14
15
20
27
20
Ua po
51
49
49
51
53
57
55
53
42
59
55
38
re
26
25
25
26
27
29
28
27
22
30
28
20
va
25
24
24
25
26
28
27
26
20
29
27
18
Gl po
612
602
628
649
685
675
666
559
562
589
604
548
re
218
214
213
215
236
204
315
222
221
217
228
197
va
394
388
415
434
449
471
351
337
341
372
376
351
In
Br po
It
295
154
141
1145
578
567
474
244
230
612
313
299
7279
2600
4679
Ja: Japan, In: South-India, Br: Brazil, It: Italy, Ma: Malaysia, Ua: United Arab
Emirates, Gl: global dataset, po: analyzed population sample, re: reference dataset, va:
validation dataset, 16* includes all individuals aged 16.00 to 16.99 years etc.
97
Country-specific age predictions
performance of the model developed for the subjects in a reference dataset
[Table 11a, b]. In addition, from each country-specific reference dataset, 100
M and 100 F subjects were randomly selected and pooled (given the
differences in age distribution between the countries, stratification by age
was not feasible). This pooled dataset will henceforth be called the global
reference set. The global validation set assembled all of the subjects from the
country-specific validation datasets. Bayes’ rule was applied to obtain age
predictions for each possible combination of the four KO third molar scores.
The details of this approach have been described above in Chapter 4
(Thevissen et al., 2010b). Briefly, our interest here is in the distribution of
age given a specific pattern of scores, the so-called posterior distribution. To
obtain this distribution, one needs to specify the multivariate distribution of
the ordinal scores given the age (the conditional distribution) and the
distribution of the age (the prior distribution). For the conditional
distribution of the scores for a given age, a generalized linear mixed model
for multivariate ordinal data is used in each reference dataset. A uniform
distribution within the age range of 16-22 years old is used as the prior
distribution. The 50th percentile of the posterior distribution is used as the
point prediction [Chap. 4].
The country-specific models were verified using the respective
country-specific, the Belgium, and the total test datasets. The difference
between the observed and the predicted age was calculated in each validation
dataset and the mean absolute difference (MAD) as well as the mean squared
error (MSE) were used to quantify the performance. To understand the
difference between these quantifications, note that larger differences receive
relatively more weight in the MSE than in the MAD. All the analyses were
performed for the F and the M subjects separately.
The juvenile and adult distinction (set at the threshold of 18 years of
age) was studied by calculating the percentages of correctly identified adults,
correctly identified juveniles, and correctly identified subjects. Exact
McNemar tests were used to compare these percentages between the various
approaches.
All the analyses were performed using SAS software, Version 9.2 of
the SAS System for Windows. Copyright © 2002 SAS Institute Inc. SAS
and all other SAS Institute Inc. products or service names are registered
trademarks of SAS Institute Inc. (Cary, NC, USA).
RESULTS
Within the range from 16 to 22 years of age, 10,185 subjects (4,668 M,
5,517 F) were analysed. 37,750 (17,335 M, 20,415 F) third molars were
scored, which means that 7.12% of the third molars were absent. The
number of F and M subjects included for each country and the global dataset
98
Country-specific age predictions
were listed separately for every reference and validation dataset per one-year
age range [Table 12a, b].
The MAD obtained using country-specific information, information
from Belgium, and information from the global dataset varied between 0.85
and 1.30 years of age, 0.88 and 1.35 years of age, and 0.87 and 1.28 years of
age, respectively. For the MSE, these ranges were between 1.19 and 2.33
years of age, 1.29 and 2.80 years of age; and 1.19 and 2.48 years of age,
respectively.
The information from Belgium compared to the information used
from the specific country increased the MAD on average only 0.87 months
with a maximal MAD increase of 2.60 months observed for F in Brazil. For
some countries, the MAD even decreased. The MSE was increased on
average with 0.31 months.
Information from the global dataset compared to country-specific
information altered on average the MAD with 0.34 months and the MSE
with 0.12 months. Maximal increases were 1.00 and 0.40 months,
respectively. Almost an equal number of situations were observed with
decreased MAD or MSE values compared to increased values [Fig 13a, b;
Fig. 14a, b].
In a specificity test, the percentage of correctly identified juveniles
was evaluated. Consequently, the proportion of incorrectly identified
juveniles (= correctly identified adults) provided information on sensitivity.
Accuracy was tested observing the proportions of correctly identified
juveniles and adults. Among the evaluated countries the specificity,
sensitivity and accuracy ranges were 40.4% to 85.3%, 66.7% to 83.2% and
71.0% to 85.0%, respectively [Fig.15, 16, 17]. There is no indication at all
that not using country-specific information influences the percentage of
correctly identified subjects. However, using information from Belgium
leads to a higher percentage of correctly identified juveniles at the price of a
lower percentage of correctly identified adults. Note that this phenomenon is
more outspoken for F, i.e., the observed differences are clearer and more
often significant. The effect of the use of information from the total dataset
on the percentages is less clear. In most situations, there are no significant
differences between the use of the total dataset and country-specific
information. The significant differences are both positive as well as negative.
DISCUSSION
The size of the study sample needs to be considered in relation to the
research aim. In the current study, the differences in age prediction
performances between country-specific samples and the related populations
are sought. As an example, the calculated sample size needed to detect
clinically a difference of 6 months with 80% power between M from
Belgium and China having the same pattern of third molar scores would be
99
Country-specific age predictions
Figure 13a: Mean absolute difference, based on country-specific information,
information from Belgium and information from all countries pooled (Global
dataset) calculated for male subjects.
Be: Belgium, Br: Brazil, Ch: China, It: Italy, Ja: Japan, Ko: Korea, Ma:
Malaysia, Po: Poland, Sa: Saudi-Arabia, In: South India, Th: Thailand, Tu:
Turkey, Ua: United Arab Emirates, T = Significant (p<0.05, paired t-tests)
difference in mean absolute difference (MAD) obtained when using global dataset
compared to country-specific information. B = Significant (p<0.05, paired t-test)
difference in MAD obtained when using Belgian dataset compared to countryspecific information. The MAD obtained with information from Be is higher than
the country-specific MAD except for Br, Ko, Th and Tu. The lines connecting the
reported MAD points were drawn to illustrate the trend per information group.
The MAD values from the global dataset were most like the MAD values obtained
from country-specific information. Four differences in MAD were statistically
significant with low actual values.
143 patients in each country (based on a two-sided t-test with alpha = 5%
and assuming a SD of 1.5 years). However, the required sample size will
differ according to the objective of the considered analysis: more Chinese
subjects are needed to detect differences with Koreans than to detect
differences with Belgians. Therefore, when taking into account the research
aim, the planned sample size will be based more on practical limitations than
on statistical considerations. Consequently, in this study, the pragmatic rule
was applied, and the intention was to collect in each country 50 F and M
subjects within each considered age range of 1 year. During data collection,
a substantial difference appeared between the countries in their number of
subjects and in their age distribution. Trying to perform the ideal distribution
set up, which includes an equal number of F and M subjects for each age
100
Country-specific age predictions
Figure 13b: Mean absolute difference, based on country-specific information,
information from Belgium and information from all countries pooled (Global)
calculated for female subjects
Be: Belgium, Br: Brazil, Ch: China, It: Italy, Ja: Japan, Ko: Korea, Ma:
Malaysia, Po: Poland, Sa: Saudi-Arabia, In: South India, Th: Thailand, Tu:
Turkey, Ua: United Arab Emirates, T = Significant (p<0.05, paired t-tests)
difference in mean absolute difference (MAD) obtained when using global dataset
compared to country-specific information. B = Significant (p<0.05, paired t-test)
difference in MAD obtained when using Belgian dataset compared to countryspecific information. The MAD obtained with information from Be is higher than
the country-specific MAD except for Br, Ko, In and Tu. The lines connecting the
reported MAD points were drawn to illustrate the trend per information
group.The MAD values from the global dataset were most like the MAD values
obtained from country-specific information. Nine differences in MAD were
statistically significant with low actual values.
range of 1 year would have reduced the amount of information in most
countries. Consequently, all the subjects within the age range between 16
and 22 years of age are included. In further research, in each country, the
number of subjects will be adjusted to obtaining the numbers prescribed in
the pragmatic rule and enabling one to evaluate these results with respect to
the current findings.
The evaluation based on the four repeated third molar scores renders
information about third molar development in all third molar positions. It
also integrates information related to the frequent occurrence of agenesis of
third molars (Garn et al., 1963; Levesque et al., 1981; Baba-Kawano et al.,
2002; Rozkovcová et al., 2004;Callahan et al. 2009; Nieminen, 2009; De
Coster et al., 2009). The observed missing third molars could be considered
agenetic because only subjects without a history of third molar extraction
101
Country-specific age predictions
were included and also because the initial third molar development is
observed at the latest when the second molar development has reached the
stage of ¾ of the completed root length (Liversidge, 2008a). This second
molar maturation stage corresponds to a chronological age of around 14
years old. Therefore, in the studied age category (16-22 years of age),
missing third molar development information has to be diagnosed as
agenesis. The finding that 7.12% third molars were absent in the current
sampling does not provide accurate information about the prevalence of third
molar agenesis. Indeed, in the current study, subjects with four missing third
molars were excluded, and no information was reported concerning the
number of subjects missing a specific number of third molars.
Information from Belgium increased the MAD at most by 2.6
Figure 14a: Mean squared error, based on country-specific information,
information from Belgium and information from all countries pooled (Global)
calculated for male subjects
Be: Belgium, Br: Brazil, Ch: China, It: Italy, Ja: Japan, Ko: Korea, Ma:
Malaysia, Po: Poland, Sa: Saudi-Arabia, In: South India, Th: Thailand, Tu:
Turkey, Ua: United Arab Emirates, T = Significant (p<0.05, paired t-tests)
difference in mean squared error (MSE) obtained when using global dataset
compared to country-specific information. B = Significant (p<0.05, paired t-test)
difference in obtained when using Belgian dataset compared to country-specific
information. The MSE obtained with information from Be is higher than the
obtained country-specific MSE, except for Br, Ko, and Tu (with equal MSE). The
lines connecting the reported MSE points were drawn to illustrate the trend per
information group. The MSE values from the global dataset were most
approaching the MSE values obtained with country-specific information. Eight
differences in MSE were statistically significant, with low actual values.
102
Country-specific age predictions
months above the information provided from the country itself. Using
information from the total dataset reduces the maximum added error to 1.0
month. This is a relatively small increase compared to the discovered
maximum country-specific MAD of 1.30 years (Brazil F) and can be
concluded that the ascertained aging results based on country-specific
information do not overrule the predicted age outcomes obtained with
information from Belgium or the global dataset. This implies that, if countryspecific information is absent, information from Belgium or combined
countries can be used with an increased error of at least 2.6 and 1.0 months,
respectively. The constructed models also permit the calculation of the exact
age differences to consider, if information of another studied country instead
of Belgium would be needed.
Figure 14b: Mean squared error, based on country-specific information,
information from Belgium and information from all countries pooled (Global)
calculated for female subjects
Be: Belgium, Br: Brazil, Ch: China, It: Italy, Ja: Japan, Ko: Korea, Ma:
Malaysia, Po: Poland, Sa: Saudi-Arabia, In: South India, Th: Thailand, Tu:
Turkey, Ua: United Arab Emirates, T = Significant (p<0.05, paired t-tests)
difference in mean squared error (MSE) obtained when using global dataset
compared to country-specific information. B = Significant (p<0.05, paired t-test)
difference in obtained when using Belgian dataset compared to country-specific
information. The MSE obtained with information from Be is higher than the
obtained country-specific MSE, except for Br, Ko, In and Tu. The lines connecting
the reported MSE points were drawn to illustrate the trend per information group.
The MSE values from the global dataset were most approaching the MSE values
obtained with country-specific information. Twelve differences in MSE were
statistically significant, with low actual values.
103
Country-specific age predictions
Figure 15: Percentage of correctly identified juveniles, based on countryspecific information, information from Belgium and information from all
countries pooled (global).
M upper panel, F lower panel, Be: Belgium, Br: Brazil, Ch: China, It: Italy, Ja:
Japan, Ko: Korea, Ma: Malaysia, Po: Poland, Sa: Saudi-Arabia, In: South India,
Th: Thailand, Tu: Turkey, Ua: United Arab Emirates,T=Significant (p<0.05,
McNemar test) difference between percentage obtained when using global dataset
compared to country-specific information. B=Significant (p<0.05, McNemar test)
difference between percentage obtained when using Belgian dataset compared to
country-specific information. Compared to information used of the own country
and the total dataset, information obtained from Belgium results for all countries
in more correctly classified juveniles. These observations are less explicit for
males compared to females.
104
Country-specific age predictions
Figure 16: Percentage of correctly identified adults, based on country-specific
information, information from Belgium and information from all countries
pooled (global).
Male upper panel, Female lower panel, Be: Belgium, Br: Brazil, Ch: China, It:
Italy, Ja: Japan, Ko: Korea, Ma: Malaysia, Po: Poland, Sa: Saudi-Arabia, In:
South India, Th: Thailand, Tu: Turkey, Ua: United Arab Emirates, T = Significant
(p<0.05, McNemar test) difference between percentage obtained when using
global dataset compared to country-specific information. B = Significant (p<0.05,
McNemar test) difference between percentage obtained when using Belgian
dataset compared to country-specific information. Compared to information used
of the own country and the total dataset, information obtained from Belgium
results for all countries in less correctly classified adults. These observations are
less explicit for males compared to females.
105
Country-specific age predictions
Liversidge et al.(2006) found, based on a meta-analysis of earlier published
data from eight countries (Australia, Belgium, Canada, England, Finland,
France, South Korea, Sweden) of information on the development of all left
mandibular permanent teeth (except third molars), a lack of consistent
Figure 17: Percentage of correctly classifiedsubjects, based on country-specific
information, information from Belgium and information from all countries
pooled (global).
Male upper panel, Female lower panel, Be: Belgium, Br: Brazil, Ch: China, It:
Italy, Ja: Japan, Ko: Korea, Ma: Malaysia, Po: Poland, Sa: Saudi-Arabia, In:
South India, Th: Thailand, Tu: Turkey, Ua: United Arab Emirates. The
percentage correctly classified subjects are for each gender and for the three
different approaches overlapping considerably.
106
Country-specific age predictions
population difference in the timing of dental formation. Braga et al. (2005)
collected three geographically dispersed population samples (European,
Asian, African) and compared their age predictions with a sample of French
children (having at least one grandparent not originating from Europe). Their
results indicate that non-adult age predictions on the dental mineralization
sequences of the seven left mandibular permanent teeth do not guarantee
better predictions than the geographically specific estimates. These results
are completely in line with the findings in the current study. However, in
studies comparing different populations based on third molar development,
Liversidge (2008b), Harris (2007) and Blankenship et al. (2007) found
significant evidence of earlier third molar development in black populations
than in white populations. Martin de las Heras et al. (2008) reported slower
third molar mineralization when comparing a Spanish with a Magrebian
population, Olze et al. (2003) detected significant differences in the
chronology of third molar mineralization between German and Japanese
populations, and Kasper et al. (2009) concluded on the comparison of
findings from an Hispanic population and the results of Mincer et al. (1993),
from a Caucasian population that third molar development is more rapid in
Hispanics. These differences in third molar development allow one to
presume that using third molar development information from another
country for the prediction of an individual’s age would result in estimations
with much greater margins of errors. In the current study, each countryspecific validation yielded wide country-specific age prediction intervals due
to the high degree of inherent human variability of third molar development
within specific populations (Thevissen et al., 2010b,c),. Accordingly the
additional error in age prediction made using information from another
country was relatively small.
The quantification of the differences in estimated age between
countries was possibly affected by fluctuations in the sample size between
the evaluated countries. This probably ignorable influence on the current
results will have to be checked by means of additional data collection and
sample resizing in the future.
The population differences found by Liversidge (2008b) were
described as the result of an earlier initiation and completion of third molar
maturation in a Black South African population and suggest a shift in timing
of initiation relative to that of the other investigated populations. Harris
(2007) reported unequal temporal differences between all morphological
stages among all of the races considered. Blankenship et al. (2007) detected
in Blacks an earlier root development compared to Whites and described
more complex variations between both groups in the final developmental
stages. The analysis of Martin de las Heras et al. (2008) started with the
stage of crown completion. Olze et al. (2004b) observed significant
differences between Japanese and Germans in the stages between crown
completion and root length equal to or greater than crown height. Based on
107
Country-specific age predictions
these reports, one can conclude that population-related third molar
maturation differences occur during each part of the maturation sequence.
Therefore, in the present study, limiting the subjects to the 16-22 year age
category and thus restricting the study to the late third molar developmental
sequence did not interfere with the observation of related differences
between the populations.
Separate evaluation of the age performances for F and M subjects
revealed gender-related outcome differences based on country-specific,
Belgian and total reference information. No overall gender-specific trends
were detected. Therefore, gender-specific models should be applied during
age examination procedures in order to obtain scientifically correct and
legally indisputable age predictions. Because the diagnostic value of the
evaluated performances is related to living individuals with known sex, it
was not relevant to our present concern to determine possible faults made by
exchanging the gender in the present study. Furthermore, pooling F and M
subjects to establish a reference when the sex of the examined individual is
not known was, for the same reason, not investigated.
Using information from Belgium resulted in less sensitivity in
juvenile discrimination than did using country-specific information. Indeed
age estimates based on the Belgium reference model are under estimated,
which implies that fewer subjects older than 18 years (adults) were detected.
Moreover, the specificity of juvenile discrimination is higher because more
subjects younger than 18 years old were found in the under-estimated age
outcomes obtained from the Belgium reference model. When discriminating
juveniles from adults, in particular, the Belgium information provides an
excellent judicial reference because its lower sensitivity and related higher
specificity in juvenile discrimination results in a higher number of selected
juveniles. Legally, this can be interpreted as a benefit of the doubt provided
to the individual under examination. Therefore, Research Hypothesis 5 can
be retained. In the absence of a country-specific reference model, the
Belgian reference model is legally most suited for forensic dental age
discrimination of unaccompanied minors. In general, using information from
all the countries taken together revealed an accuracy in juvenile adult
discrimination that corresponds to the observed country-specific proportions.
Consequently, the model based on the reference sample combining all of the
countries is the most accurate substitute for the country-specific reference
model in sub-adult age discrimination. Thus, Research Hypothesis 6 can be
accepted [Fig. 14a, b].
CONCLUSION
Verification with 13 country-specific databases using information from
Belgium or all of the countries pooled together changes the difference
108
Country-specific age predictions
between observed and predicted age obtained on country-specific
information only slightly. For the adult-juvenile discrimination, information
from Belgium provides an overall better legal reference, and information
from all the countries pooled provides similar results compared to the
outcomes based on information from each specific country. As such,
Research Hypotheses 5 and 6 were accepted. In the absence of a countryspecific reference model, the Belgium reference model provides the best
benefit of the doubt during forensic dental age estimation examinations of
unaccompanied young refugees. The reference model based on all the pooled
countries replaces the country-specific reference model most accurately for
sub-adults.
109
Chapter 8
Influence of tooth
morphological age predictors
on age estimation based on
third molar development
THIS CHAPTER IS BASED ON THE FOLLOWING MANUSCRIPT.
Human dental age estimation combining third molar(s) development and tooth
morphological age predictors.
Thevissen PW, Galiti D, Willems G
Published in International Journal of Legal Medicine
2012 Nov;126(6):883-7
Oral presentation at the annual scientific meeting of the American Academy of
Forensic Sciences, Atlanta 2012 and at the Fünfzehnte Treffen der
Arbeitsgemeinschaft für Forensische Altersdiagnostik (AGFAD). Berlin 15 03
2012
TESTING RESEARCH HYPOTHESIS 7:
In sub-adults, the accuracy of age estimations based on third molar
development is improved by adding age-related information from tooth
morphological age predictors
111
Third molar and tooth morphological predictors
INTRODUCTION
Dental age estimation in the sub-adult age group is based mainly on third
molar development observed in panoramic radiographs [Table 1]. These
radiographs provide morphological age-related dental information. In fact,
all of the permanent teeth are mature in the period of late third molar
development so that they have closed apices (Liversidge et al., 2010) and, by
definition, secondary dentine formation has commenced (Benzer, 1948;
Philippas and Applebaum 1966; Moore, 1970; Solheim, 1992a). The amount
of secondary dentine apposition was observed, measured and quantified in
both peri-apical (Kvaal et al., 1995; Sharma and Srivastava, 2010; KanchanTalreja et al., 2012) and panoramic radiographs (Bosmans et al., 2005;
Paewinsky et al., 2005; Meinl et al., 2007; Landa et al., 2009; Erbudak et al.,
2012). The quantifications were related to age and modelled for age
estimation purposes by Kvaal et al. (Kvaal et al., 1995). These findings
allow one to combine different dental variables observed on a particular
diagnostic tool for age estimation.
The aim of this study is to analyse in sub-adults the age predicting
performances of adding tooth morphological measurements from permanent
teeth to the developmental stages of third molars as evaluated on panoramic
radiographic data.
MATERIALS AND METHODS
Digital panoramic radiographs from 450 different individuals were
retrospectively collected from the dental clinic files of the Katholieke
Universiteit Leuven, Belgium. The individuals had Belgian nationality and
were of Caucasian origin. For each gender, 25 radiographs were selected
within each age category of one year in the range between 15 and 23 years.
On each selected panoramic radiograph, at least one third molar was present,
and the image quality allowed one to measure the length and width of the
monoradicular teeth and their pulp chambers. The selected individuals had
no medical history that could have influenced tooth development and no
history of tooth extraction.
The development of all the available third molars was classified and
registered with the 10-point staging and scoring technique described by
Köhler et al. (1994) (KO). The Kvaal et al. (1995) measuring technique was
used for teeth on the left side. In particular, the upper central and lateral
incisor and the second premolar as well as the lower lateral incisor, the
canine, and the first premolar were considered. If, due to tooth positioning,
tilting, or overlapping, insufficient tooth information was available, the
corresponding tooth on the right side was measured. The lengths and widths
of tooth and pulp were measured. Their ratios, mean ratios (L, W, MK), and
difference of ratios (W-L) were calculated separately for each tooth for all
113
Third molar and tooth morphological predictors
Figure 18: Measurements according the Kvaal technique performed in image
improvement software.
To obtain optimal measurements the panoramic radiographs were imported in
Adobe Photoshop CS4®. The images were zoomed 300% and arbitrarily rotated
to be parallel to the left (or right) working canvas side. Guides were dragged at
the selected tooth points, and the measurements were made using the
measurement tool snapped to the guides. The left panel illustrates the horizontal
guides placed for the length measurements of tooth # 33: T = total tooth length, P
= pulp length, R = root length. The right panel illustrates the vertical guides
placed for the width measurements at the level of the cementum enamel junction
of tooth # 33: A = root width, A’ = pulp width
the upper, the lower, and all six teeth. The staging and measuring was
performed in image enhancement software (Adobe Photoshop CS4, Adobe
Systems Incorporated, San Jose, CA, USA) [Fig. 18].
All the radiographs were staged and measured by one observer.
After a month, 20 radiographs were randomly selected from the sample and
re-evaluated by the same as well by as a second observer. Linear regression
models with age as response and scored third molar developmental stages as
explanatory variables were developed. To these models, MK and W-L the
measurement ratios were added for the six teeth separately, the upper, the
lower, and all six teeth together. From the models' determination coefficients
(R2) and root mean squared errors (RMSE) were calculated. The R2
calculation indicates the predictive value of the set of explanatory variables:
the higher the R2, the more variance in age is explained by these variables.
Smaller RMSE’s denote minor differences between the predicted and the
chronological age. The analyses were performed on the entire group and
114
Third molar and tooth morphological predictors
separately for M and F. Pearson correlation between the four third molar
stages showed strong relations [0.86-0.93]. This relation was strongest
between the molars of the same arch. Therefore, multicollinearity problems
in the regression models were reduced by using third molar stages of one
side. For standardization, the left side was chosen. If a left third molar was
missing, the score of the corresponding right third molar was used. All
analyses were done using the SAS software, Version 9.2 of the SAS system
for windows (SAS statistical software, SAS Institute, Cary, NC, USA).
RESULTS
High intra- and inter-observer reliabilities were obtained for both the third
molar staging (84% perfect agreement) as well as the tooth measurements
(maximal difference 2%).
For the combined F and M sample, the regression model, including
only third molar stages provided an R2 of 60% and an RMSE of 1.63 years
[Table 12]. Adding to this model, Kvaal ratios (MK, W-L) of one tooth,
maximally increased R2 with 1% (Tooth #22, 61%) and maximally decreased
RMSE with 0.02 years (Tooth #22, 1.61 years). Adding to the same model,
Kvaal ratios of the upper or lower teeth increased R2 at most 1% (upper teeth
together, 61%) and decreased RMSE at most 0.01 years (upper teeth
together, 1.62 years). Added information of all six teeth together increased
R2 by 1% (61%) and decreased RMSE by 0.02 years (1.61 years).
Similar analyses performed on the M sample resulted in analogue
Table 12: R² and RMSE calculated from the third molar regression model and the
multiple regression models combining third molar information and information based
on Kvaal et al. (1995)
Regression model
TM
TM + 21
TM + 22
TM + 25
TM + 34
TM + 33
TM + 32
TM + U
TM + L
TM + U+L
R²
0.60
0.60
0.61
0.61
0.61
0.60
0.61
0.61
0.61
0.61
M+F
RMSE
1.63
1.64
1.61
1.62
1.63
1.63
1.63
1.62
1.62
1.62
R²
0.70
0.70
0.71
0.70
0.70
0.70
0.70
0.70
0.70
0.70
M
RMSE
1.43
1.43
1.42
1.43
1.43
1.43
1.43
1.43
1.43
1.43
R²
0.52
0.53
0.56
0.56
0.54
0.54
0.56
0.57
0.56
0.58
F
RMSE
1.78
1.78
1.71
1.72
1.76
1.76
1.73
1.69
1.72
1.68
TM: third molar; M: males; F: females; 21, 22, 25, 34, 33, 32, U, L, U+L: Kvaal ratios
of too(ee)th 21, 22, 25, 34, 33, 32, 21+22+25, 34+33+32, 21+22+25+34+33+32 (FDI
standard) respectively
115
Third molar and tooth morphological predictors
increases of the R² and comparable decreases of the RMSE values [Table
12].
The largest added value of age predicting information was detected
in the regression analyses performed on the F sample. In fact, adding the
Kvaal’s ratios of all six teeth increased R² by 6% and decreased RMSE by
0.10 years [Table 12].
R² and RMSE values from the regression models, including only
Kvaal’s ratios ranged between 0.1% and 29% and 2.21 years and 2.60 years,
respectively [Table 13].
DISCUSSION
In the present study, models combining third molar developmental
information with morphological dental variables resulted in a maximal
increase of explained variance in age of 6% and a maximal decrease of 0.1
year in RMSE compared to models based only on third molar(s)
development. On average, the 9 studied models combining developmental
and morphological variables disclosed ignorable and clinically insignificant
differences compared with the corresponding third molar models. This
finding reflects the poor dental age-related morphological information
available in the studied age range. Indeed the explained variance in age
detected in the models based on tooth morphology varied between 0.1 and
29% and the RMSE were between 2.21 and 2.60 years [Table 13]. The cause
of these inferior age-related performances could be explained by the lack of
ample amounts of secondary dentine formed in this age category. Indeed,
Philippas et al. (1966) studied secondary dentine formation in 14 age groups
Table 13: Determination coefficient (R²) and Root Mean Squared Error (RMSE)
calculated from the third molar development and the Kvaal et al. (1995) regression
models.
Regression model
TM
21
22
25
34
33
32
U
L
U+L
R²
0.60
0.03
0.06
0.10
0.06
0.06
0.04
0.11
0.09
0.13
M+F
RMSE
1.63
2.57
2.53
2.47
2.54
2.53
2.56
2.46
2.50
2.44
R²
0.70
0.001
0.02
0.05
0.03
0.02
0.01
0.03
0.03
0.05
M
RMSE
1.43
2.60
2.58
2.55
2.57
2.58
2.59
2.57
2.57
2.55
R²
0.52
0.09
0.15
0.19
0.13
0.16
0.14
0.25
0.24
0.29
F
RMSE
1.78
2.50
2.40
2.36
2.45
2.41
2.43
2.26
2.29
2.21
TM: third molar model; M: males; F: females; 21, 22, 25, 34, 33, 32, U, L, U+L: Kvaal
model containing calculated measures of too(ee)th 21, 22, 25, 34, 33, 32, 21+22+25,
34+ 33+ 32, 21+22+25+34+33+32 (FDI standard) respectively
116
Third molar and tooth morphological predictors
of 5 year, starting at the age of 6 years, and concluded that with beginning of
the 21 to 25 year group there was a gradual increase in the amount of
irregular secondary dentine formation, attaining a pronounced increase in the
46 to 50 year group. Moreover the teeth considered by Philippas et al.
(1966), were upper central incisors. From all developing permanent
monoradicular teeth, incisors are maturing earliest and thus are advanced in
secondary dentine formation. In the current study, beside incisors, canines
and premolars were measured, implicating that within this group of
combined tooth types the mean threshold of beginning gradual increase of
secondary dentine formation has to be set at older ages (>21 to 25 year
group). Furthermore, it has to be taken into consideration that in the
Phillippas et al. (1966) study, secondary dentine formation was
microscopically evaluated on sectioned teeth with magnifications up to 200
times. In the present study, this initial secondary dentine formation was
measured on panoramic radiographs 3 times magnified in Adobe Photoshop
CS4®, (Adobe Systems Incorporated, San Jose, CA, USA). Associated with
the knowledge that on radiographs no distinction can be made between
primary and secondary dentine, it has to be concluded that certainly on
panoramic radiographs, initial secondary dentine formation is hardly or even
not measurable when evaluating ratios of tooth lengths and root-pulp widths.
The R² values reported in the Kvaal et al. (1995) study were
outperforming (56%<R²<76%) compared to the obtained R² values
evaluating the tooth morphological variables in the current study
(0.1%<R²<29%) [Table 13]. The major difference in research set up between
both studies concerns the age range and distribution of the investigated
sample. The age range of the reference sample in the current study was
restricted to young individuals (15-23 year). In the Kvaal et al. (1995) study
adult individuals were sampled (20-87 year). Although in the present study,
a bigger sample size with a more homogenous gender and age distribution
was used, this couldn’t compensate for the poor variance in age explained by
the considered tooth morphological variables. Meinl et al. (2007) reported
similar poor age predicting performances when validating the Kvaal et al.
(1995) method on a sample of individuals between 13 and 24 year. Further
on the contrasting performance between the young and adult age groups,
applying the Kvaal et al. (1995) method, was reported in the Paewinsky et al.
(2005) study. The results were plotted as relation between pulp-root width
ratios and age and fitted as well as linear, cubic as logistic functions. For
each of the fitted curves the young individuals (between 14 and 20 year)
could be considered as outliers. Hereby again a marked deviation from the
performance of the adult part of the considered sample (between 20 and 81
year) was expressed.
Erbudak et al. (2012) reported as the main disadvantage for the
application of the Kvaal et al. (1995) method on panoramic radiographs that
these images do not display the fine anatomic details available on periapical
117
Third molar and tooth morphological predictors
radiographs. Landa et al. (2009) had to exclude measurements of all
indicated upper teeth due to overlap and the lack of sharpness in their
selected panoramic radiographs. It has to be denoted that the potency to
perform the secondary dentin measurements on all tooth positions indicated
in the Kvaal et al. (1995) method, indeed greatly depended on the image
quality of the selected panoramic radiographs. But, the described differences
in performance between the Kvaal et al. (1995) and the current study were
not related to the better image quality found on periapical x-rays. Indeed, in
the Bosmans et al. (2005) study panoramic radiographs were evaluated on an
adult age group. It was concluded that no significant differences were
detected comparing the results based on periapical versus panoramic
radiographic data. Therefore, in the Bosmans et al. (2005) study the sampled
reference data were selected on criteria requiring good quality panoramic
radiographs with clear radiological image. It was not quantified how many
panoramic radiographs had to be eliminated from sampling. In the current
study, strict image quality selection criteria were used based on criteria
allowing to perform optimal variable measurements on each indicated tooth.
Therefore, on average, 80% of the archived radiographs had to be excluded.
The use of the image ameliorating tools in Adobe Photoshop CS4®, didn’t
allow to narrow the obtained exclusion result. Moreover, in the current study
it was aimed to measure the teeth on the left side. Due to the altering image
quality according to specific tooth positions, in almost every selected
panoramic radiograph at least one contra lateral tooth was chosen to enable
optimal variable measurements for that particular tooth position. In forensic
practice, it is impermissible to apply a method only applicable on 20% of the
population.
The poor age predicting performance of the models based on
morphological too(ee)th information indicate that in sub-adult individuals
who are missing all four third molars, the use of Kvaal et al. (1995)
measurements on all related permanent teeth is not a good alternative to
perform (dental) age estimations. In these cases, specific dental age
estimations can just be performed if other permanent teeth are still maturing.
If all teeth are fully developed the only dental age prediction that can be
reported is that the investigated individual is at least 16 years of age
(Liversidge et al., 2010).
CONCLUSION
Due to the inherent image quality of panoramic radiographs, Kvaal‘s
measurements could be obtained only on a restricted sample. Clinically, the
gain in age prediction accuracy was negligible in view of the timeconsuming additional tooth morphological measurements for the staged third
molar development. Forensic dental age estimations in the sub-adult group
118
Third molar and tooth morphological predictors
should consider third molar development as the only reliable age predictor
on panoramic radiographs. Thus, Research Hypothesis 7 cannot be accepted.
119
Chapter 9
Influence of skeletal age
predictors on age estimation
based on third molar
development
THIS CHAPTER IS BASED ON THE FOLLOWING MANUSCRIPT.
Human age estimation combining third molar and skeletal development.
Thevissen PW, Kaur J, Willems G
Published in International Journal of Legal Medicine
2012 Mar;126(2):285-92
Oral presentation at the annual scientific meeting of the American Academy of
Forensic Sciences, Chicago 2011
TESTING RESEARCH HYPOTHESIS 8:
In sub-adults, the accuracy of age estimations based on third molar
development is improved by adding age-related information from cervical
vertebrae maturation
121
Third molar and skeletal predictors
INTRODUCTION
To evaluate the chronological age of sub-adult individuals, dental age
estimation methods regard primarily the radiologically observed maturation
stages of developing third molars (Mincer et al., 1993). Compared to all the
other developing teeth, the timing of third molar development shows the
highest variability (Liversidge, 2008b). Consequently, age estimations based
on third molar development have wide prediction intervals (Thevissen et al.,
2010a). Therefore, worldwide, various forensic protocols for determining the
age of unaccompanied young refugees combine dental and other age
estimation methods (Garamendi et al., 2005; Solheim and Vonen, 2006;
Nuzzolese and Di Vella, 2008; Schmeling et al., 2008; Santoro et al., 2009;
Cunha et al., 2009; Lewis and Senn, 2010) [Fig.3, Chap. 1]. The objective of
pooling age estimation methods is to report the obtained age results and to
interpret them as a conclusive age outcome with narrow prediction intervals.
For sub-adult age estimation purposes, small-scale research was performed
on samples including living individuals with known chronological age about
whom dental and other age predictors were compiled at the same time
(Garamendi et al., 2005). As a result, the interpretations of pooled age
estimation outcomes are not solidly based and result, to a certain extent, in
investigator dependant conclusions.
Forensic age estimation protocols can combine methods based on third
molar development and socio-psychological maturity (Nelki and Bailey,
2010), physical appearance (Healy, 1992; Wright et al., 2002), secondary
sexual development (Tanner, 1986), clinical dental observations (Solheim
and Vonen, 2006), radiologically observed secondary dentine apposition
(Kvaal et al., 1995; Star et al., 2011), visibility on panoramic radiographs of
the root pulp and the periodontal ligament in third molars (Olze et al.,2010a),
and skeletal variables. The last group includes mainly non-invasive methods
based on the degree of ossification of hand-wrist bones (Greulich and Pyle,
1959; Tanner, 1975; Fishman, 1982; Leite et al., 1987; Tanner et al., 2001;
Gilsanz and Ratib, 2005), the medial part of the collar bone (Schmeling et
al., 2004; Schulz et al., 2005; Schulze et al., 2006; Schulz et al., 2008a,b;
Quirmbach et al., 2009; Kellinghaus et al., 2010; Hillewig et al., 2011), and
the costal cartilage of the first rib (Moskovitch et al.,2010; Garamendi et al.,
2011). For orthodontic treatment planning, correlations between skeletal and
dental development have frequently been investigated in an attempt to detect
the maturity status of the examined patient (Demisch and Wartmann, 1956;
Sierra, 1987; Lewis, 1991; Başaran et al.,2007; Cho and Hwang, 2009; Chen
et al., 2010a; Perinetti et al., 2011; Różyło-Kalinowska et al., 2011).
123
Third molar and skeletal predictors
Figure19:
Figure
19:Different
Differentskeletal
skeletaldevelopment
developmentregistration
registrationtechniques
techniquesused
applied
withon
cephalometric radiographs
The upper panels
panels illustrate
illustrate six
6 developmental
developmental stages
stages of cervical vertebrae:
vertebrae. on
At the
left side as described by Baccetti et al. (2005) (BA) and evaluated on the cervical
vertebrae C2,C3,C4.
C2, C3, C4;
At on
thethe
right
right
sideasasdescribed
describedby
bySeedat
Seedat and
and Forsberg
Forsberg (2005)
(SE) and evaluated on the cervical vertebra C3. The lower panels illustrate two
measuring techniques:
techniques. on
At the
the left,
left the
sidetechnique
the technique
described
described
by Caldas
by Caldas
et al. (2007,
et al.
(2007, (CAL)
2010)
2010) that
(CAL)
considers
considers
ratios
ratio’s
of the
of the
shown
shown
height
height
andand
width
width
measures
measures
of the
of
the cervical
cervical
vertebrae
vertebrae
C3 C3
andand
C4;C4.
on At
thethe
right,
rightthe
side
technique
the technique
described
described
by RaibyetRai
al.
et al. (2008)
(2008)
(RAI) (RAI)
considers
considers
3 length
3 length
measurements:
measurements:
# 1 between
#1 between
Gonion
Gonion
(Go)(Go)
and
and Condylion
Condylion
(Co),(Co),
# 2 #2
between
between
Condylion
Condylion
(Co)
(Co)and
andGnation
Gnation(Gn),
(Gn),##3
3 between
Gnation (Gn) and Gonion (Go).
(Go)
On cephalometric radiographs, the developmental changes of
cervical vertebrae were described to evaluate the degree of physiological
maturity of a growing individual and to calculate cervical vertebral bone age.
The development of the mandibular bone was registered and used as an age
predictor. The development of the radiologically observed cervical vertebrae
bodies was evaluated and registered using a staging and a measuring system.
More specifically, Baccetti et al. (2005) (BA) classified the vertebral body
growth of cervical vertebrae C2, C3 and C4 in six stages. Seedat et al. (2005)
(SE) considered a 6-stage system on C3 discriminating the different stages as
described by Hassel et al. (1995). Caldas et al. (2007, 2010) (CAL)
registered ratios of length and height of the corpus of C3 and C4 to
determine cervical vertebrae bone age using multiple regression analysis.
Rai et al. (2008) (RAI) measured two mandibular lengths and a height on
cephalometric radiographs and presented three related formulae to calculated
age [Fig. 19].
124
Third molar and skeletal predictors
The aim of the present study is, first, to compare existing skeletal
maturation evaluation methods developed on cephalometric radiographs
(BA, SE, CAL, RAI) in order to determine the most accurate age predicting
variable and its related registration system; second, to verify whether adding
the detected most accurate age predicting skeletal variable to third molar
development variables (obtained on panoramic radiographs) resulted in
improved age estimations. The research hypothesis to evaluate was that, in
sub-adults, the accuracy of age estimations based on third molar
development is improved by adding age-related information from cervical
vertebrae maturation.
MATERIALS AND METHODS
In an initial study, 496 cephalograms (283 M; 213 F) [Table 14] were
collected to determine the most accurate age predicting variables and the
related registration system among the published BA, SE, CAL, and RAI
methods. All the radiographs were taken from individuals of Central Indian
origin. Their chronological age on the day of x-ray exposure was calculated
based on a valid birth certificate. None of these individuals presented
congenital or acquired malformations affecting their dental or skeletal
development. The cephalograms were taken in an analogous way using a
Soredex unit (Soredex, Tuusula, Finland) and scanned digitally (HewlettPackard 8300,NY, USA). Care was taken that the individuals were
positioned correctly during cephalographic radiography and that all the
radiographs were of good image quality. The cephalograms were imported
into Adobe Photoshop CS3 (Adobe Systems Incorporated, San Jose, CA)
and scored or measured following the techniques described by BA, SE,
CAL, and RAI. Scatter plots with a smoothed trend line were used to explore
the shape of the relationship between age and the stages or measurements
obtained, and found to be nonlinear. Regression models were derived with
age as response and the stages or measurements as explanatory variables.
From each model determination, coefficients (R²) and root mean square
errors (RMSE) were analysed. R² indicates the proportion of the explained
variability in the response variable, namely, age. Alternatively, RMSE
denotes the magnitude of the error in age prediction. Interaction effects with
gender and main gender effects were checked for all of the variables.
The main study included 460 Caucasian individuals (234 F, 226 M)
from whom an orthopantogram and a cephalogram were taken on the same
day. The radiographs were retrospectively selected out of the dental clinics'
files of the Katholieke Universiteit Leuven that had been taken at intake for a
dental check-up. None of the collected individuals presented congenital or
acquired malformations affecting dental or skeletal development. The
chronological age on the day of the x-ray exposures was calculated based on
identity card records and ranged between 3 and 25 years old [Table 15]. The
125
Third molar and skeletal predictors
Table 14: Age and gender distribution of individuals included in the studied samples
Age
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
33
Total
Initial sample
M+F F
M
.
.
.
2
1
1
3
1
2
3
1
2
7
6
1
12
6
6
25
12
13
18
6
12
37
17
20
53
27
26
51
24
27
46
24
22
33
22
11
38
28
10
31
19
12
25
14
11
16
13
3
14
9
5
12
9
3
20
10
10
14
9
5
14
7
7
6
6
.
6
3
2
5
5
.
5
3
2
1
1
.
496
283 213
Main sample
M+F F
M
2
2
.
7
2
5
7
5
2
14
7
7
18
5
13
16
5
11
3
2
1
19
7
12
19
9
10
40
26
14
65
30
35
58
30
28
57
29
28
42
26
16
25
11
14
10
3
7
11
5
6
10
5
5
3
.
3
4
2
2
11
8
3
8
7
1
11
8
3
.
.
.
.
.
.
.
.
.
.
.
.
460
234 226
Additional sample
M+F F
M
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3
2
1
19
7
12
19
9
10
40
26
14
65
30
35
58
30
28
57
29
28
42
26
16
25
11
14
10
3
7
11
5
6
10
5
5
3
.
3
4
2
2
11
8
3
8
7
1
11
8
3
.
.
.
.
.
.
.
.
.
.
.
.
396
208 188
F: female, M: male, Age-3 includes subjects between 3.00 and 3.99 years old, etc.
radiographs were taken digitally: the orthopanograms on a Cranex unit
(Soredex, Tuusula, Finland) and the cephalograms on a Orthoceph OC100
unit (Instrumentarium Corp, Graven, Finland), both using phosphor plate
technology (Siemens, Berlin, Germany).
On the orthopantomograms, third molar development was evaluated
using the ten point staging technique of Köhler et al., (1994) (KO). Pearson’s
correlation coefficients between the developmental scores of different
wisdom tooth positions were calculated and induced multicollinearity in the
regression model. Because of the highly correlated left and right third molar
scores, this problem was reduced using only stages of the left third molars.
The development of third molars cannot be measured before the onset of the
calcification of third molars or when third molars are absent. This missing
126
Third molar and skeletal predictors
information may in itself contain some information about age. Therefore,
four prediction models including this information were constructed based on
the present third molars: first: upper and lower molar present, second: upper
molar present, third: lower molar present; fourth: no third molars present. To
apply the maximal available information, for each subject the predictions
were used agreeing with the absence pattern of this subject.
The cephalometric radiographs were scored following the most (BA)
and second most (SE) accurate age predicting skeletal variables and related
registration systems identified in the initial study.
Interaction effects with gender and main gender effects were
checked, first, for regression models with age as response and the third
molar scores (KO) as explanatory variable. Second, the same models were
fitted including additional information of respectively BA, SE and BA+SE.
From each model, determination coefficients (R²) and root mean square
errors (RMSE) were calculated and analysed.
Because no third molar development was observed for any subject
younger than 9 years old, in an additional study a sample that included all of
the individuals of the main study older than 9 years of age were selected
[Table 14]. All the analyses from the main study were repeated on this
reduced sample.
All analyses were performed using PROC GLM in SAS Version 9.2
(SAS Institute Inc., Cary, NC, USA)
RESULTS
In the initial study, the age predicting variables and related
registration system providing the most information on age was BA (58%),
immediately followed by SE (55%). Combining these two techniques
provided a gain of 5% of explained variability in age (63%). The two CAL
ratios jointly explained 26% of the variability. All RAI measures clearly
contained very little information on age, and their regression models
explained at most3% of the variability in age. The calculated RMSE values
ranged between 3.20 and 4.90 years placing the magnitude of the error in the
age prediction of the age predicting variables and related registration
systems in the same order of best performance as based on the detected R²
values (BA>SE>CAL>RAI) [Table 15].
In the main study, a Pearson correlation coefficient of 0.98 was
determined between the left and the right third molars from both the upper
and the lower jaw. Between the upper and lower third molars, this
coefficient was 0.91 and 0.90 for the left and the right side, respectively.
Inclusion of information from cephalograms based on the BA as well as the
SE technique improved the amount of explained variance in age acquired
from the panoramic radiographs using the KO technique by 48%. Inclusion
of cephalometric BA+SE information marginally improved the previous
127
Third molar and skeletal predictors
Table 15: Determination coefficient (R²) and root mean squared error (RMSE) from
the regression models with age as response and the indicated explanatory variable(s)
developed on the initial sample.
R²
Explanatory variable
BA
SE
BA+SE
CAL
RAI(Go-Co)
RAI(Gn-Co)
RAI(Go-Gn)
M+F F
M
RMSE
M+F F
M
0.58
0.55
0.63
0.26
0.03
0.02
0.01
0.61
0.59
0.66
0.30
0.12
0.05
0.04
3.20
3.29
3.01
4.23
4.85
4.88
4.90
0.55
0.50
0.60
0.20
0.01
0.01
0.01
3.33
3.43
3.13
4.36
4.99
4.98
4.99
3.00
3.10
2.84
4.11
4.43
4.60
4.61
M: male, F: female, BA: Baccetti et al. (2005), SE: Seedat and Forsberg (2005), CAL:
Caldas et al. (2007, 2010), RAI: Rai et al. (2008), Go-Co: length from Gonion to
Condylion, Gn-Co: length from Gnation to Condylion, Go-Gn: length from Gonion to
Gnation
result (+1%). The RMSE decreased by 1.93, 1.85 and 2.03 years of age
adding, respectively, BA, SE and BA+SE information to the KO model
[Table 16].
In the additional study, the magnitude of explained variance in age
adding BA, SE and BA+SE information to the KO model was reduced to 19,
17 and 21%, respectively. The RMSE ranged for the models, including
cephalometric information between 1.62 and 1.78 years [Table 17].
In all the study samples and for all variable(s) and related
registration systems, age was better predicted for M than for F.
DISCUSSION
It should be recommended not to use the RAI technique for age estimations.
The initial study revealed that measures from the RAI technique provide an
extremely low maximal explained variability in age (3%), and high RMSE
values (approximately 5 year). Moreover, Dibbets et al. (2002) and Cohen
(2005) reported that the magnification inherent to the technique of
radiographic projection should be taken into account when comparing linear
dimensions on cephalometric data. The RAI technique is not allowing to
correct for magnifications of data from different sources. In contrast, the
CAL technique is overcoming this problem using ratios of linear dimensions
obtained on the same radiograph.
The best to age-related age predicting variable(s) and related registration
system were the staging and corresponding scoring techniques of BA and
SE. The measuring technique of CAL was remarkably less performing.
Thevissen et al. (2011) ascertained that scorings of third molar stages
128
Third molar and skeletal predictors
Table 16: Determination coefficient (R²) and root mean squared error (RMSE)
obtained from the regression models with age as response and the indicated
explanatory variable(s) developed on the main sample.
R²
KO
KO+BA
KO+SE
KO+BA+SE
M+F F
M
RMSE
M+F F
M
0.39
0.87
0.87
0.88
0.53
0.90
0.91
0.92
3.60
1.67
1.75
1.57
0.30
0.86
0.84
0.87
3.99
1.81
1.97
1.75
2.99
1.39
1.33
1.22
M: male, F: female, KO: Köhler et al. (1994), BA: Baccetti et al. (2005), SE: Seedat and
Forsberg (2005)
(categorical data) were best related to age and provided the most accurate
age predictions compared to tooth measurements and ratio’s of tooth
measurements from third and second molars (continuous data). They stated
that measures and related ratios used to register molar development,
incorporate the variance in tooth size between individuals. A similar
explanation for the minor performance of the measured observations of
skeletal development of cervical vertebrae is that these measurements (and
their ratios) incorporated the human variability in corpus vertebrae size. The
used staging techniques ignored corpus vertebrae size differences between
individuals, delivering higher percentages of explained variability in age and
smaller magnitudes of error in the age estimates. Caldas et al. (2010)
reported that a computerized CAL technique was used in their study because
it allowed skeletal age to be measured and calculated in an objective manner.
Their decision was based upon the findings of Özer et al. (2006) indicating
that in the Lamparski (1972) technique, which was modified into the SE
technique (Seedat and Forsberg, 2005), the initial and final developmental
stage was most accurate, because compared to the intermediate stages the
borderline cases blended into each other. The current initial study denoted
that, for age estimation purposes, the staging techniques were outperforming
the measuring technique. In addition, no agreement exists on the
reproducibility of cervical vertebrae staging techniques. On one hand,
Gabriel et al. (2009) detected inter observer agreement levels below 50%
and slightly better intra observer agreement, on the other hand Jaqueira et al.
(2010) reported good inter observer agreement for the staging techniques of
BA, SE and Hassel et al. (1995). Further on, Baccetti et al. (2005), San
Romàn et al. (2002) and Chen et al. (2010b) observed that the concavity of
the lower border of the vertebral bodies is more outspoken with increased
maturity. This finding should be considered in an attempt to diminish the
borderline cases. Moreover, during forensic age estimations on living
individuals the advantage of the doubt has to be given to the examined
individual. In most cases, this advantage has to be accorded to the youngest
129
Third molar and skeletal predictors
Table 17: Determination coefficient (R²) and root mean squared error (RMSE)
obtained from the regression models with age as response and the indicated
explanatory variable(s) developed on the additional sample
R²
KO
KO+BA
KO+SE
KO+BA+SE
M+F F
M
RMSE
M+F F
M
0.59
0.78
0.76
0.80
0.62
0.81
0.79
0.85
2.29
1.69
1.78
1.62
0.60
0.79
0.74
0.80
2.47
1.80
1.99
1.76
1.97
1.41
1.33
1.25
M: male, F: female, KO: Köhler et al. (1994), BA: Baccetti et al. (2005), SE: Seedat and
Forsberg (2005)
age outcomes. Therefore, borderline cases detected during the staging of
cervical vertebrae development have to be classified in the earliest of the
questioned stages.
Clinically, the SE technique allows a faster and easier registration of the
degree of development of the cervical vertebrae compared with the BA
technique. Both techniques classify the observed cephalometric radiographs
into six stages, but against the more complex combined examination of three
vertebrae (C2, C3, C4) considered in the BA technique, the SE technique
simplifies the evaluations to the examination of one vertebral corpus (C3).
Since statistically the performances of SE or BA, added to KO are likely in
as well the main as the additional sample, based on clinical conveniences,
SE is the technique of choice to classify the added cervical vertebrae
development. Further on, optimal accuracies in age predictions are obtained
using gender-specific regression models [Table 18]. In particular, for M all
obtained RMSE values are reduced compared to the gender independent
values. In a forensic context, the use of gender-specific models is no
constrain because the sex of examined unaccompanied young asylum
seekers is always known.
The main study indicated that registrations of cervical vertebrae
development added to stages of third molar development, improved
drastically the age predictions. This improvement was largely ascribed to the
fact that the main sample contained subjects with ages between 3 and 25
year, while no dental stages and related scores were available on any subject
younger than 9 years. For these young subjects, the cervical vertebrae
development was the only age-related information available apart from the
fact that absence of third molar development, especially in this young age
category is age-related. As expected, in the additional sample adding skeletal
information (BA) to dental information (KO) for age prediction reduced the
explained variability in age from 87% in the main sample to 78%. The
explained variability for the model with only dental information (KO)
increased from 39% in the main sample to 59% for the additional sample.
130
Third molar and skeletal predictors
Table 18: Gender specific regression formulae for KO plus SE added, and SE fitted
on the additional sample
Present third molars Gender Regression formula
Age=7,38+1,27UL-0,25UL²+0.02UL3+0.03LL+0,84SE
ul+ll
F
Age=8,19+0,32UL-0,19UL²+0,02UL3+0,37LL+0,01SE
M
Age=8,35+1,17UL-0,23UL²+0,02UL3+0,84SE
ul
F
Age=8,51+0,48UL-0,14UL²+0,02UL3+1,07SE
M
Age=7,63+1,40LL-0,31LL²+0,28LL3+0,92SE
ll
F
Age=10,57+0,20LL-0,08LL²+0,01LL3+1,10SE
M
Age=9,27+1,96SE
F
Age=4,92+2,21SE
M
KO: Tooth score according to Köhler et al. (1994), SE: Cervical vertebrae score
according to Seedat and Forsberg (2005), ul: Upper left, ll: Lower left, F: Female, M:
Male, UL: Score upper left third molar, LL: Score lower left third molar
Subtracting both previous results reduced the gain of explained variability in
age from 48% in the main sample to 19% in the additional sample.
Combining KO and BA techniques diminished in the additional sample the
RMSE with 0.6 years compared to the KO technique alone. These findings
indicate a considerable gain in accuracy of age prediction combining third
molar and cervical vertebrae information. However, the period of vertebral
development is not completely overlapping the span of third molar(s)
development. The older the considered individual gets, the minor the extent
of overlap is. This was reflected in a related decrease of the R² values and a
gain in calculated RMSE. Indeed, the values obtained from the models (KO
and KO+BA) based on all individuals from the main sample older than 14
years (n=250), and those older than 16 years (n=135) revealed a decrease in
added R² for both groups to 3%, and the gain in RMSE was respectively 0.12
and 0.09 years. Further on, the latest BA stage with potential of cervical
vertebrae development (BA, stage 5) ranged between 11.51 and 19.47 year
while the last stage with potential third molar development (KO, stage 9)
ranged between 17.27 and 25.7 years. Consequently, during the period of
late third molar development no or a neglect able gain in accuracy of age
prediction is obtained after adding cervical vertebrae information to third
molar(s) information. In summary it should be recommended to take
additional cephalometric radiographs when aging individuals with third
molar development lower than KO stage 7 (root ¾ developed). As such,
research hypothesis 8 could not be accepted and no age-related information
from cervical vertebrae maturation was implemented in the triple test.
The effective radiation dose needed for a cephalometric exposure varies
between 2 and 3 micro sievert (µSv). In forensic age estimation
investigations, additional skeletal age information is most frequently
obtained from data observed on hand-wrist and chest radiographs with
131
Third molar and skeletal predictors
respective effective radiation doses around 5 and 30 µSv (EC, 2004). The
relative low cephalometric dose is usually less than one day of natural
background radiation. This has to be considered as an advantage of the
proposed age estimation technique, especially because the examined
individuals are maturing children.
Gabriel et al. (2010) analyzed facial proportions of developing juveniles for
age estimation purposes. The authors will assemble frontal and lateral
anthropometric data in age-related reference samples. Since cephalometric
radiographs visualize soft-tissue contours, certain of the above mentioned
lateral measurements can be obtained from these radiographs. These way
cephalometric radiographs could be considered as a source of soft tissue and
skeletal age-related information. In future research, the accuracy of age
predictions adding these non skeletal lateral measurements to the dental and
skeletal information described throughout the current study could be
searched.
The major problem in establishing the present research was to
collect retrospectively individuals on which on the same day a panoramic
and a cephalometric radiograph was taken. In future research, the present
main sample will be extended to obtain a sample, including for both genders,
individuals homogenously distributed in age categories of maximally one
year, allowing for optimal statistical analysis (Smith, 1991; Bocquet-Appel
and Masset, 1996; Gelbrich et al., 2010; Liversidge et al., 2010). Further on
under the same inclusion conditions a validation sample will be collected to
verify the established regression models. Age estimation methods based on
tooth development for the age categories until 16 years consider all
developing teeth except third molars and for age categories of young
individuals above 16 years the developing third molars [Fig. 1, Chap. 1].
Since cervical vertebrae development is not equally overlapping both age
categories, in future research the skeletal information will be added to dental
information obtained using two techniques. For the age group below 16
years the Willems et al., (2010) technique will be applied on all lower left
permanent teeth, and for the group above 16 year the third molar(s)
development will be staged as described in the current study.
CONCLUSION
On cephalometric radiographs, the skeletal age predicting variables and
related registration systems providing the most information on age were the
BA and SE cervical vertebrae scoring systems. Because the SE technique
provides clinically the easiest and fastest registration of the degree of
development of the cervical vertebrae, it is the technique of choice to
classify the added cervical vertebrae development. Adding the BA or the SE
information to the third molar model developed on the basis of KO stages
132
Third molar and skeletal predictors
improved the age predictions greatly in the period of early third molar
development. Sub-adult dental age estimations are based on the late third
molar development and consequently the research hypothesis cannot be
accepted.
133
Chapter 10
General discussion and
conclusion
135
General discussion and conclusion
The principal and the most requested forensic application of dental age
estimation is based on third molar development. It provides age assessments
for the sub-adult group and allows the status as child or adult of young
unaccompanied refugees to be determined. To report legally indisputable age
estimation outcomes, the age estimation examinations need to be based on
scientifically sound evidence. Therefore, the general aim of the current
research was to provide a scientific basis for the optimisation of dental age
estimates based on third molar development. Hence, multiple research
hypotheses were evaluated and, on the basis of the results, the Triple Test
established at the KULeuven was modified.
The initial step in third molar development data collection concerns
the registration of the third molar maturation available in a subject.
Panoramic radiographs allow one to observe at a specific moment on one
image and with a minimal radiation dose, the developmental status of all
third molars. Testing Research Hypothesis 1, was determined if measuring
the observed third molar dimensions provided better age prediction
performances than does staging its morphological status in relation to an
arbitrarily established ordinal developmental sequence. The measurements
were less informative about age and added no age-related information once
the stages of third molar development were used for age estimation.
Different tooth development staging techniques were described that were
based mainly on the number of steps the tooth maturation process was
arbitrarily divided into.
Therefore, with Research Hypothesis 2, the influence of the number
of stages described in the third molar development registration techniques on
their age prediction performances was evaluated. The age predictions were
negligibly influenced by the number of stages described in the registration
technique. As such, the choice of staging technique should depend on the
availability of stages in the period of interest and should allow a precise
distinction between them. Therefore, the 10-point staging technique
according Gleiser et al. (1955) and modified by Köhler et al. (1994) was
found most suitable for age predictions based on the late third molar
development. Accordingly, it was used in the further research and in the
Triple Test.
In a classic approach, third molar development data collected from a
reference sample is modelled using regression analysis. As such, the age of
an individual can be predicted by examining the registered third molar
development of one, or more, of the third molars present in the
corresponding regression model. Therefore, the residuals in the regression
model are assumed to be normally distributed around the regression line with
a constant variance. In practice, the conditions for this assumption are often
absent and, accordingly, the correctness of the quantified uncertainty of the
age prediction is affected. Furthermore, the high correlation between the
developmental third molar variables restricts the number of third molars
137
General discussion and conclusion
integrated in the model due to multicollinearity. To include the frequent
agenesis of third molars, different models covering the patterns of missing
third molars need to be constructed. As such, the age predictions obtained
with regression models based on third molar development depend on the
choice of the regression formula. Aykroyd et al. (1997, 1999) have described
a systematic bias in age estimation using regression analyses. Indeed,
depending upon the degree of agreement between the stages and age, the
estimated ages are too young for old individuals and too old for young
individuals. The legal context of forensic investigations requires indisputable
results. This is not obtained fully with regression analysis for age estimation
because of the distribution of the residuals, multicollinearity and the
improper integration of missing third molars. Moreover, in case of doubt, the
benefit of the doubt has to be given to the applicant. The systematic bias
detected in regression analyses used for age estimations leads to
overestimating the age of young individuals. This effect is the opposite of
the benefit of the doubt to the young. Therefore, in an attempt to reduce the
limitations in age estimation using regression analyses, in the present
research a Bayesian approach for age estimation based on third molar
development was constructed. With this approach, no assumptions about the
specific shape and variability in the age distribution conditional on the
observed stages are assumed. The developed model allows, at the cost of
higher computational complexity, one to incorporate all the age-related
information of the four third molar positions. Indeed, specific to the tooth
position, the developmental stage of the present third molars and evidence
about the missing third molars were integrated. The probability distributions
obtained from the model enabled to calculate point predictions and
confidence intervals that provide an age estimate with a corresponding
prediction interval. This model normalizes the total surface of the posterior
distribution equal to 1, thus allowing one to discriminate the proportion of
individuals younger or older than a set age threshold. The differences
between the observed and predicted age, the precision in age estimation and
the coverage of the prediction intervals were equal comparing the regression
models with the Bayesian approach. Consequently, Research Hypothesis 3
had to be rejected.
However, the Bayesian approach reduced the systematic bias present
in the regression model. The age of juveniles was less overestimated, which
means that subjects younger than 18 years old were more often classified
correctly. As such, especially in a legal context, the Bayesian approach
permits fewer disputable age estimation examinations than does regression
analysis. In essence, the Bayesian approach administers a computerized table
that provides age estimates for all possible combinations of developmental
stages or tooth absence observed at the four third molar positions. In the
literature, no age estimation tables combining all this third molar information
have been published [Table 1, Chap.1]. For these reasons, in the present
138
General discussion and conclusion
research, reference samples were modelled using the established Bayesian
approach and applied for age estimation. In the Triple Test, if all the
permanent teeth (except the third molars) are mature, tooth-position specific,
the KO developmental stages and information of missing third molars are
integrated in a Bayesian model constructed on a reference sample from
Belgium with 1106 subjects [Table 12a, Chap. 7]. Further on, the probability
of an applicant being older than 18 years of age is calculated with the same
model based on the applicants’ third molar development status.
Inherent to unaccompanied young refugees migrating throughout the
world, forensic age estimations need to be performed on applicants from
diverse geographic and biologic origin. Because dental age estimations in
this age category depend on third molar development, it was set out to
determine in a standardized way if there are country-specific differences in
third molar development. Based on a quantification of the degree of third
molar development in a pooled sample of 13 country-specific third molar
datasets, it was determined that, indeed, there are differences in third molar
development between countries. The differences in third molar development
between countries were heterogenic, without clear patterns, and changed
over age. Although some of the pairwise differences in third molar
development between countries were statistically significant (p = 0.05), they
were clinically small and negligible. In fact, the maximal difference in third
molar development detected between 2 countries could be quantified as 14
months or as a succeeding pattern of four equal KO third molar scores.
Research Hypothesis 4 was accepted and, for forensic applications, it had to
be investigated what the consequences of the differences in third molar
development between countries were on the corresponding age predictions.
Therefore, the 13 collected country-specific samples were divided into a
reference and a validation sample to establish country-specific Bayesian age
estimation models and to verify and compare their performance on the
respective country-specific, the Belgian, and the global validation datasets. It
appeared that the model developed on the Belgian reference sample
provided, in the juvenile-adult discriminations, a greater number of selected
juveniles compared to the models constructed on the country-specific or the
global reference samples. Therefore, Research Hypothesis 5 was accepted.
This implies that, in the absence of a country-specific reference model, the
Belgian reference model is legally the most suitable for forensic dental age
discrimination of unaccompanied young refugees. Information from
Belgium increased the MAD maximally 2.6 month compared to information
provided from the country itself; using information from the global dataset
reduced the maximally added error to 1.0 month. In the sub-adult group, age
estimations made with the model constructed on the global reference dataset
provide age estimates approaching the most the estimates based on the
model constructed on the country-specific reference sample. As a result,
Research Hypothesis 6 was accepted. In practice, with the Triple Test, an
139
General discussion and conclusion
applicants’ age is estimated using the Bayesian model established on the
country-specific reference sample corresponding to the applicants’
nationality. If the nationality of the examined individual is not included
within the series of established country-specific reference models, the
applicant’s age is estimated based on the global reference dataset.
Accordingly, the possible added maximal error in age prediction of 1.0
month is reported. Regarding juvenile-adult discrimination, the applicant is
evaluated on the country-specific model established with a reference sample
in function of her or his nationality. If no corresponding model has been
established, the discrimination is performed using the model established on
the Belgium reference sample. As such, a maximal benefit of the doubt is
provided for juvenile adult discrimination.
Forensic age estimation methods provide wide prediction intervals.
Therefore, specifically for forensic age estimations of unaccompanied young
refugees, protocols combining age estimation methods based on various agerelated variables were established. Because age estimations based on third
molar development require panoramic radiographic inspection, the current
research evaluated if on these radiographs age-related dental morphological
variables could be detected and used as an additional age predictor. For that
reason on panoramic radiographs, the apposition of secondary dentine in the
left permanent teeth was quantified using the Kvaal et al. (1995) method.
Multiple regression models combining the registered third molar
development and the morphologic permanent tooth information provided an
ignorable and a clinically insignificant quantity of added age information
compared to the corresponding third molar models. A poor amount of dental
age-related morphologic information was available in the sub-adult age
range. Research Hypothesis 7 was not accepted, and accordingly, no dental
morphological age predictors were added to the Triple Test. In dental
practice, in addition to panoramic radiographs, cephalometric radiographs
are frequently used as a diagnostic tool, especially for orthodontic treatment.
In the current research, the best to age-related skeletal variable detectable on
cephalometric radiographs was searched. The development of the cervical
vertebrae C2, C3 and C4 contained the most age-related information and was
best registered with the technique described by Baccetti et al. (2005) and
Seedat and Forsberg (2005). Further it was verified whether adding agerelated information of cervical vertebrae to third molar development
information resulted in improved age estimations for the sub-adult group.
The period of vertebral development does not completely overlap the span of
third molar development and with increasing sub-adult age the extent of
overlap diminishes. Consequently, during the period of late third molar
development no or negligible gain in accuracy of age prediction was
obtained, and Research Hypothesis 8 could not be accepted. As a practical
consequence, no age-related variables based on cervical vertebrae maturation
were added in the Triple Test.
140
General discussion and conclusion
In an optimal approach, dental age estimations in sub-adults are based on the
radiologically observed developmental status or the absence of each third
molar. The information is registered with a staging technique. The collected
reference information is modelled as dependent ordinal data in a multivariate
Bayesian approach. It allows one to integrate in a legally indisputable
manner all the age-related information from the four third molars. The
Bayesian model provides, in addition to age estimates and corresponding
prediction intervals, a more correct classification between adults and
juveniles, especially for the latter. Despite the country-specific differences in
third molar development, the clinical impact on age estimation is minimal.
Based on on-going country-specific data collection, a quantification of the
maximal difference in prediction error using non- county-specific reference
information is established. Further on, the Belgian reference information
classifies more juveniles compared to country-specific information. As such,
in a forensic context, the benefit of the doubt for the examined individual is
integrated using the Belgium data as the reference information. Ideally, age
estimation protocols combining age estimation methods based on several
age-related variables are based on reference information collected from the
subject in whom the concerned variables are observed and registered at the
same time. In sub-adults, tooth morphological age predictors based on
secondary dentine formation observed in panoramic radiographs and skeletal
age predictors based on cervical vertebrae development observed in
cephalometric radiographs add no extra age-related information to third
molar development.
Diverse aspects of the current research are subject for future
research. An ongoing collection of country-specific third molar data-sets
enables a continuous validation of the obtained age predictions and juvenileadult discriminations between countries. In the time frame of the current
research, it was impossible to collect a country-specific sample from a
(black) colored population. Its integration in the already collected data would
create a reference covering the major ethnic groups.
An ethical issue related to age estimations in the living, concerns the
application of ionizing imaging techniques for non medical diagnostic
purposes (Thevissen et al., 2012). Magnetic resonance imaging (MRI) is a
non-ionizing medical imaging technique, and its applicability on dental
structures has to be investigated. In particular, it should be explored if the
related MRI image quality allows to register tooth development according to
the most suitable staging technique (KO). Ideally, an automated technique
for tooth recognition and staging its developing status on MRI images is
established. The three-dimensional character of MRI images provides an
extra challenge to this investigation and opens perspectives for more
accurate and higher reproducible registrations of tooth development.
The error in age predictions could be improved combining in one
model the dental and skeletal information used in the triple test. Therefore, a
141
General discussion and conclusion
data set containing age-related information from hand wrist, clavicle and
third molar development, registered at the same moment in sub-adult
subjects should be collected. An ethically approvable collection of the
necessary information could be obtained using MRI imaging techniques in
living subjects or full body computed tomography in deceased. On the
reference set an age prediction model could be constructed using a Bayesian
approach. To permit the integration of all skeletal and dental variables a
relaxed model assuming independence of the included variables should be
constructed and verified.
To establish uniform and indisputable age estimations in sub-adults
future research has to investigate automated data registration and its
Bayesian modelling.
142
References
Abbing, H. (2011). Age Determination of Unaccompanied Asylum Seeking Minors in the
European Union:A Health Law Perspective. Eur J Health Law , 18(1):11-25.
Acharya, A. (2011a). Accuracy of predicting 18 years of age from mandibular third molar
development in an Indian sample using Demirjian's ten-stage criteria. Int J Legal Med ,
125(2):227-33.
Acharya, A. (2011b). Age estimation in Indians using Demirjian's 8-teeth method. J
Forensic Sci , 56(1):124-7.
Alkass, K., Buchholz, B., Ohtani, S., Yamamoto, T., Druid, H., & Spalding, K. (2010a). Age
estimation in forensic sciences: application of combined aspartic acid racemization and
radiocarbon analysis. Mol Cell Proteomics , 9(5):1022-30.
Alkass, K., Buchholz, B., Druid, H., & Spalding, K. (2010b). Analysis of 14C and 13C in
teeth provides precise birth dating and clues to geographical origin. Forensic Sci Int ,
15;209(1-3):34-41.
AlQahtani, S., Hector, M., & Liversidge, H. (2010). Brief communication: The London atlas
of human tooth development and eruption. Am J Phys Anthropol , 142(3):481-90.
Anderson, D., Thompson, G., & Popovich, F. (1976). Age of attainment of mineralization
stages of the permanent dentition. J Forensic Sci , 21(1):191-200.
Ando, S., Aizawa, K., Nakashima, T., Shinbo, K., Sanka, Y., Kiyokawa, K., et al. (1965).
Studies on the consecutive survey of succedaneous and permanent dentition in the Japanese
children. 1. Eruption processes of permanent teeth. J Nihon Univ Sch Dent , 7(4):141-81.
Arany, S., Iino, M., & Yoshioka, N. (2004). Radiographic survey of third molar development
in relation to chronological age among Japanese juveniles. J Forensic Sc , 49(3):534-8.
Aykroyd, R., Lucy, D., Pollard, A., & Solheim, T. (1997). Technical note: regression
analysis in adult age estimation. Am J Phys Anthropol , 104:259–265.
Aykroyd, R., Lucy, D., Pollard, A., & Roberts, C. (1999). Nasty, brutish, but not necessarily
short: a reconsideration of the statistical methods used to calculate age at death from adult
human skeletal and dental indicators. American antiquity , 64:55–70.
Aynley–Green, A. (2011, march). The assessment of age in undocumented migrants. A
report for the office of the Denfensor del Pueblo, Madrid , Spain. Retrieved march 04, 2013,
from
humanrights.gov.au
http://humanrights.gov.au/ageassessment/submissions/
Sir%20Al%20 Aynsley-Green%20Kt%20(Submission%2038).pdf.
Baba-Kawano, S., Toyoshima, Y., Regalado, L., Sa'do, B., & Nakasima, A. (2002).
Relationship between congenitally missing lower third molars and late formation of tooth
germs. Angle Orthod , 72(2):112-7.
Baccetti, T., Franchi, L., & McNamara, J. (2005). The Cervical Vertebral Maturation
(CVM) Method for the Assessment of Optimal. Treatment Timing in Dentofacial
Orthopedics. Semin Orthod , 11:119–129 .
Bagherpour, A., Anbiaee, N., Partovi, P., Golestani, S., & Afzalinasab, S. (2012). Dental
age assessment of young Iranian adults using third molars: A multivariate regression study.
J Forensic Leg Med , 19(7):407-12.
143
References
Bai, Y., Mao, J., Zhu, S., & Wei, W. (2008). Third-molar development in relation to
chronologic age in young adults of central China. J Huazhong Univ Sci Technolog Med Sci
, 28(4):487-90.
Bang, G., & Ramm, E. (1970). Determination of age in humans from root dentin
transparency. Acta Odontol Scand , 28(1):3-35.
Başaran, G., Ozer, T., & Hamamci, N. (2007). Cervical vertebral and dental maturity in
Turkish subjects. Am J Orthod Dentofacial Orthop , 131(4):447.e13-20.
Bassed, R., Briggs, C., & Drummer, O. (2011). Age estimation and the developing third
molar tooth: an analysis of an Australian population using computed tomography. J
Forensic Sci , 56(5):1185-91.
Bengston, R. (1935). A study of the time of eruption and root development of the permanent
teeth between six and thirteen years. Nortwest University Bulletin , 35.3-9.
Benomran, F. (2009). The medico-legal scene in Dubai: 2002-2007. Forensic Leg Med ,
16(6):332-7. .
Benzer, G. (1948). The development and morphology of physiological secondary dentin. J
Dent Res , 27:640–646.
Bhat, V., & Kamath, G. (2007). Age estimation from root development of mandibular third
molars in comparison with skeletal age of wrist joint. Am J Forensic Med Pathol ,
28(3):238-41.
Blankenship, J., Mincer, H., Anderson, K., Woods, M., & Burton, E. (2007). Third molar
development in the estimation of chronologic age in american blacks as compared with
whites. J Forensic Sci , 52(2):428-33.
Blenkin, M., & Taylor, J. (2012 ). Age estimation charts for a modern Australian
population. Forensic Sci Int. , 10;221(1-3):106-12.
Bocquet-Appel, J., & Masset, C. (1996). Paleodemography: expectancy and false hope. Am
J Phys Anthropol , 99(4):571-83.
Bolaños, M., Moussa, H., Manrique, M., & Bolaños, M. (2003). Radiographic evaluation of
third molar development in Spanish children and young people. Forensic Sci Int ,
5;133(3):212-9.
Boonpitaksathit, T., Hunt, N., Roberts, G., Petrie, A., & Lucas, V. (2011). Dental age
assessment of adolescents and emerging adults in United Kingdom Caucasians using
censored data for stage H of third molar roots. Eur J Orthod , 33(5):503-8.
Bosmans, N., Ann, P., Aly, M., & Willems, G. (2005). The application of Kvaal's dental age
calculation technique on panoramic dental radiographs. Forensic Sci Int , 2005. 29;153(23):208-12.
Boyde, A. (1997). Microstructure of enamel. Ciba Found Symp , 205:18-2.
Braga, J., Heuze, Y., Chabadel, O., Sonan, N., & Gueramy, A. (2005). Non-adult dental age
assessment: correspondence analysis and linear regression versus Bayesian predictions. Int
J Legal Med , 119(5):260-74.
Bullitin, E. (2012). A report from the European Migration network for the period January to
May
2012.
Retrieved
march
3
2013,
from
http://emn.intrasoftintl.com/Newsletter/previewNews.do?id=17
144
References
Caldas, I., Júlio, P., Simões, R., Matos, E., Afonso, A., & Magalhães, T. (2011).
Chronological age estimation based on third molar development in a Portuguese
population. Int J Legal Med , 125(2):235-43.
Caldas, M. d., Ambrosano, G., & Haiter Neto, F. (2007). New formula to objectively
evaluate skeletal maturation using lateral cephalometric radiographs. Braz Oral Res ,
21(4):330-5.
Caldas, M. d., Ambrosano, G., & Haiter Neto, F. (2010). Computer-assisted analysis of
cervical vertebral bone age using cephalometric radiographs in Brazilian subjects. Braz
Oral Res , 24(1):120-6.
Caldas, I., Carneiro, J., Teixeira, A., Matos, E., Afonso, A., & Magalhães, T. (2012).
Chronological course of third molar eruption in a Portuguese population. Int J Legal Med ,
126(1):107-12.
Callahan, N., Modesto, A., Meira, R., Seymen, F., Patir, A., & Vieira, A. (2009). Axis
inhibition protein 2 (AXIN2) polymorphisms and tooth agenesis. Arch Oral Biol , 54(1):459.
Calonius, P., Lunin, M., & Stout, F. (1970). Histologic criteria for age estimation of the
developing human dentition. Oral Surg Oral Med Oral Pathol , 29(6):869-76.
Cameriere, R., Ferrante, L., & Cingolani, M. (2004). Variations in pulp/tooth area ratio as
an indicator of age: a preliminary study. J Forensic Sci , 2004 , 49(2):317-9.
Cameriere, R., Ferrante, L., & Cingolani, M. (2006). Age estimation in children by
measurement of open apices in teeth. Int J Legal Med. , 120: 49-52.
Cameriere, R., Ferrante, L., De Angelis, D., Scarpino, F., & Galli, F. (2008). The
comparison between measurement of open apices of third molars and Demirjian stages to
test chronological age of over 18 year olds in living subjects. Int J Legal Med , 122(6):4937.
Cantekin, K., Yilmaz, Y., Demirci, T., & Celikoglu, M. (2012). Morphologic analysis of
third-molar mineralization for eastern Turkish children and youth. J Forensic Sci ,
57(2):531-4.
Cardoso, H. (2007). A test of the differential accuracy of the maxillary versus the
mandibular dentition in age estimations of immature skeletal remains based on developing
tooth length. J. Forensic Sci. , 52: 434-7.
Carlson, H. (1944-1945). Studies on the rate and amount of eruption of certain human teeth.
Am J Orthod Oral Surg , 1944-1945 (42:78-91), 42:78-91.
Chaillet, N., Nyström, M., Kataja, M., & Demirjian, A. (2004a). Dental maturity curves in
Finnish children: Demirjian's method revisited and polynomial functions for age estimation.
J Forensic Sci , 9(6):1324-31.
Chaillet, N., Willems, G., & Demirjian, A. (2004b). Dental maturity in Belgian children
using Demirjian's method and polynomial functions: new standard curves for forensic and
clinical use. J Forensic Odontostomatol , 22:18-27.
Chaillet, N., & Demirjian, A. (2004). Dental maturity in South France: A comparison
between Demirjian's method and polynomial functions. J Forensic Sci , 49:1059-1066.
Chen, J., Hu, H., Guo, J., Liu, Z., Liu, R., Li, F., et al. (2010a). Correlation between dental
maturity and cervical vertebral maturity. Oral Surg Oral Med Oral Pathol Oral Radiol
Endod , 110(6):777-83.
145
References
Chen, L., Liu, J., Xu, T., Long, X., & Lin, J. (2010b). Quantitative skeletal evaluation based
on cervical vertebral maturation: a longitudinal study of adolescents with normal occlusion.
Int J Oral Maxillofac Surg , 39(7):653-9.
Cho, S., & Hwang, C. (2009). Skeletal maturation evaluation using mandibular third molar
development in adolescents. Korean J Orthod , 39(2):120-129.
Cody, C., & Plan. (2009). Count every child: The right to birth registration. Woking, UK:
Publication production and design by Plan’s Global Publications Team.
Cohen, J. (2005). Comparing digital and conventional cephalometric radiographs. Am J
Orthod Dentofacial Orthop , 128(2):157-60.
Corradi, F., Pinchi, V., Barsanti, I., & Garatti, S. (2013). Probabilistic classification of age
by third molar development: the use of soft evidence. J Forensic Sci. , 58(1):51-9.
Crossner, C., & Mansfeld, L. (1983). Determination of dental age in adopted non-European
children. Swed Dent J , 7(1):1-10.
Cunha, E., Baccino, E., Martrille, L., Ramsthaler, F., Prieto, J., Schuliar, Y., et al. (2009).
The problem of aging human remains and living individuals: a review. Forensic Sci Int ,
15;193(1-3):1-13.
Dalitz, G. (1962). Age determination of adult human remains by teeth examination. J
Forensic Sci , 3,11-21.
De Coster, P., Marks, L., Martens, L., & Huysseune, A. (2009). Dental agenesis: genetic
and clinical perspectives. J Oral Pathol Med , 38(1):1-17.
de Oliveira, F., Capelozza, A., Lauris, J., & de Bullen, I. (2012). Mineralization of
mandibular third molars can estimate chronological age--Brazilian indices. Forensic Sci Int
, 10;219(1-3):147-50.
De Salvia, A., Calzetta, C., Orrico, M., & De Leo, D. (2004). Third mandibular molar
radiological development as an indicator of chronological age in a European population.
Forensic Sci Int , 146 Suppl:S9-S12.
Dean, M. (2010). Retrieving chronological age from dental remains of early fossil hominins
to reconstruct human growth in the past. Philos Trans R Soc Lond B Biol Sci ,
27;365(1556):3397-410.
Demirjian, A., Goldstein, H., & Tanner, J. (1973). A new system of dental age assessment.
Hum Biol , 45(2):211-27.
Demirjian, A., & Goldstein, H. (1976). New systems for dental maturity based on seven and
four teeth. Ann Hum Biol , 3(5):411-21.
Demisch, A., & Wartmann, P. (1956). Calcification of the mandibular third molar and its
relation to skeletal and chronological age in children. Child Dev , 27(4):459-73.
Dibbets, J., & Nolte, K. (2002). Effect of magnification on lateral cephalometric studies. Am
J Orthod Dentofacial Orthop , 122(2):196-201.
EC. (2004). European Guidelines on Radiation Protection in Dental Radiology. European
Commission RP Luxembourg , 136.
Engebretsen, L., Steffen, K., Bahr, R., Broderick, C., Dvorak, J., Janarv, P., et al. (2010).
The International Olympic Committee Consensus statement on age determination in highlevel young athletes. 2010 (44(7): 476-84).
146
References
Engström, C., Engström, H., & Sagne, S. (1983). Lower third molar development in relation
to skeletal maturity and chronological age. Angle Orthod , 53(2):97-106.
Erbudak, H., Ozbek, M., Uysal, S., & Karabulut, E. (2012). Application of Kvaal et al.'s age
estimation method to panoramic radiographs from Turkish individuals. Forensic Sci Int ,
219(1-3):141-6.
Fishman, L. (1982). Radiographic evaluation of skeletal maturation. A clinically oriented
method based on hand-wrist films. Angle Orthod , 52(2):88-112.
Gabriel, D., Southard, K., Qian, F., Marshall, S., Franciscus, R., & Southard, T. (2009).
Cervical vertebrae maturation method: poor reproducibility. Am J Orthod Dentofacial
Orthop , 136(4):478.e1-7; discussion 478-80.
Gabriel, P., Obertová, Z., Ratnayake, M., Arent, T., Cattaneo, C., Dose, M., et al. (2010).
Schätzung des Lebensalters kindlicher Opfer auf Bilddokumenten Rechtliche Implikationen
und Bedeutung im Ermittlungsverfahren. Rechtsmedizin , 21:7-11.
Garamendi, P., Landa, M., Ballesteros, J., & Solano, M. (2005). Reliability of the methods
applied to assess age minority in living subjects around 18 years old. A survey on a
Moroccan origin population. Forensic Sci Int , 10;154(1):3-12.
Garamendi, P., Landa, M., Botella, M., & Alemán, I. (2011). Forensic age estimation on
digital X-ray images: Medial epiphyses of the clavicle and first rib ossification in relation to
chronological age. J Forensic Sci , 56 Suppl 1:S3-12.
Garn, S., Lewis, A., Koski, K., & Polacheck, D. (1958). The sex difference in tooth
calcification. J Dent Res , 37(3):561-7.
Garn, S., Lewis, A., & Vicinus, J. (1963). Third molar polymorphism and its significance to
dental genetics. J Dent Res , 42:SUPPL1344-63.
Gelbrich, B., Lessig, R., Lehmann, M., Dannhauer, K., & Gelbrich, G. (2010).
Altersselektion in Referenzstichproben. Auswirkung auf die forensische Altersschätzung.
Rechtsmedizin , 20:459-463.
Gilsanz, V., & Ratib, O. (2005). Hand bone age: a digital atlas of skeletal maturity. Berlin:
Springer.
Gleiser, I., & Hunt, E. (1955). The permanent first molar: its calcification, eruption and
decay. Am. J. Phys. Anthropol. , 13:253–284.
Golovcencu, L., Scripcaru, C., & Zegan, G. (2009). Third molar development in relation to
chronological age in Romanian children and young adults. Rom J Leg Med , (4) 277 -282.
Greulich, W., & Pyle, S. S. (1959). Radiographic atlas of skeletal development of the hand
and wrist. Stanford, CA: Stanford University Press.
Griffin, R., Moody, H., Penkman, K., Fagan, M., Curtis, N., & MJ., C. (2008a). A new
approach to amino acid racemization in enamel: testing of a less destructive sampling
methodology. Forensic Sci Int , 53(4):910-6.
Griffin, R., Moody, H., Penkman, K., Fagan, M., Curtis, N., & Collins, M. (2008b). The
application of amino acid racemization in the acid soluble fraction of enamel to the
estimation of the age of human teeth. Forensic Sci Int , 25;175(1):11-6).
Guha-Sapir, D., Vos, F., Below, R., & Ponserre, S. (2012). Annual Disaster Statistical
Review 2011: The numbers and trends. Brussels: CRED.
147
References
Gunst, K., Mesotten, Carbonez, A., & Willems, G. (2003). Third molar root development in
relation to chronological age: a large sample sized retrospective study. Forensic Sci Int ,
136(1-3):52-7.
Gustafson, G. (1950). Age determination on teeth. J Am Dent Assoc, 41(1):45-54.
Gustafson, G., & Koch, G. (1974). Age estimation up to 16 years of age based on dental
development. Odontol Revy, 25(3):297-306.
Haavikko, K. (1970). The formation and the alveolar and clinical eruption of the permanent
teeth. An orthopantomographic study. Suom Hammaslaak Toim, 66(3):103-70.
Haavikko, K. (1974). Tooth formation age estimated on a few selected teeth. A simple
method for clinical use. Proceedings of the Finnish Dental Society, 70: 15–19.
Harrell, F. (2001). Regression modeling strategies. New-York: Springer.
Harris, E. (2007). Mineralization of the mandibular third molar: a study of American blacks
and whites. Am J Phys Anthropol, 132(1):98-109.
Harris, M., & Nortjé, C. (1984). The mesial root of the third mandibular molar. A possible
indicator of age. J Forensic Odontostomatol, 2(2):39-43.
Hassel, B., & Farman, A. (1995). Skeletal maturation evaluation using cervical vertebrae.
Am J Orthod Dentofacial Orthop, 107(1):58-66. Erratum in: 1995 Am J Orthod Dentofacial
Orthop, 107(6):19.
Healy, M. (1992). Normalizing transformations for growth standards. Ann Hum Biol ,
19(5):521-6.
Hedeker, D., & Gibbons, R. (1994). A random-effects ordinal regression model for
multilevel analysis. Biometrics , 50:933-944 .
Helfman, P., & Bada, J. (1976). Aspartic acid racemisation in dentine as a measure of
ageing. Nature. 262(5566):279-81.
Hickman, M., Hughes, K., Strom, K., & Ropero-Miller, J. (2007). Medical Examiners and
Coroners’ Offices, 2004. Washington, DC 20531: The Bureau of Justice Statistics.
Hillewig, E., De Tobel, J., Cuche, O., Vandemaele, P., Piette, M., & Verstraete, K. (2011).
Magnetic resonance imaging of the medial extremity of the clavicle in forensic bone age
determination: a new four-minute approach. Eur Radiol , 21(4):757-67.
Hoogeveen, J., Tesliuc, E., Vakis, R., & Dercon, S. (2005). Guide to the Analysis of
Risk,Vulnerability and Vulnerable Groups. (Mimeo. Social Protection Unit).
Hurme, V. (1949). Ranges of Normalcy in the Eruption of Permanent Teeth. J Dent for
Children , 16:11-15.
ICRP. (2007). The 2007 Recommendations of the International Commission on Radiological
Protection. ICRP publication 103 , Ann ICRP 37:1-332.
IOFOS. (2008, 02 12). Dental age estimation, Quality assurance. Retrieved 04 05, 2013, on
IOFOS.eu: http://www.iofos.eu/Quality-Ass/Age-IOFOS.htm
Israel, H., & Lewis, A. (1971). Radiographically determined linear permanent tooth growth
from age 6 years. J Dent Res. , 50: 334.
Jafari, A., Mohebbi, S., Khami, M., Shahabi, M., Naseh, M., Elhami, F., et al. (2012).
Radiographic evaluation of third molar development in 5- to 25 year olds in tehran, iran. J
Dent (Tehran) , 9(2):107-15.
148
References
Jaqueira, L., Armond, M., Pereira, L., Alcântara, C., & Marques, L. (2010). Determining
skeletal maturation stage using cervical vertebrae: evaluation of three diagnostic methods.
Braz Oral Res , 24(4):433-7.
Johan, N., Khamis, M., Abdul Jamal, N., Ahmad, B., & Mahanani, E. (2012). The variability
of lower third molar development in Northeast Malaysian population with application to
age estimation. J Forensic Odontostomatol , 1;30(1):45-54.
Johanson, G. (1971). Age determination from human teeth. Thesis. Odontologish Revy , 22:
suppl. 21: 1-126.
Kagerer, P., & Grupe, G. (2001). Age-at-death diagnosis and determination of life-history
parameters by incremental lines in human dental cementum as an identification aid.
Forensic Sci Int , 118(1):75-82.
Kahl, B., & Schwarze, C. (1988). Updating of the dentition tables of I. Schour and M.
Massler of 1941. Fortschr Kieferorthop , 49(5):432-43.
Kalt, A, Hossain, M, Kiss, L, & Zimmerman, C. (2013) Asylum Seekers, Violence and
Health: A Systematic Review of Research in High-Income Host Countries , Am J Public
Health., 103(3): e30-42.
Kanchan-Talreja, P., Acharya, A., & Naikmasur, V. (2012). An assessment of the versatility
of Kvaal's method of adult dental age estimation in Indians. Arch Oral Biol , 57(3):277-84.
17.
Karadayi, B., Kaya, A., Kolusayın, M., Karadayi, S., Afsin, H., & Ozaslan, A. (2012).
Radiological age estimation: based on third molar mineralization and eruption in Turkish
children and young adults. Int J Legal Med , 126(6):933-42.
Kasper, K., Austin, D., Kvanli, A., Rios, T., & Senn, D. (2009). Reliability of third molar
development for age estimation in a Texas Hispanic population: a comparison study. J
Forensic Sci , 54(3):651-7.
Kellinghaus, M., Schulz, R., Vieth, V., Schmidt, S., & Schmeling, A. (2010). Forensic age
estimation in living subjects based on the ossification status of the medial clavicular
epiphysis as revealed by thin-slice multidetector computed tomography. Int J Legal Med ,
124(2):149-54 .
Kim, Y., Kho, H., & Lee, K. (2000). Age estimation by occlusal tooth wear. J Forensic Sci ,
45:303-309.
Knell, B., Ruhstaller, P., Prieels, F., & Schmeling, A. (2009). Dental age diagnostics by
means of radiographical evaluation of the growth stages of lower wisdom teeth. Int J Legal
Med , 123(6):465-9.
Köhler, S., Schmelzle, R. L., & Püschel, K. (1994). Development of wisdom teeth as a
criterion of age determination. Ann. Anat , 176, 339–345.
Kondo-Nakamura, M., Fukui, K., Matsura, S., Kondo, M., & Iwadate, K. (2011). Single
tooth tells us the date of birth. Int J Legal Med , 125(6):873-7.
Kraus, B., & Jordan, R. (1965). The human dentition before birth (Vol. 1965). Philadelphia:
Lea and Febiger.
Kullman, L., Johanson, G., & Akesson, L. (1992). Root development of the lower third molar
and its relation to chronological age. Swed Dent J , 16(4):161-7.
Kvaal, S., & Solheim, T. (1995). Incremental lines in human dental cementum in relation to
age. Eur J Oral Sci , 103(4):225-30.
149
References
Kvaal, S., Kolltveit, K., Thomsen, I., & Solheim, T. (1995). Age estimation of adults from
dental radiographs. Forensic Sci Int , 28;74(3):175-85.
Lamendin, H., Baccino, E., Humbert, J., Tavernier, J., Nossintchouk, R., & Zerilli, A.
(1992). A simple technique for age estimation in adult corpses: the two criteria dental
method. J Forensic Sci , 37(5):1373-9.
Lamparski, D. (1972). Skeletal age assessment utilizing cervical vertebrae. [Thesis.] .
Pittsburgh: University of Pittsburgh .
Landa, M., Garamendi, P., Botella, M., & Alemán, I. (2009). Application of the method of
Kvaal et al. to digital orthopantomograms. Int J Legal Med , 23(2):123-8.
Lee, S., Lee, J., Park, H., & Kim, Y. (2009). Development of third molars in Korean
juveniles and adolescents. Forensic Sci Int ,188(1-3):107-11.
Legović, M., Sasso, A., Legović, I., Brumini, G., Cabov, T., Slaj, M., et al. (2010). The
reliability of chronological age determination by means of mandibular third molar
development in subjects in Croatia. J Forensic Sci , 55(1):14-8.
Leite, H., O'Reilly, M., & Close, J. (1987). Skeletal age assessment using the first, second,
and third fingers of the hand. Am J Orthod Dentofacial Orthop , 92(6):492-8.
Leroy, R., Cecere, S., Lesaffre, E., & Declerck, D. (2008 ). Variability in permanent tooth
emergence sequences in Flemish children. Eur J Oral Sci , 116(1):11-7.
Levesque, G., Demirijian, A., & Tanguay, R. (1981). Sexual dimorphism in the development,
emergence, and agenesis of the mandibular third molar. J Dent Res , 60(10):1735-41.
Lewis, A. (1991). Comparisons between dental and skeletal ages. Angle Orthod , 61(2):8792.
Lewis, J., & Senn, D. (2010). Dental age estimation utilizing third molar development: A
review of principles, methods, and population studies used in the United States. Forensic Sci
Int , 10;201(1-3):79-83.
Li, G., Ren, J., Zhao, S., Liu, Y., Li, N., Wu, W., et al. (2012). Dental age estimation from the
developmental stage of the third molars in western Chinese population. Forensic Sci Int ,
10;219(1-3):158-64.
Liversidge, H., & Molleson, T. (1999). Developing permanent tooth length as an estimate of
age. J Forensic Sci. , 44(5): 917-20.
Liversidge, H., Chaillet, N., Mörnstad, H., Nyström, M., Rowlings, K., Taylor, J., et al.
(2006). Timing of Demirjian's tooth formation stages. Ann Hum Biol , 33(4):454-70.
Liversidge, H. (2008a). Predicting Mandibular Third Molar Agenesis. Acta Stomatol Croat ,
42(4):311-317.
Liversidge, H. (2008b). Timing of human mandibular third molar formation. Ann Hum Biol ,
35(3):294-321.
Liversidge, H. (2009). Permanent tooth formation as a method of estimating age. Front Oral
Biol , 13:153-7.
Liversidge, H., & Marsden, P. (2010). Estimating age and the likelihood of having attained
18 years of age using mandibular third molars. Br Dent J , 23;209(8):E13.
Liversidge, H., Smith, B., & Maber, M. (2010). Bias and accuracy of age estimation using
developing teeth in 946 children. Am J Phys Anthropol , 143(4):545-54.
150
References
Logan, W., & Kronfeld, R. (1933). Development of the human jaws and surrounding
structures from birth to the age of fifteen years. J Am Dent Assoc , 20: 379-427.
Lorentsen, M., & Solheim, T. (1989). Age assessment based on translucent dentine. J
Forensic Odontostomatol , 7(2):3-9.
Lucy, D., Aykroyd, R., & Pollard, A. (2002). Non-parametric calibration for age estimation.
Appl Stat , 52: 185-196.
Massoglia, M, & Uggen, C. (2010). Settling down and aging out: toward an interactionist
theory of desistance and the transition to adulthood. AJS 116(2):543-82
Maat, G., Gerretsen, R., & Aarents, M. (2006). Improving the visibility of tooth cementum
annulations by adjustment of the cutting angle of microscopic sections. Forensic Sci Int ,
159 Suppl 1:S95-9.
Maples, W. (1978). An improved technique using dental histology for estimation of adult
age. J Forensic Sci , 23(4):764-70.
Martin-de las Heras, S., García-Fortea, P., Ortega, A., Zodocovich, S., & Valenzuela, A.
(2008). Third molar development according to chronological age in populations from
Spanish and Magrebian origin. Forensic Sci Int , 15;174(1):47-53.
Massler, M., & Schour, I. (1946). The Appositional Life Span of the Enamel and DentinForming Cells : I. Human Deciduous Teeth and First Permanent Molars. Introduction. J
Dent Res. , 25:145-50.
Mattila, K., & Haavikko, K. (1969). The correspondence between the orthopantomographic
and the clinical apperance of an erupting tooth (first molar). Odontol Tidskr , 77(1):39-45.
Meinl, A., Tangl, S., Pernicka, E., Fenes, C., & Watzek, G. (2007a). On the applicability of
secondary dentin formation to radiological age estimation in young adults. J Forensic Sci ,
52:438-441.
Meinl, A., Tangl, S., Huber, C., Maurer, B., & Watzek, G. (2007b). The chronology of third
molar mineralization in the Austrian population--a contribution to forensic age estimation.
Forensic Sci Int , 4;169(2-3):161-7.
Melsen, B., Wenzel, A., Miletic, T., Andreasen, J. V.-H., & Terp, S. (1986). Dental and
skeletal maturity in adoptive children: assessments at arrival and after one year in the
admitting country. Ann Hum Biol. , 153-9. .
Mesotten, K., Gunst, K., Carbonez, A., & Willems, G. (2002). Dental age estimation and
third molars: a preliminary study. Forensic Sci Int , 26;129(2):110-5.
Mesotten, K., Gunst, K., Carbonez, A., & Willems, G. (2003). Chronological age
determination based on the root development of a single third molar: a retrospective study
based on 2513 OPGs. J Forensic Odontostomatol , 21(2):31-5.
Mincer, H., Harris, E., & Berryman, H. (1993). The A.B.F.O. study of third molar
development and its use as an estimator of chronological age. J Forensic Sci , 38(2):379-90.
Molenberghs, G., & Verbeke, G. ( 2005). Models for discrete longitudinal data. New-York:
Springer.
Moore, G. (1970). Age changes occurring in the teeth. J Forensic Sci Soc , 10:179–180.
Moore, D., & McCabe, G. (1993). Introduction to the practice of statistics. New York: p.
854.
151
References
Moorrees, C., Fanning, E., & Hunt, E. J. (1963). Age variation of formation stages for ten
permanent teeth. J Dent Res , 42:1490-502.
Moskovitch, G., Dedouit, F., Braga, J., Rougé, D., Rousseau, H., & Telmon, N. (2010).
Multislice computed tomography of the first rib: a useful technique for bone age assessment.
Forensic Sci.Int , 55(4):865-70.
NCHS, (2003). Medical examiners' and coronors' handbook on death registration and fetal
death reporting. Hyattsville Maryland: DHHS Publication.
Nelki, J., P, G., & Bailey, S. (2010). Challenges of psychological assessments of maturity.
Dans S. Black, A. Aggrawal, & J. Payne-James, Age estimation in the living, 1st Edn. (pp.
55–76.). UK: Wiley-Blackwell, UK.
Nieminen, P. (2009). Genetic basis of tooth agenesis. J Exp Zool B Mol Dev Evol ,
15;312B(4):320-42. .
Nortjé, C. (1983). The permanent mandibular third molar. Its value in age determination. J
Forensic Odontostomatol , 1(1):27-31.
Nuzzolese, E., & Di Vella, G. (2008). Forensic dental investigations and age assessment of
asylum seekers. Int Dent J , 58(3):122-6.
Nyström, M., Ranta, H., Peltola, J., & Kataja, J. (2007). Timing of developmental stages in
permanent mandibular teeth of Finns from birth to age 25. Acta Odontol Scand , 65(1):3643.
Ohtani, S. (1994). Age estimation by aspartic acid racemization in dentin of deciduous teeth.
Forensic Sci Int , 68(2):77-82.
Ohtani, S. (1995a). Estimation of age from the teeth of unidentified corpses using the amino
acid racemization method with reference to actual cases. Am J Forensic Med Pathol ,
16(3):238-42.
Ohtani, S. (1995b). Studies on age estimation using racemization of aspartic acid in
cementum. J Forensic Sci , 40(5):805-7.
Ohtani, S., Ito, R., Arany, S., & Yamamoto, T. (2005). Racemization in enamel among
different types of teeth from the same individual. Int J Legal Med , 119(2):66-9.
Olze, A., Taniguchi, M., Schmeling, A., Zhu, B., Yamada, Y., Maeda, H., et al. (2003).
Comparative study on the chronology of third molar mineralization in a Japanese and a
German population. Leg Med (Tokyo) , 5 Suppl 1:S256-60.
Olze, A., Schmeling, A., Taniguchi, M., Maeda, H., van Niekerk, P., Wernecke, K., et al.
(2004a). Forensic age estimation in living subjects: the ethnic factor in wisdom tooth
mineralization. Int J Legal Med , 118(3):170-3.
Olze, A., Taniguchi, M., Schmeling, A., Zhu, B., Yamada, Y., Maeda, H., et al. (2004b).
Studies on the chronology of third molar mineralization in a Japanese population. Leg Med
(Tokyo) , (2):73-9.
Olze, A., Bilang, D., Schmidt, S., Wernecke, K., Geserick, G., & Schmeling, A. (2005).
Validation of common classification systems for assessing the mineralization of third
molars. Int J Legal Med. , 119: 22-6.
Olze, A., Reisinger, W., Geserick, G., & Schmeling, A. (2006a). Age estimation of
unaccompanied minors. Part II. Dental aspects. Forensic Sci Int , 15;159 Suppl 1:S65-7.
152
References
Olze, A., van Niekerk, P., Schmidt, S., Wernecke, K., Rösing, F., Geserick, G., et al. (2006b).
Studies on the progress of third-molar mineralisation in a Black African population. Homo ,
57(3):209-17.
Olze, A., Solheim, T., Schulz, R., Kupfer, M., Pfeiffer, H., & Schmeling, A. (2010a).
Assessment of the radiographic visibility of the periodontal ligament in the lower third
molars for the purpose of forensic age estimation in living individuals. Int J Legal Med ,
124(5):445-8
Olze, A., Pynn, B., Kraul, V., Schulz, R., Heinecke, A., Pfeiffer, H., et al. (2010b). Studies on
the chronology of third molar mineralization in First Nations people of Canada. Int J Legal
Med , 124(5):433-7..
Orhan, K., Ozer, L., Orhan, A., Dogan, S., & Paksoy, C. (2006). Radiographic evaluation of
third molar development in relation to chronological age among Turkish children and
youth. Forensic Sci Int , 5;165(1):46-51.
Ozer, T., Kama, J., & Ozer, S. (2006). A practical method for determining pubertal growth
spurt. Am J Orthod Dentofacial Orthop , 130(2):131.e1-6.
Paewinsky, E., Pfeiffer, H., & Brinkmann, B. (2005). Quantification of secondary dentine
formation from orthopantomograms-a contribution to forensic age estimation methods in
adults. Int J Legal Med , 2005 (119(1):27-30).
Perinetti, G., Contardo, L., Gabrieli, P., Baccetti, T., & Di Lenarda, R. (2011). Diagnostic
performance of dental maturity for identification of skeletal maturation phase. Eur J Orthod
, 34(4):487-92.
Philippas, G., & Applebaum, E. (1966). Age factor in secondary dentin formation. J Dent
Res , 45:778–789.
Prahl-Anderson, B., & van der Linder, F. (1972). The estimation of dental age. Trans Eur
Orthod Soc , 535-41.
Prieto, J., Barbería, E., Ortega, R., & Magaña, C. (2005). Evaluation of chronological age
based on third molar development in the Spanish population. Int J Legal Med , 119(6):34954.
Prince, D., & Konigsberg, L. (2008). New formulae for estimating age-at-death in the
Balkans utilizing Lamendin's dental technique and Bayesian analysis. J Forensic Sci ,
53:578-587.
Prince, D., Kimmerle, E., & Konigsberg, L. (2008). A Bayesian approach to estimate
skeletal age-at-death utilizing dental wear. J Forensic Sci , 53:588-593.
Quirmbach, F., Ramsthaler, F., & Verhoff, M. (2009). Evaluation of the ossification of the
medial clavicular epiphysis with a digital ultrasonic system to determine the age threshold
of 21 years. Int J Legal Med , 23(3):241-5.
Rai, B., Krishan, K., Kaur, J., & Anand, S. (2008). Age estimation from mandible by lateral
cephalogram: a preliminary study. J Forensic Odontostomatol , 27:1:24-28.
Rai, B., Kaur, J., & Anand, S. (2009). Mandibular third molar development staging to
chronologic age and sex in north Indian children and young adults. J Forensic
Odontostomatol , 1;27(2):45-9.
Raungpaka, S. (1988). The study of tooth-developmental age of Thai children in Bangkok.
Journal of the Dental Association of Thailand , 38: 72-81.
153
References
Resolution, G. A. (44/25 1989, Novembre 20). Convention on the Rights of the Child,
resolution 44/25 1989. Retrieved march 03, 2013, on Child rights international network:
http://www.crin.org/docs/resources/treaties/uncrc.asp
Risnes, S. (1987 ). Multiplane sectioning and scanning electron microscopy as a method for
studying the three-dimensional structure of mature dental enamel. Scanning Microsc ,
1(4):1893-902.
Ritz-Timme, S., Cattaneo, C., Collins, M., Waite, E., Schütz, H., Kaatsch, H., et al. (2000).
Age estimation: the state of the art in relation to the specific demands of forensic practise.
Int J Legal Med , 113(3):129-36.
Ritz-Timme, S., Kaatsch, H., & Marré. (2002). Empfehlungen für die alterdiagnostik bei
lebenden im rentenverfahren. German academy of forensic odontostomatology , 9/1-3; 9,93
.
Roberts, G., Parekh, S., Petrie, A., & Lucas, V. (2008). Dental age assessment (DAA): a
simple method for children and emerging adults. Br Dent J. 23;204(4):E7; discussion 1923.
Rozkovcová, E., Marková, M., Láník, J., & Zvárová, J. (2004). Development of third molar
in the Czech population. Prague Med Rep , 105(4):391-422.
Rozkovcová, E., Marková, M., & Mrklas, L. (2005). Third Molar as an Age Indicator in
Young Individuals. Prague Medical Report , 106 No. 4, p. 367–398.
Rozkovcova, E., Dostalova, T., Markova, M., & Brouka, l. Z. (2012). The third molar as an
age marker in adolescents: new approach to age evaluation. J Forensic Sci , 57(5):1323-8.
Różyło-Kalinowska, I., Kolasa-Rączka, A., & Kalinowski, P. (2011). Relationship between
dental age according to Demirjian and cervical vertebrae maturity in Polish children. Eur J
Orthod , 33(1):75-83.
Rushton, M. (1933). On the fine contour lines of the enamel of milk teeth. Dent Res ,
53:170–171.
Sabel, N., Johansson, C., Kühnisch, J., Robertson, A., Steiniger, F., Norén, J., et al. (2008).
Neonatal lines in the enamel of primary teeth--a morphological and scanning electron
microscopic investigation. 53(10):954-63.
Sakuma, A., Ohtani, S., Saitoh, H., & Iwase, H. (2012). Comparative analysis of aspartic
acid racemization methods using whole-tooth and dentin samples. Forensic Sci Int , 223(13):198-201.
San Román, P., Palma, J., Oteo, M., & Nevado, E. (2002). Skeletal maturation determined
by cervical vertebrae development. Eur J Orthod , 24(3):303-11.
Santoro, V., De Donno, A., Marrone, M., Campobasso, C., & Introna, F. (2009). Forensic
age estimation of living individuals: a retrospective analysis. Forensic Sci Int , 15;193(13):129.
Santoro, V., Lozito, P., Mastrorocco, N., & Introna, F. (2008). Morphometric analysis of
third molar root development by an experimental method using digital orthopantomographs.
J Forensic Sci. , 53(4):904-9.
Scheurer, E., Quehenberger, F., Mund, M., Merkens, H., & Yen, K. (2011). Validation of
reference data on wisdom tooth mineralization and eruption for forensic age estimation in
living persons. Int J Legal Med , 125(5):707-15.
154
References
Schmeling, A., Olze, A., Reisinger, W., & Geserick, G. (2001). Age estimation of living
people undergoing criminal proceedings. Lancet , 14;358(9276):89-90.
Schmeling, A., Schulz, R., Reisinger, W., Mühler, M., Wernecke, K., & Geserick, G. (2004).
Studies on the time frame for ossification of the medial clavicular epiphyseal cartilage in
conventional radiography. Int J Legal Med , 118(1):5-8.
Schmeling, A., Reisinger, W., Geserick, G., & Olze, A. (2006). Age estimation of
unaccompanied minors. Part I. General considerations. Forensic Sci Int , 15;159 Suppl
1:S61-4
Schmeling, A., Geserick, G., Reisinger, W., & Olze, A. (2007). Age estimation. Forensic Sci
Int , 17;165(2-3):178-81.
Schmeling, A., Grundmann, C., Fuhrmann, A., Kaatsch, H., Knell, B., Ramsthaler, F., et al.
(2008). Criteria for age estimation in living individuals. Int J Legal Med , 122(6):457-60.
Schmeling, A., Olze, A., Pynn, B., Kraul, V., Schulz, R., Heinecke, A., et al. ( 2010). Dental
age estimation based on third molar eruption in First Nation people of Canada. J Forensic
Odontostomatol , 1;28(1):32-8.
Schopf, P. (1970). Root calcification and tooth eruption in the mixed dentition. A study in
panoramic x-rays. Fortschr Kieferorthop , 1970 (31(1):39-56.), 31(1):39-56.
Schour, I., & Massler, M. (1941). Development of the human dentition. J Am Dent Assoc ,
28: 1153-1160.
Schulz, R., Mühler, M., Mutze, S., Schmidt, S., Reisinger, W., & Schmeling, A. (2005).
Studies on the time frame for ossification of the medial epiphysis of the clavicle as revealed
by CT scans. Int J Legal Med , 119:142-145.
Schulze, D., Rother, U., Fuhrmann, A., Richel, S., Faulmann, G., & Heiland, M. (2006).
Correlation of age and ossification of the medial clavicular epiphysis using computed
tomography. Forensic Sci Int , 10;158(2-3):184-9.
Schulz, R., Mühler, M., Reisinger, W., Schmidt, S., & Schmeling, A. (2008a). Radiographic
staging of ossification of the medial clavicular epiphysis. Int J Legal Med , 122:55-58.
Schulz, R., Zwiesigk, P., Schiborr, M., Schmidt, S., & Schmeling, A. (2008b). Ultrasound
studies on the time course of clavicular ossification. Int J Legal Med , 122(2):163-7.
Seedat, A., & Forsberg, C. (2005). An evaluation of the third cervical vertebra (C3) as a
growth indicator in Black subjects. SADJ , 60(4):156, 158-60.
Sharma, R., & Srivastava, A. (2010). Radiographic evaluation of dental age of adults using
Kvaal's method. J Forensic Dent Sci , 2(1):22-6.
Shumaker, D., & El Hadary, M. (1960). Roentgenograhic study of eruption. J.A.D.A. , 61:
535-41.
Sierra, A. (1987). Assessment of dental and skeletal maturity. A new approach. Angle
Orthod , 57(3):194-208.
Sisman, Y., Uysal, T., Yagmur, F., & Ramoglu, S. (2007). Third-molar development in
relation to chronologic age in Turkish children and young adults. Angle Orthod ,
77(6):1040-5.
Smith, B. (1991). Standards of Human Tooth Formation and Dental Age Assessment. Dans
K. MA, L. CS, & editors, Advances in Dental Anthropology (pp. 143-168). New York: WileyLiss, Inc.
155
References
Smith, S., & Buschang, P. (2010). An examination of proportional root lengths of the
mandibular canine and premolars near the time of eruption. Am J Orthod Dentofacial
Orthop , Orthod Dentofacial Orthop , 138:795-803).
Smith, T, & Brownlees, L. (2011) Age assessment practices: a literature review & annotated
bibliography.. United Nations Children’s Fund (UNICEF), New York .
Solari, A., & Abramovitch, K. (2002). The accuracy and precision of third molar
development as an indicator of chronological age in Hispanics. J Forensic Sci , 47(3):5315.
Solheim, T. (1984). Dental age estimation. An alternative technique for tooth sectioning. Am
J Forensic Med Pathol , 5(2):181-4.
Solheim, T. (1988a). Dental attrition as an indicator of age. Gerodontics , 4(6):299-304.
Solheim, T. (1988b). Dental color as an indicator of age , (4(3):114-8. ).
Solheim, T. (1990). Dental cementum apposition as an indicator of age. Scand J Dent Res ,
98(6):510-9.
Solheim, T. (1992a). Amount of secondary dentin as an indicator of age. Scand J Dent Res ,
100(4):193-9.
Solheim, T. (1992b). Recession of periodontal ligament as an indicator of age. J Forensic
Odontostomatol , 10:32-42.
Solheim, T. (1993). A new method for dental age estimation in adults. Forensic Sci Int ,
59(2):137-47.
Solheim, T., & Kvaal, S. (1993). Dental root surface structure as an indicator of age. J
Forensic Odontostomatol , 11(1):9-21.
Solheim, T., & Vonen, A. (2006). Dental age estimation, quality assurance and age
estimation of asylum seekers in Norway. Forensic Sci Int , 15;159 Suppl 1:S56-60.
Someda, H., Saka, H., Matsunaga, S., Ide, Y., Nakahara, K., Hirata, S., et al. (2009). Age
estimation based on three-dimensional measurement of mandibular central incisors in
Japanese. Forensic Sci Int , 185(1-3):110-4.
Spalding, K., Buchholz, B., Bergman, L., Druid, H., & Frisén, J. (2005). Forensics: age
written in teeth by nuclear tests. Nature , 437(7057):333-4.
Star, H., Thevissen, P., Jacobs, R., Fieuws, S., Solheim, T., & Willems, G. (2011). Human
dental age estimation by calculation of pulp-tooth volume ratios yielded on clinically
acquired cone beam computed tomography images of monoradicular teeth. J Forensic Sci ,
56 Suppl 1:S77-82.
Sun-Mi, C., & Chung-Ju, H. (2009). Skeletal maturation evaluation using mandibular third
molar development in adolescents. Korean J Orthod , 39(2):120-129).
Tanner, J., Whitehouse, R., & Healy, M. (1962). A new system for estimating skeletal
maturity from the hand and wrist, with standards derived from a study of 2.600 healthy
British children. Paris: International Children's Centre.
Tanner, J. (1975). Assessment of skeletal maturity and prediction of adult height (TW2
method). New York.: Academic Press, London .
Tanner, J. (1986). Normal growth and techniques of growth assessment . Clin Endocrinol
Metab , 15(3):411-51.
156
References
Tanner, J., Healey, M., Goldstein, H., & Cameron, N. (2001). Assessment of Skeletal
Maturity and Prediction of Adult Height (TW3 Method) Ed 3. . Philadelphia: Saunders.
Thevissen, P., Pittayapat, P., Fieuws, S., & Willems, G. (2009). Estimating age of majority
on third molarsdevelopmental stages in young adults from Thailand using a modified
scoring technique. J Forensic Sci.54 , 54, 428-32.
Thevissen, P., Alqerban, A., Asaumi, J., Kahveci, F., Kaur, J., Kim, Y., et al. (2010a).
Human dental age estimation using third molar developmental stages: Accuracy of age
predictions not using country specific information. Forensic Sci Int. , 10;201(1-3):106-11.
Thevissen, P., Fieuws, S., & Willems, G. (2010b). Human dental age estimation using third
molar developmental stages: does a Bayesian approach outperform regression models to
discriminate between juveniles and adults? Int J Legal Med. , 124:35-42.
Thevissen, P., Fieuws, S., & Willems, G. (2010c). Human third molars development:
Comparison of 9 country specific populations. Forensic Sci Int , 10;201(1-3):102-5.
Thevissen, P., Fieuws, S., & Willems, G. (2011). Third molar development: measurements
versus scores as age predictor. Arch Oral Biol. , 56(10):1035-40.
Thevissen, P., Kvaal, S., Dierickx, K., & Willems G. (2012). Ethics in age estimation of
unaccompanied minors. J Forensic Odontostomatol. 2012, 30 Suppl 1:84-102.
Thorson, J., & Hägg, U. (1991). The accuracy and precision of the third mandibular molar
as an indicator of chronological age. Swed Dent J , 15(1):15-22.
Tonge, C. (1969). The time-structure relationship to tooth development in human
embryogenesis. J Dent Res , 48(5):745-52.
Tukey, J. (1977). Exploratory data analysis. Addison-Wesley, Reading (1977).: AddisonWesley, Reading.
Ubelaker, D. (1978). Human skeletal remains. Excavation, analysis, and interpretation.
Chicago: Aldine.
UNICEF. (2007). BIRTH REGISTRATION AND ARMED CONFLICT. Florence It:
UNICEF Innocenti Research Centre.
Vandevoort, F., Bergmans, L., Van Cleynenbreugel, J., Bielen, D., Lambrechts, P., Wevers,
M., et al. (2004). Age calculation using X-ray microfocus computed tomographical scanning
of teeth: a pilot study. J Forensic Sci , 49(4):787-90.
Waite, E., Collins, M., Ritz-Timme, S., Schutz, H., Cattaneo, C., & Borrman, H. (1999). A
review of the methodological aspects of aspartic acid racemization analysis for use in
forensic science. Forensic Sci Int , 103(2):113-24.
Wedl, J., & Friedrich, R. (2005). Measuring the distance of the wisdom teeth from the
occlusal plane as forensic-odontological method for chronological age determination. Arch
Kriminol , 215(3-4):77-84.
Wetgeving. (2002). belgiëlex . be - Kruispuntbank Wetgeving. Retrieved march 3, 2013, on
24 DECEMBER 2002. - Programmawet (I) (art. 479) - Titel XIII - Hoofdstuk VI :
<Voogdij>
over
niet-begeleide
minderjarige
vreemdelingen.:
http://www.ejustice.just.fgov.be/cgi_loi/loi_a1.pl?DETAIL=2002122445%2FN&caller=list
&row_id=1&numero=1&rech=2&cn=2002122445&table_name=wet&nm=2002A21488&
la=N&ddfm=12&chercher=t&dt=WET&language=nl&choix1=EN&choix2=EN&text1=v
oogdij&fromtab=wet_all&nl=n&sql=
157
References
Willems, G. (2001). A review of the most commonly used dental age estimation techniques. J
Forensic Odontostomatol , 19:9-17.
Willems, G., Van Olmen, A., Spiessens, B., & Carels, C. (2001). Dental age estimation in
Belgian children: Demirjian's technique revisited. J Forensic Sci , 46(4):893-5.
Willershausen, B., Löffler, N., & Schulze, R. (2001). Analysis of 1202 orthopantograms to
evaluate the potential of forensic age determination based on third molar developmental
stages. Eur J Med Res , 28;6(9):377-84.
Wright, C., Booth, I., Buckler, J., Cameron, N., Cole, T., Healy, M., et al. (2002). Growth
reference charts for use in the United Kingdom. Arch Dis Child , 86(1):11-4.
Yang, F., Jacobs, R., & Willems, G. (2006). Dental age estimation through volume matching
of teeth imaged by cone-beam CT. Forensic Sci Int , 15;159 Suppl 1:S78-83.
Zanolli, C., Bondioli, L., Manni, F., Rossi, P., & Macchiarelli, R. (2011). Gestation length,
mode of delivery, and neonatal line-thickness variation. Hum Biol , 83(6):695-713.
Zeng, D., Wu, Z., & Cui, M. (2010). Chronological age estimation of third molar
mineralization of Han in southern China. Int J Legal Med. , 24(2):119-23.
158
Summary
Increasing global human migration, raises management concerns in the
countries where immigrants seek shelter. A special protective status must be
given to immigrating unaccompanied children. Therefore, most national
laws enforce specialized medical investigations to get proof of the age of
unaccompanied youngsters with no, or lacking official identification
documents and claiming to be minors. Dental age estimation in this
particular age group relies on the only dental age predictor(s) available,
namely the developing third molar(s). Hence, scientific correct dental age
estimations in sub-adults, especially when originating from distant countries
and diverse biological origin are requested. The general research aim was to
optimize dental age estimation based on third molar development.
Panoramic radiographs were retrospectively and cross-sectional
sampled to collect data registering third molar development. For that
registration, two techniques were described. The sequence of third molar
development was divided in succeeding stages, and the observed third molar
development was classified in the corresponding stage. Otherwise, the
dimensions of third molars increase during its maturation and measures of
the observed third molar sizes were registered. In the current thesis, both
registration techniques were compared. Third molar stages (categorical data)
were best related to age and provided the most accurate age predictions
compared to all collected tooth measurements and ratios of tooth
measurements (continuous data). Combining the scored third molar stages
with tooth measurements or ratios did not contribute to a clinical relevant
information gain for age prediction.
Multiple tooth development staging techniques were reported, based
on the described and considered borderlines between succeeding stages the
quantity of stages covering the third molar development process differs
between techniques. Therefore, it was studied if the number of stages used in
a staging technique is influencing the age prediction performances. The
number of stages utilized in the third molar registration technique slightly
influenced the age predictions. The choice of third molar development
registration technique has to depend on its stages described for the
developmental period of interest and should not compromise the feasibility
of correctly registering all these stages.
The classical approach for age estimation uses regression analysis to
model the collected reference data. Drawbacks of this technique concern the
age distribution of the residuals, the high correlation between the
independent variables, often observed missing values of the independent
variables, and a systematic bias in the age predictions. Therefore, a Bayesian
approach of age estimation on third molar development was established in
159
Summary
the current study. The age prediction performances of both approaches were
compared. Both models provided similar accuracy, precision and coverage in
age estimation outcome. The Bayesian approach reduced the bias that is
typically present in the regression models. The age of juveniles was less
overestimated, yielding a better discrimination between subjects older or
younger than 18 years. Moreover, the Bayesian model integrated all
available third molar information.
Sub-adult age estimations are mostly requested to discriminate a
child from an adult during migration and asylum procedures. Due to the
migration aspect, frequently the age of an applicant with a particular
geographical and biologic origin was estimated using methods or models
developed on a reference sample, including subjects with unlike origin. It
was investigated whether differences in third molar development between
populations with different geographic and biological origin exist. Therefore,
third molar development was analysed and compared on 13 country-specific
samples using a factor analysis. Differences in third molar development
between countries exist, but they were not constant over age and varied in an
unordered way. Because the magnitude of the differences turned out to be
small there was no evidence for important differences in third molar
development between the countries.
Age estimation models developed on a particular country-specific
reference sample were validated on their age prediction performances using
a validation sample from a different geographic and biological origin as the
reference sample. Validated on 13 country-specific databases using
information from Belgium, or all countries pooled together changes the
difference between observed and predicted age obtained on country-specific
information only slightly. For the adult-juvenile discrimination, the Belgium
reference model provided a maximal advantage of the doubt to investigated
unaccompanied minor fugitives. The reference model based on all pooled
countries, substituted the country-specific reference model most accurately
in sub-adults.
The age prediction performances of age estimation models
constructed on a single age-related variable are possibly ameliorated, adding
age-related information of one or more variables present in the considered
period of life. Therefore, reference samples registering at a specific moment
third molar development, as well as tooth morphological or skeletal agerelated variables, were collected, modelled and validated. Due to the inherent
image quality of panoramic radiographs tooth morphological measurements
based on secondary dentine apposition, could only be achieved on a
restricted sample. Clinically the gain in age prediction accuracy was
negligible when adding the time consuming additional tooth morphological
measurements to the staged third molar development. On cephalometric
radiographs the skeletal age predicting variable(s) and related registration
systems providing the most information on age were cervical vertebrae
160
Summary
scoring systems. Combining the information from cervical vertebrae and
third molars improved the age predictions drastically in the period of early
third molar development. In sub-adults no, or a negligible, gain in accuracy
of age prediction was obtained.
In an optimal approach, dental age estimations in sub-adults
are based on the radiologically observed developmental status or the
absence, of each third molar. The observed information is registered
according to a staging technique. The collected reference information is
modeled as dependent ordinal data in a multivariate Bayesian approach.
Despite detected country-specific differences in third molar development,
the clinical impact on age estimation is minimal. Based on an ongoing
country-specific data collection, a quantification of the maximal difference
in prediction error using not county-specific reference information is
established. Further on, the Belgium reference information classifies more
juveniles compared to country-specific information and is recommended in
lack of a country-specific reference model for age estimations of young
unaccompanied fugitives.
161
Samenvatting
Wereldwijd is er een toenemende migratie van mensen. Dit veroorzaakt
beheerproblemen in de landen waar migranten onderkomen zoeken. In het
bijzonder moet elk land een specifieke beschermende status toekennen aan
immigrerende, niet begeleide kinderen. Daartoe voorzien de meeste
nationale rechtspraken in medische testen om leeftijd te bepalen. Die kunnen
worden uitgevoerd wanneer niet begeleide minderjarige jongelingen, geen of
ontbrekende officiële identificatie documenten kunnen voorleggen en
beweren minderjarig te zijn. Dentale leeftijdsschattingen zijn gebaseerd op
de enig aanwezige dentale leeftijdsschatter die in deze leeftijdsgroep
voorkomt, namelijk de zich ontwikkelende derde molaar. Deze
leeftijdsschattingen moeten wettelijk ontegensprekelijk zijn en dienen
daarom gebaseerd te zijn op wetenschappelijke kennis. Deze kennis is
voornamelijk vereist omdat de onderzochte personen afkomstig zijn uit
diverse landen en een verschillende biologische herkomst hebben. Daarom
omvat het algemeen onderzoeksdoel van deze thesis de optimalisatie van
leeftijdsschattingen gebaseerd op de ontwikkeling van derde molaren.
Panoramische röntgenopnames werden retrospectief en ad random
geselecteerd en gegroepeerd om data die de derde molaarontwikkeling
registreren te kunnen verzamelen. Deze registratie werd beschreven in twee
technieken. Enerzijds werd het ontwikkelingstraject van de derde molaar
opgedeeld in opeenvolgende stadia en de waargenomen ontwikkeling werd
geclassificeerd in een overeenstemmend stadium. Anderzijds, nemen de
dimensies van de derde molaar toe tijdens zijn ontwikkeling en de
geobserveerde afmetingen kunnen worden gemeten en geregistreerd. In deze
studie werden beide registratietechnieken met elkaar vergeleken. Derde
molaar stadia (categorische data) werden best gerelateerd met leeftijd en
verstrekten de meest accurate leeftijdsvoorspellingen vergeleken met alle
tandmetingen en hun verhoudingen (continue data). Een gecombineerd
gebruik van gescoorde derde molaar stadia en tandafmetingen (of hun
verhoudingen) resulteerden niet in een klinisch relevante verbetering van de
leeftijdsvoorspelling.
Meerdere technieken ontwikkeld op de tandontwikkelingsstadia
werden beschreven, gebaseerd op het aantal stadia en het aantal vastgelegde
grenzen tussen opeenvolgende stadia. Daarom werd onderzocht wat de
invloed van het aantal stadia in een techniek was op de leeftijdsschatting.
Het aantal stadia gebruikt in een bepaalde techniek beïnvloedt slechts gering
de leeftijdsvoorspelling. De keuze van gebruikte registratie techniek dient af
te hangen van het aantal stadia voorzien in de betrokken leeftijdszone en
moet toelaten om elk geobserveerd stadium correct te classificeren.
163
Samenvatting
In een klassieke aanpak wordt door toepassing van regressie
analyses een leeftijdsbepaling model geconstrueerd op verzamelde referentie
data. Dit model heeft echter beperkingen. Die zijn het gevolg van de
verdeling van de restwaarden, de hoge correlatie tussen de onafhankelijke
variabelen, de vaak voorkomende afwezigheid van deze variabelen en een
systematische vertekening van de leeftijdsvoorspellingen. Daarom werd in
deze studie een Bayesiaans model voor leeftijdsbepaling, gebaseerd op de
derde molaarontwikkeling, ontworpen. De leeftijdschattingsprestaties van
beide modellen werden vergeleken. Elk model leverde een gelijke
accuraatheid, precisie en “coverage” in leeftijdsschatting. Het Bayesiaanse
model verminderde de systematische vertekening die regressie-analyse
typeert. Bovendien werd hiermee de leeftijd van jonge individuen minder
overschat, en zorgde het voor een beter onderscheid tussen individuen jonger
of ouder dan 18 jaar. Bovendien kon in het Bayesiaans model de
leeftijdsinformatie van alle derde molaren worden geïntegreerd.
Eigen aan migratie dient dikwijls de leeftijd van een individu met
een specifieke geografische of biologische oorsprong te worden geschat op
basis van een referentie methode of -model ontwikkeld op een steekproef
met individuen van andere origine. Daarom werd onderzocht of er
onderscheid in derde molaarontwikkeling bestaat tussen populaties met
verschillende oorsprong. Hiertoe werd met behulp van een factor analyse, de
derde molaarontwikkeling geanalyseerd en vergeleken tussen 13 landspecifieke steekproeven. Er werden verschillen in derde molaarontwikkeling
tussen deze groepen vastgesteld, maar deze waren niet constant in functie
van leeftijd en varieerden op een ongeordende wijze. De grootte van deze
verschillen was gering.
Een leeftijdsschattingsmodel geconstrueerd op een bepaalde landspecifieke referentie steekproef kan worden gevalideerd op leeftijdschatting
met behulp van een land-specifieke validatie steekproef. Een validatie van
13 land-specifieke modellen met behulp van een Belgische validatie
steekproef en een validatie steekproef die de 13 landen groepeert, veranderde
nauwelijks het verschil tussen geschatte en chronologische leeftijd verkregen
bij een land-eigen validatie. Tijdens de kind-volwassen classificatie van niet
begeleide jonge vluchtelingen gaf de Belgische referentie steekproef een
maximaal voordeel van de twijfel. Het referentie model gebaseerd op de 13
landen gegroepeerd, verving bij jong volwassenen het meest accuraat het
land specifieke referentie model.
Leeftijdsvoorspellingen verkregen met leeftijdsbepaling modellen
gebaseerd op één leeftijdsgerelateerde veranderlijke, worden mogelijks
accurater wanneer één of meer leeftijdsgerelateerde veranderlijken, die
aanwezig zijn in dezelfde leeftijdsgroep, aan het model worden toegevoegd.
Daarom werden referentie steekproeven waarin op een bepaald moment
zowel derde molaarontwikkeling als leeftijdsgebonden tandmorfologische
veranderlijken en derde molaarontwikkeling en skeletale leeftijdsgebonden
164
Samenvatting
veranderlijken aanwezig waren, verzameld, gemodelleerd en gevalideerd.
Bij jongvolwassenen was de winst in accuraatheid van leeftijdsvoorspelling
verwaarloosbaar door toevoeging van tandmorfologische variabelen
gebaseerd op de afzetting van secondair dentine. De toegevoegde skeletale
variabelen gebaseerd op de ontwikkeling van de halswervellichamen
zorgden voor een toegevoegde waarde bij kinderen maar niet bij
jongvolwassenen.
In een optimale benadering worden dentale leeftijdsbepalingen bij
jongvolwassenen verricht aan de hand van het radiografisch geobserveerde
ontwikkelingstadium, of de afwezigheid, van elke derde molaar. De
waarnemingen worden geregistreerd door middel van een stadium techniek.
De verzamelde referentie informatie wordt als afhankelijke ordinale data
gemodelleerd in een Bayesiaanse aanpak. Ondanks de vaststelling dat
verschillen in derde molaarontwikkeling bestaan tussen land-specifieke
populaties, is de klinische impact op de leeftijdsvoorspellingen minimaal.
Gebaseerd op een land-specifieke data verzameling is de grootte van het
maximale verschil in leeftijdsschatting gebruikmakend van niet landspecifieke informatie bepaald. De Belgische referentie informatie
classificeert meer jongeren correct vergeleken met land-specifieke
informatie. Daarom is het aanbevolen om, bij gebrek aan land-specifieke
informatie, tijdens leeftijdsschattingen bij jonge niet begeleide vluchtelingen
de Belgische referentie informatie te gebruiken.
165
Curriculum vitae
Personal data
Last Name
Given Names
Address
Date of birth
Place of birth
Email
Thevissen
Patrick, Werner, Cyrille
Dendermondsesteenweg 483 9040 Gent (Belgium)
04 06 1956
Gent
Denthepa@telenet.be
Diplomas
1974
1980
2005
High School, Koninklijk Atheneum Gent West
Dentist (DDS), Rijksuniversiteit te Gent
Master after master Forensic Odontology (MSc), Katholieke
Universiteit Leuven
Oral presentations
2006
2006
2009
2009
2009
2009
2009
2009
2009
2010
2010
2010
2010
2010
2011
AAFS annual meeting, Seattle (USA), RFID tags: working principle
IOFOS meeting, Leuven (Belgium), RFID tags: physical properties.
AAFS annual meeting, Denver (USA), Pulp/ tooth volume ratio’s on
CBCT images of mono radicular teeth
AAFS annual meeting Denver (USA), Portable X-ray units
AAFS annual meeting Denver (USA), Bite mark case report
AAFS annual meeting Denver (USA), Computerized facial
reconstruction
Rettsodontologi mote, Oslo (Norway), Age estimation of young
asylum seekers in Belgium
KBGGG meeting, Brussels (Belgium), Bite mark evidence
recognition and its registration protocols: An awareness for all
involving (child) abuse investigations.
MAFS meeting, Antalya (Turkey), Forensic odontological
disciplines highlighted
AAFS meeting, Seattle (USA), Third molar development: differences
between 9 country specific populations
AAFS meeting, Seattle (USA), Measurements of third and preceding
second molar related to age.
Talking Points, Brussels (Belgium), Forensic odontological tasks: A
constant awareness for each dental practitioner
IOFOS meeting, Leuven (Belgium), Age estimation on third molars
development: Comparison of country specific data
International workshop on methods for age estimation in teenagers
and young adult, Oslo (Norway), Age estimation: comparison of
country specifics
AAFS annual meeting, Chicago (USA), Effects of combining
radiological third molar and cervical vertebrae development on
human age estimation
167
Curriculum vitae
2011
2011
2011
2011
2011
2012
2012
2012
2012
2012
2012
2013
AAFS annual meeting, Chicago (USA), Third Molar Development:
Comparison of Nine Tooth Development Scoring and Measuring
Techniques
Vierzehnte treffen der (AGFAD), Berlin (Germany), Dental age
estimation based on third molars development: a Baeysian approach
19th World IAFS Meeting, 9th WPMO Triennial Meeting, 5th MAFS
Meeting, Funchal, (Madeira), Quality assurance in age estimation
Giornate di Odontologia Forense, Florence (Italy. Dental age
estimation based on dental development
Meeting regarding Transition Analysis, Copenhagen (Denmark).
Age Estimation Unaccompanied Young Asylum Seekers: Triple test.
Forensische Krans, Department forensic medicine KU.Leuven
(Belgium), Bite mark evidence, recognition and registration
protocols: An awareness for all potentially involved
AAFS annual meeting, Atlanta (USA), Protocol for a systematic
review of human dental age estimation studies.
AAFS annual meeting, Atlanta (USA), Dental age estimation
combining developmental and morphological age predictors
Fünf Zehntel treffen der AGFAD, Berlin (Germany), Dental age
estimation combining developmental and morphological age
predictors
IOFOS meeting, Leuven (Belgium). Ethics in age estimation of
unaccompanied asylum seeking children
Wetenschappelijke dag NMT, Antwerpen (Belgium). On the border
between forensic odontology and general dentistry
AAFS annual meeting, Washington (USA), Human third molars
development: Comparison of 13 country specific populations.
Moderator Congress section
2009
2010
2012
AAFS annual scientific meeting, Denver (USA)
IOFOS meeting, Leuven (Belgium)
IOFOS meeting, Leuven (Belgium)
Lecturer workshop
2006
2009
2009
2010
2010
2013
Dental age estimation workshop, IOFOS meeting, Leuven
(Belgium).
Identification workshop, KU Leuven (Belgium)
Dental age estimation workshop, MAFS meeting, Antalya (Turkey).
Dental age estimation workshop, Adult part. AAFS annual meeting,
Seattle (USA)
Dental age estimation workshop, IOFOS meeting, Leuven (Belgium)
Dental age estimation workshop, IOFOS meeting, Florence (Italy)
Associate consensus workshop
2010
2010
168
International workshop on methods for age estimation in teenagers
and young adults, Oslo, (Norway)
Unaccompanied Minors: children crossing the external borders of
Curriculum vitae
2010
2012
the EU in search of protection, Brussel (Belgium)
Seminarie leeftijdsbepalingen onbegeleide jonge asielzoekers. FOD
Justitie Directoraat-generaal Wetgeving, Fundamentele Rechten en
Vrijheden Dienst Voogdij, Brussel (Belgium)
Seminar Belgian group “Children on the run”, Brussel (Belgium)
Associate Congress organizing committee
2006
2007
2010
2013
IOFOS Leuven
AFIO Gent
IOFOS Leuven
IOFOS Florence
2006
Thevissen PW, Poelman G, De Cooman M, Puers R, Willems G.
Implantation of an RFID-tag into human molars to reduce hard
forensic identification labor. Part I: working principle.
Forensic Sci Int. 2006, 159 Suppl 1:S33-9.
Thevissen PW, Poelman G, De Cooman M, Puers R, Willems G.
Implantation of an RFID-tag into human molars to reduce hard
forensic identification labor. Part 2: physical properties.
Forensic Sci Int. 2006, 159 Suppl 1:S40-6.
Thevissen PW, Willems G.
Nieuwigheden in de forensische tandheelkunde.
Het Tandheelkundig Jaar 2009, Bohn Stafleu van Loghum, Houten.
Thevissen PW, Pittayapat P, Fieuws S, Willems G.
Estimating age of majority on third molars developmental stages in
young adults from Thailand using a modified scoring technique.
J Forensic Sci, 2009, 54(2):428-32
Thevissen PW, Fieuws S, Willems G.
Human dental age estimation using third molar developmental
stages: does a Bayesian approach outperform regression models to
discriminate between juveniles and adults?
Int J Legal Med. 2009, 124:35-42.
Pittayapat P, Thevissen PW, Fieuws S, Jacobs R, Willems G.
Forensic oral imaging quality of hand-held dental X-ray devices:
comparison of two image receptors and two devices.
Forensic Sci Int. 2010, 194(1-3):20-7.
Thevissen PW, Alqerban A, Asaumi J, Kahveci F, Kaur J, Kim YK,
Pittayapat P, Van Vlierberghe M, Zhang Y, Fieuws S, Willems G.
Human dental age estimation using third molar developmental
stages: Accuracy of age predictions not using country specific
information
Forensic Sci Int. 2010, 201(1-3):106-11.
Thevissen PW, Fieuws S, Willems G.
Human third molars development: Comparison of 9 country specific
populations.
Forensic Sci Int. 2010, 201(1-3):102-5.
Willems G, Thevissen PW, Belmans A, LiversidgeHM,
Willems II. Non-gender-specific dental maturity scores.
Forensic Sci Int. 2010,201(1-3):84-5.
Van Vlierberghe M, Bołtacz-Rzepkowska E, Van Langenhove L,
Łaszkiewicz J, Wyns B, Devlaminck D, Thevissen PW, Boullart L,
Publications
2006
2009
2009
2009
2010
2010
2010
2010
2010
169
Curriculum vitae
2010
2011
2011
2012
2012
2012
2012
2012
2012
2013
2013
2013
170
Willems G.
Dental age estimation on third molars in polish youngsters.
Forensic Sci Int. 2010, 201(1-3):86-94.
Pittayapat P, Oliveira-Santos C, Thevissen PW, Michielsen K,
Bergans N, Willems G, Debruyckere D, Jacobs R..
Image quality assessment and medical physics evaluation of
different portable dental X-ray units.
Forensic Sci Int. 2010, 201(1-3):112-7.
Star H, Thevissen PW, Jacobs R, Fieuws S, Solheim T, Willems G.
Human dental age estimation by calculation of pulp-tooth volume
ratios yielded on clinically acquired cone beam computed
tomography images of monoradicular teeth.
J Forensic Sci. 2011, 56 Suppl 1:S77-82
Thevissen PW, Fieuws S, Willems G.
Third molar development: measurements versus scores as age
predictor.
Arch Oral Biol. 2011, 56(10):1035-40.
Thevissen PW, Kaur J, Willems G.
Human age estimation combining third molar(s) and skeletal
development
Int J Legal Med. 2012;126(2):285-92
Thevissen PW, Galiti D, Willems G.
Human dental age estimation combining third molar(s) development
and tooth morphological age predictors
Int J Legal Med. 2012, 126(6):883-7.
Franco do Rosário Junior A, Couto Souza P, Coudyzer W,
Thevissen PW, Willems G, Jacobs R.
Virtual autopsy in forensic sciences and its applications in the
forensic odontology.
Rev Odonto Cienc 2012;27(1):5-9
Franco A, Thevissen PW, Coudyzer W, Develter W, Van de voorde
W, Oyen R, Vandermeulen D, Jacobs R, Willems G.
Feasibility and validation of virtual autopsy for dental identification
using the Interpol dental codes.
JFLM 2012, in Press, Corrected Proof, Available online 10 October
2012
Thevissen P.W., Kvaal S.I., Dierickx K., Willems G.
Ethics in age estimation of unaccompanied minors.
J Forensic Odontostomatol. 2012, 30 Suppl 1:84-102.
Ramanan N, Thevissen P, Fleuws S, Willems G.
Dental Age Estimation in Japanese Individuals Combining
Permanent Teeth and Third molars.
J Forensic Odontostomatol. 2012, 2(30):34-8.
Thevissen PW, Willems G.
De Triple Test: Het KU Leuven protocol voor leeftijd schattingen op
niet begeleide minderjarige vluchtelingen.
Het Tandheelkundig Jaar 2013. Bohn Stafleu van Loghum. Houten
Franco A, Thevissen PW, Fieuws S, Couto Souzac P, Willems G.
Applicability of Willems model for dental age estimations in
Brazilian children.
Accepted for publication Forensic Sci Int. 2013, Ref.: Ms. No. FSID-12-00959R1
Thevissen PW, Fieuws S, Willems G.
Curriculum vitae
2013
2014
Third molar development: Evaluation of nine tooth development
registration techniques for age estimations.
J Forensic Sci. 2013, Feb.13
Yusof MY, Thevissen PW, Fieuws S, Willems G.
Dental age estimation in Malay children based on all permanent
teeth types.
Int J Legal Med. 2013 Jan 31. Epub ahead of print
Altalie S, Thevissen PW, Fieuws S, Willems G.
Optimal Dental Age Estimation Practice in United Arab
Emirates’Children.
Accepted for publication JFS 03/2014 ref # JOFS-12-693
AAFS: American Academy of Forensic Sciences, IOFOS: International Organisation for
Forensic Odonto-Stomatology, KBGGG: Koninklijke Belgisch Genootschap voor
Gerechtelijke Geneeskunde, MAFS: Mediterranean Academy of Forensic Sciences, AGFAD:
Arbeitsgemeinschaft für Forensische Altersdiagnostik, IAFS: International Association of
Forensic Sciences, WPMO: Association of World Police Medical Officers in Clinical
Forensic Medicine, NMT: Nederlandse Maatschappij tot bevordering der Tandheelkunde
171
Appendix A
BAYESIAN APPROACH.
A multivariate ordinal regression model to obtain the likelihoods
f (x i1,..., x i4 | age i ) :
Formally, let xij denote the j-th third molar, j = 1,…,4, for subject i, with K
possible values, then

 P( xij  k ) 

(C)
log 
   0 k  1kU ij  h(agei )  bi ,
1

P
(
x

k
)


ij


Where  0 k are the K-1 intercept terms used to model the marginal frequencies
in the K ordered categories of the stage. The left-hand side of the equation
represents various logits, i.e., natural logarithms of a specific odds (the odds
of observing a stage lower than a specific value k). Observe that if a
developmental stage would only have two different values (say 1 and 2), the
left-hand side would pertain to a single logit, which yields a binary
regression model. A binary indicator U is valued 1 if the third molar is
located in the upper jaw and 0 elsewhere. The α1k quantify the difference in
stage between upper and lower jaw. The subscript k indicates in the latter
that the effect of jaw is allowed to be non-constant over the intercepts, which
implies that a proportional odds assumption is not made for this effect. A
flexible function h (.) is used to relate age to the logit scale, more
specifically, restricted cubic splines have been used (Harrell, 2001). The key
idea is to allow non-linearity (on the logit scale) in a flexible way without
over fitting the data. Finally, the bi denote the random subject effect,
assumed to be normal distributed. By including this term in (C), each subject
i is allowed to have its own stage level (on logit scale), thereby accounting
for the correlation between the four repeated stage measures. The resulting
model is a generalized linear mixed model, where the term mixed refers to
the simultaneous presence of fixed effects (i.e., age and jaw) and a random
effect (the bi). See, for example, Molenberghs and Verbeke (2005). Due to
the low incidence of stages less than or equal to 5, those stages are combined
into one category. Moreover, no distinction is made between the locations
(left/right) of a stage. As such, a stage pattern ‘8 8 6 7’ pertains to two stages
equal to 8 in the upper jaw and one stage 6 (left or right) and one stage 7
(left or right) in the lower jaw. The generalized linear mixed model is fitted
with the procedure PROC NLMIXED in the SAS 9.1 statistical package
(SAS Institute Inc., Cary, NC, USA), using adaptive Gaussian quadrature.
Once model (C) is fitted on the data, the likelihood
f (x i1,..., x i4 | age i )
173
Appendix A
can be calculated for all possible patterns (xi1,…,xi4) given a specific age.
This has been done in steps of 0.1 years, hence the integral in the
denominator of (B, Chap. 5) is replaced by a sum over age intervals of 0.1
years, and the posterior distribution in (B, Chap. 5) will also have steps of
0.1 years as support points. For the prior distribution, a uniform distribution
has been used, which implies that each age-category within the considered
range (16-22 years old for the comparison of the approaches) is given the
same prior probability.
174
175