Why cancer survival may worsen
(Or, alternative: Explanations for worsening cancer survival)
Esther de Vries,1,2 Henrike E Karim-Kos,1 Maryska LG Janssen-Heijnen,2 Isabelle
Soerjomataram,1,2 Lambertus A Kiemeney,3,4,5 Jan Willem W Coebergh1,2
1
Department of Public Health, Erasmus MC University Medical Centre, Rotterdam;
2
Eindhoven Cancer Registry, Comprehensive Cancer Centre South, Eindhoven;
3
Department of Epidemiology, Biostatistics and Health Technology Assessment,
Radboud University Nijmegen Medical Centre, Nijmegen;
4
Department of Urology, Radboud University Nijmegen Medical Centre, Nijmegen;
5
Comprehensive Cancer Centre East, Nijmegen
Correspondence:
Dr Esther de Vries
Department of Public Health, Erasmus MC
P.O. Box 2040, 3000 CA Rotterdam
The Netherlands
Tel: +31 10 704 3730
Fax: +31 10 704 8475
E-mail: e.devries@erasmusmc.nl
1
Abstract
When cancer survival is reported to be worsening over time or inferior compared to
other countries, politicians and healthcare workers may get blamed for presumed
suboptimal care. Yet, there is a variety of reasons for cancer survival statistics to
change for the worse, deterioration of (access to) care being only one. Other
explanations include improved diagnosis of premalignant lesions, causing only the
most aggressive cancers, with a worse prognosis, to remain. Deleterious changes in
the distribution of prognostic factors, and changes in the distribution of sociodemographic characteristics may also negatively affect survival proportions. In this
overview article, we identify the pitfalls of comparisons of published population-based
survival data from different time periods or populations.
2
Introduction
Cancer survival statistics attract a lot of attention, particularly when comparisons
between calendar periods or countries show that survival decreases over time or is
lower than expected based on the average of surrounding countries. Populationbased cancer survival proportions generally tend to remain stable or increase over
time in most industrialized countries and for most cancer types.1,2 Such increases in
survival proportions, however, do not necessarily reflect true improvements in cancer
treatment. For example, early detection and screening practices have artificial effects
on survival due to lead time or length bias (Box 1).3,4 Decreasing survival proportions
can sometimes be observed over time. Such deteriorating survival proportions have
a variety of causes, even after adjustment for age and all-cause mortality.1,5 We
briefly explain the principles of cancer survival and illustrate that a decrease in
survival proportion can be attributed to four factors: deterioration of (access to) care,
improved diagnosis of premalignant lesions, deleterious changes in the distribution of
prognostic factors, and changes in the distribution of socio-demographic
characteristics that negatively affect survival proportions. [AU: Please confirm edit OK] In this Perspectives article, we identify possible pitfalls when comparing
published survival data from different time periods or in different populations.
Short explanation of cancer survival
Cancer survival is estimated for cohorts of newly diagnosed patients, based on
cytological or histological criteria and sometimes clinical criteria of patients when
entered into a cohort study. Cancer patient survival is measured as the time from
cancer diagnosis until death; a 5-year follow-up period is most frequently used as an
indicator of outcome, although survival at 10 years would be more suitable for many
cancer types, such as those amenable to screening, for example breast and prostate
cancer. Endpoints for calculating survival can vary. Death due to any cause allows
the overall (all-cause) survival proportion to be calculated, that is, the proportion of
3
cancer patients alive at a certain point in time after diagnosis. Death due to the
cancer under study or its treatment is reflected in the disease-specific survival
proportion.
Many practical problems are encountered in correctly determining and
registering cause of death. For example, it can often be difficult to determine the
proper underlying cause of death.3 Relative survival circumvents the need to
determine the cause of death, because it represents the ratio of the overall survival
proportion for a cohort of cancer patients and the expected overall survival proportion
for the general population with the same sex and age distribution as the cancer
patient cohort. Relative survival therefore measures mortality associated directly
and/or indirectly with the diagnosis of a cancer and thus includes deaths due to
complications of cancer treatment. In order for relative survival to be interpreted as a
measure of excess mortality among cancer patients it is important to have accurately
estimated expected mortality data for the general population. It is assumed that
cancer mortality and non-cancer mortality are independent. Moreover, in theory,
relative survival is dependent on trends in other causes of death in the general
population. These assumptions, however, should sometimes be questioned. For
example, a markedly decreased incidence and mortality from cardiovascular
diseases, which results in an increase in the expected overall survival proportion for
the general population, would lead to a decrease in the relative survival ratio of
cancer, even when the observed disease-specific survival proportion remains stable.
Also, it is assumed that cancer patients would have had an average life expectancy if
they had not been diagnosed with cancer. This is of course an arbitrary assumption
for some types of cancer. For example, patients with lung cancer may have a lower
life expectancy because the cause of their cancer (smoking) is also a strong risk
factor for mortality from other diseases.
Calculation of relative survival ratios (also referred to as relative survival
rates) requires mortality data by sex, age, and calendar year from the general
4
population.3 Relative survival ratios can be compared over time or between
geographical regions, despite the ageing of populations, but the results are
meaningful only when they are age-adjusted. The most reliable survival comparisons
are based on age-specific risks in case of differential periods of observation.6 The
relative survival ratio is very useful for specific cancer (sub)sites, but it is not suitable
when it is calculated for all cancers combined because all cancers combined do not
represent a small proportion of deaths due to all causes of death combined. The
"expected" survival figures for relative survival calculations in this scenario are based
on competing risk mortality data that are too high and so the estimates become less
valid. Moreover, survival statistics for all-sites should generally not be compared
directly because the case-mixes can differ greatly between different populations and
time periods.
In order to judge the quality of survival estimates given by a registry, the
inclusion and exclusion rules of the registry database should be clear. If these rules
are changed, for example, in coding systems such as the ICD-O international
classification of diseases or TNM, survival estimates might change rapidly. When
previously non-invasive ‘cancers’ were included as ‘new’ cancer cases (for example,
the change of classification of bladder papillomas into papillocarcinomas in 1978),
survival estimates increased markedly.7 Moreover, completeness of study follow-up
is essential for accurate survival estimates. Unfortunately, administrative
completeness can vary with time across and between countries. Survival proportions
tend to become lower when the completeness of follow-up improves, because
patients who were initially lost to follow-up usually turn out to have died. The number
of death certificate only (DCO) cases in cancer registries depends on the quality of
the registry and access to death certificates. It is common practice to exclude these
cases for survival analysis because the date of diagnosis (and hence survival time) of
DCO cases is unknown.8 Registries with a high proportion of DCO cases will
5
overestimate their survival. Therefore, when the proportion of DCO cases decreases
over time survival estimates may seem to worsen.9
Deterioration of access to care
The most obvious, but uncommon, reason for decreasing survival proportions is less
aggressive or substandard cancer care, resulting in failing early detection or less
effective treatment. For example, decreasing relative survival ratios for patients with
laryngeal cancer were observed in the mid-1990s in the US compared with the
1980s.10 [During the mid-1990s, many clinicians preferred irradiation above
laryngectomies. Detailed analyses over time revealed shorter survival for patients
with laryngeal squamous-cell carcinoma following non-surgical treatment.11
Likewise, a comparison of patients with high-grade T1 bladder cancer who
underwent radical surgery in the US showed a 5-year disease-free survival of 70%
for patients who had radical surgery before 1998 versus only 40% after 1998. During
the 1990s, intravesical therapy (for example, immunotherapy and/or chemotherapy)
facilitated bladder-sparing strategies for these patients. Before 1998, 74% of these
patients underwent radical surgery without prior intravesical therapy versus 43% after
1998. The observed decrease in survival was attributed to delayed radical surgery
after the increased use of intravesical therapy.12
The economic collapse of the former socialistic countries of Central Europe
coincided with decreasing survival proportions during the transition period; survival
proportions of ovarian, cervical and uterine cancer, childhood soft-tissue sarcomas,
and hepatic and germ-cell tumours temporarily decreased between 1988–1992 and
1993–1997.13,14 These temporary decreases were likely due to disintegrating health
care systems and infrastructures.
When cancers are detected at a later stage or mistakenly staged as a lowstage cancer because of deteriorating screening or staging availability (for example,
due to poor imaging capacities), treatment will be less effective and survival
6
proportions will decrease. It is possible to adjust for stage at diagnosis,15 although
improvements in staging methods and changes in stage-coding could still be
problematic when comparing stage-adjusted survival estimates over time.
Improved diagnosis of precancers
Survival proportions may decrease while therapeutic options remain stable or
improve over time and/or early detection remains unchanged or improve. This
occurred for cervical cancer in many European countries, where large-scale and high
quality population-based programs for cervical screening gradually became available
(Table 1).1 The same phenomenon might be observed in the future for colorectal
cancer.
The aim of screening for cervical and colorectal cancers is not only the
detection of cancers in early stages but also the detection of pre-malignant lesions,
thereby preventing the development of ‘invasive’ cancer. However, the cancers that
still occur, despite screening, may consist of a selected group of rapidly growing,
aggressive tumors that may be more difficult to treat and thus might result in
decreasing survival proportions, preceded by a lower incidence.
Screening for most other types of cancer (for example, breast or prostate) will
detect early stages of cancer rather than pre-malignant lesions. Survival increases
due to lead time bias and even length bias (Box 1), and possibly as a result of moreeffective treatment. Survival proportions, however, may decrease again when the
awareness or willingness to participate in screening programs decreases, possibly
leading to higher stage disease at diagnosis. If the date of death is not postponed by
treatment this will result in worsening survival proportions, that is, an inverse ‘lead
time’ effect.
New and improved methods for diagnosing cancer frequently result in
improved survival proportions because these methods are more precise and/or
detect the cancers earlier, hopefully resulting in better therapeutic options.
7
Conversely, the introduction of a new diagnostic technique may temporarily decrease
cancer survival proportions. Pancreatic cancer is difficult to diagnose and treat and is
associated with a poor prognosis. New diagnostic technologies have resulted in more
diagnoses during the patients’ lifetime, whereas previously the diagnosis would have
been made at autopsy and recorded as a DCO. These tumors with a very bad
prognosis will now be included in the survival statistics resulting in lower survival
estimates.3
Changes for the worse in the distribution of prognostic factors
Subtypes and subsites
Changes in risk factor exposure may lead to shifts in the distribution of cancer
subtypes. In the Netherlands, relative survival ratios for adenocarcinomas of the lung
decreased during the 1980s despite increased application of better endoscopic
techniques by more lung physicians accompanied by better access to specialized
care. This decrease in relative survival ratios was partly attributed to the termination
of mass screening for tuberculosis in the early 1980s, which sometimes detected
slow-growing peripheral adenocarcinomas. The higher concentration of carcinogens
in the peripheral lung zone as a result of the increased use of filter cigarettes and
deep inhalation may also have caused tumors to more metastasize rapidly.16 Similar
decreases in overall lung cancer survival were observed in Malta, with stable overall
incidence and mortality rates but a relative rise in the incidence rate of
adenocarcinomas.1,17
Shifts in cancer subtype and subsite distribution may need to be studied over
time as a determinant of survival. This is illustrated by a study of changes in
incidence and survival of gastric cancer in the southeastern part of the Netherlands.
Despite marked improvements in endoscopic early detection, staging, surgery, and
peri-operative mortality, no improvement in survival occurred during the period 1982
to 1995 because of the increasing proportion of cardiac and diffuse cancers with a
8
poor prognosis.18 Another example of the negative influence of shifts in cancer
subsite on survival is laryngeal cancers.8 Laryngeal cancers of the glottis exhibit a 5year relative survival of around 60–80%, and supraglottic cancers around 40%.19
Besides earlier detection of cancer of the glottis, alcohol consumption and tobacco
smoking have different etiological effects on tumor subsite, with alcohol being more
important for supraglottic tumors and smoking for glottic cancers. With the
decreasing smoking prevalence and stable alcohol consumption rates in many
European countries, the proportion of cancers of the poor prognostic subsite
(supraglottis) will increase and will negatively affect the relative survival ratios of
laryngeal cancer over time.
Co-morbidity
Changes in prevalence of risk factors can also affect overall survival because of the
presence of concomitant diseases (co-morbidity). If a concomitant (co-morbid)
condition becomes more prevalent in the population as a whole, the relative survival
ratio is corrected for this change. However, relative survival ratios are likely to be
affected when the co-morbid condition is strongly associated with the tumor (for
example, COPD in lung or laryngeal cancer patients) or when the co-morbid
condition is not only life-threatening in itself, but also influences the eligibility of
patients to receive aggressive cancer treatments such as surgery or chemotherapy,
as is the case for example for diabetes.20
Changes in (behavioural) risk factors, such as an increase in alcohol
consumption, cause on the long term rises in alcohol-related cancer incidence,
mortality and co-morbidity, such as ischemic heart disease, stroke, hypertension,
diabetes, liver cirrhosis and depression. An increased prevalence of such co-morbid
conditions will cause lower survival proportions.21 Since alcohol consumption is an
important risk factor, particularly for squamous-cell cancer of the esophagus and
supraglottic cancer of the larynx, the above-mentioned co-morbid conditions are
9
likely to be more prevalent among esophageal and laryngeal cancer patients
resulting in less optimal treatment and more complications. This phenomenon may
also have contributed to the observed declines in survival of these cancers during the
transition period in the former socialist countries in central Europe.1,22,23
In a similar manner, the increased prevalence of infectious diseases related
to cancer, such as the human immunodeficiency virus and the hepatitis B and C
viruses,24 may thus not only cause a higher incidence of tumors such as lymphomas
and liver cancer, but may also have a negative effect on survival of patients with
these cancers.
Changes in socio-demographic characteristics
When patient populations change rapidly, survival proportions can also be affected;
for example, when certain subgroups comprise a larger part of the cancer burden
due to changes in the distribution of socio-economic status. Cancer survival
proportions among people of low socio-economic classes are generally lower than
those of the mid or higher socio-economic classes, with relative risks of death within
5 years of diagnosis in the most deprived groups being 1.3–1.5-fold higher compared
with the most affluent group.25,26 Underlying causes of the worse prognosis are
related to lower awareness, unfavourable tumor characteristics (for example, later
stage at diagnosis due to diagnostic delay), patient characteristics such as race,
screening participation rates, psychosocial factors, co-morbidity or health care factors
including treatment, screening, and quality of medical care.26 Changes in the
distribution of socio-economic classes, for example by selective emigration of the
more healthy and enterprising, may cause decreasing survival proportions.
Conclusions
In conclusion, a variety of factors can cause decreasing cancer survival proportions,
most of which do not necessarily represent deteriorating care. Unfavourable changes
10
in underlying risk factors, and early detection or screening practices that lead to
detection of relatively less aggressive lesions are often the cause of this
phenomenon as well as changing demographic profiles of the populations.
Worsening survival proportions merit in-depth investigation, taking into account
preceding trends in incidence, particularly for tumor subtype or subsite.
Acknowledgement
The work on this research was performed within the framework of the project
‘Progress against cancer in the Netherlands since the 1970s?’ (Dutch Cancer Society
grant 715401) and co-funded by the Comprehensive Cancer Centre South in
Eindhoven, the Netherlands.
11
References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
Karim-Kos, H.E. et al. Recent trends of cancer in Europe: a combined
approach of incidence, survival and mortality for 17 cancer sites since the
1990s. Eur. J. Cancer 44, 1345–1389 (2008).
Jemal, A. et al. Cancer statistics, 2007. CA Cancer J Clin 57, 43-66 (2007).
Dickman, P.W. & Adami, H.O. Interpreting trends in cancer patient survival. J.
Intern. Med. 260, 103–117 (2006).
Welch, H.G., Schwartz, L.M. & Woloshin, S. Are increasing 5-year survival
rates evidence of success against cancer? JAMA 283, 2975–2978 (2000).
Brenner, H., Gondos, A. & Arndt, V. Recent major progress in long-term
cancer patient survival disclosed by modeled period analysis. J. Clin. Oncol.
25, 3274–3280 (2007).
Pokhrel, A. & Hakulinen, T. How to interpret the relative survival ratios of
cancer patients. Eur. J. Cancer (2008).
Kiemeney, L.A. et al. Bladder cancer incidence and survival in the southeastern part of The Netherlands, 1975-1989. Eur. J. Cancer 30A, 1134–1137
(1994).
Berrino, F. The EUROCARE Study: strengths, limitations and perspectives of
population-based, comparative survival studies. Ann. Oncol. 14 Suppl 5, v9–
13 (2003).
Robinson, D., Sankila, R., Hakulinen, T. & Moller, H. Interpreting international
comparisons of cancer survival: the effects of incomplete registration and the
presence of death certificate only cases on survival estimates. Eur. J. Cancer
43, 909–913 (2007).
Induction chemotherapy plus radiation compared with surgery plus radiation
in patients with advanced laryngeal cancer. The Department of Veterans
Affairs Laryngeal Cancer Study Group. N. Engl. J. Med. 324, 1685–1690
(1991).
Hoffman, H.T. et al. Laryngeal cancer in the United States: changes in
demographics, patterns of care, and survival. Laryngoscope 116, 1–13
(2006).
Lambert, E.H. et al. The increasing use of intravesical therapies for stage T1
bladder cancer coincides with decreasing survival after cystectomy. BJU Int
100, 33–36 (2007).
Coebergh, J.W. et al. Leukaemia incidence and survival in children and
adolescents in Europe during 1978-1997. Report from the Automated
Childhood Cancer Information System project. Eur. J. Cancer 42, 2019–2036
(2006).
Aareleid, T. & Brenner, H. Trends in cancer patient survival in Estonia before
and after the transition from a Soviet republic to an open-market economy.
Int. J. Cancer 102, 45–50 (2002).
Ciccolallo, L. et al. Survival differences between European and US patients
with colorectal cancer: role of stage at diagnosis and surgery. Gut 54, 268–
273 (2005).
Janssen-Heijnen, M.L. & Coebergh, J.W. Trends in incidence and prognosis
of the histological subtypes of lung cancer in North America, Australia, New
Zealand and Europe. Lung Cancer 31, 123–137 (2001).
Micallef, R. Incidence of lung cancer in Malta for the years 1992-2005 with all
morphology subtypes. Personal communication. [AU: Is this an
unpublished reference? Is so, this does not get cited in the reference
list but as unpublished data within the text and includes the author’s
initial and surname] – OK – yes this is a personal communication – not
published
12
18.
19.
20.
21.
22.
23.
24.
25.
26.
Pinheiro, P.S., van der Heijden, L.H. & Coebergh, J.W. Unchanged survival of
gastric cancer in the southeastern Netherlands since 1982: result of
differential trends in incidence according to Lauren type and subsite. Int. J.
Cancer 84, 28–32 (1999).
Berrino, F. & Gatta, G. Variation in survival of patients with head and neck
cancer in Europe by the site of origin of the tumours. Eur. J. Cancer 34,
2154–2161 (1998).
van de Poll-Franse, L.V. et al. Less aggressive treatment and worse overall
survival in cancer patients with diabetes: a large population based analysis.
Int. J. Cancer 120, 1986–1992 (2007).
Janssen-Heijnen, M.L. et al. Prognostic impact of increasing age and comorbidity in cancer patients: a population-based approach. Crit. Rev. Oncol.
Hematol 55, 231–240 (2005).
Zaborskis, A., Sumskas, L., Maser, M. & Pudule, I. Trends in drinking habits
among adolescents in the Baltic countries over the period of transition: HBSC
survey results, 1993–2002. BMC Public Health 6, 67 (2006).
Zatonski, W. & Jha, P. (Cancer Center and Institute of Oncology, Warsaw,
2000). [AU: Please can you specify if this is a conference and provide
the date and details] No this is a report from a project. The title somehow
disappeared from the reference list and I did not notice. My apologies, the
title of the report is: The Health Transformation in Eastern Europe after
1990: A Second Look and the report can be found through this link:
http://www.hem.home.pl/files/The_Health_Transformation_in_Eastern_Europ
e_after_1990_-_A_Second_Look.pdf
Parkin, D.M. The global health burden of infection-associated cancers in the
year 2002. Int. J. Cancer 118, 3030–3044 (2006).
Coleman, M.P., Babb, P., Sloggett, A., Quinn, M. & De Stavola, B.
Socioeconomic inequalities in cancer survival in England and Wales. Cancer
91, 208–216 (2001).
Woods, L.M., Rachet, B. & Coleman, M.P. Origins of socio-economic
inequalities in cancer survival: a review. Ann. Oncol. 17, 5–19 (2006).
13
Table 1 Relative survival for cervical cancer in Europe
Start of
Type of screening
screening
invitation
Region
program
Austria (Tirol)
Iceland
Finland
Italy
Netherlands (3
regions)
Scotland
Norway
Sweden
Switzerland
Czech republic
England
Germany
(Saarland)
1970
1964
1963
1982–1998
1980
PB/OP (regional)
NRS
PB
PB/OP (regional)
PB
1988
1995, pilot
1992
1967–1977
No data
1966
1988
1971
NRS
NRS
No data
No data [AU:
okay? – yes]
No data [AU:
okay? - yes]
NRS
Malta
No data
Poland
Slovenia
Spain
Wales
2003 (1995
opportunistic)
1986 (regional)
1988
NRS
OP
OP
NRS
OP
PB/OP
NRS
5-year relative survival
ratios
1990–
1994a
63.6
68.6
66
66.6
69.4
2000–
2002b, c, d
64.2
70.6
65.8
67
69.2
60.6
61
69
69.6
68.7
65.2
63.8
63.5
67.5
66.7
66.8
59.8
58.6
55.5
64.4
46.5
48.2
56
59.9
68.7
58.7
65.2
60.4
52.6
Trends in
survival
Increased
Increased
No change
No change
No change
No change
Decreased
Decreased
Decreased
Large decrease
Large decrease
Large decrease
Large decrease
Large increase
Large increase
Large decrease
Large decrease
PB: population-based: invitational programs
NRS: only invitation to women that did not recently have an (opportunistic) smear
OP: opportunistic only
=: <0.5 percent-points difference, single arrow: 0.5 – 5 percent-points difference, double arrows: >5% point
difference
aSant M, et al. EUROCARE-3: survival of cancer patients diagnosed 1990-94-results and commentary. Ann
Oncol. 2003;14 Suppl 5:V61-V118.
bVerdecchia A, et al. Recent cancer survival in Europe: a 2000-02 period analysis of EUROCARE-4 data. Lancet
Oncol. 2007;8:784-96.
c Linos A, Riza E. Comparisons of cervical cancer screening programmes in the European Union. Eur J Cancer.
2000;36:2260-65.
d Anttila A, et al. Cervical cancer screening programmes and policies in 18 European countries. Br J Cancer.
2004;91:935-41. [AU: Please clarify what these two references pertain to – hope this is better now?]
Box 1. Definitions of lead time and length bias.Lead time: early detection or screening
aims to diagnose a disease in an earlier stage than it would be without screening. When the
moment of death is not postponed by screening, and therefore no additional life-time has
been gained, the survival time since diagnosis is longer for a screened person compared to
unscreened people. In this case, screening appears to increase survival time, and this gain is
called lead time.
Length bias: slow growing tumours often have better prognosis than tumours with high growth
rates. Early detection or screening is more likely to detect slow growing tumours as they have
a longer time-period of existing without clinical symptoms, some of them might actually never
have caused clinical disease. In such cases, screen-detected tumours appear to be
associated with better survival statistics, because they actually represent a group of tumours
which inherently already had a better prognosis. This effect is called length bias.
14