Summarizing our knowledge of
normal tissue tolerances:
the progress and future directions
of QUANTEC*.
Purpose of QUANTEC
• Both AAPM and ASTRO recognized ($$):
• Need for a systematic overhaul of our
understanding of normal tissue tolerances
ANDREW JACKSON
Memorial Sloan-Kettering Cancer Center
• For use in clinical treatment planning and
optimization
*QUAntitative Normal TissuE models in the Clinic
History of QUANTEC
• 2006 AAPM Science Council
– Ellen Yorke and Rock Mackie
– Steering Committee: Deasy, Bentzen, Yorke, Ten-Haken,
Jackson, Marks, Eisbruch, Constein
• 2007 1st QUANTEC meeting in Madison Wisconsin
– Initial review of tolerances of involving physicists, biostatisticians and physicians.
• 2008 Special Issue of Red Journal
– in preparation, to be published in the fall.
Organ specific QUANTEC articles
• Significance of injury
• Clinically relevant endpoints
– Time course
– Ambiguities
• Anatomic definitions
– Variations in contouring practice
• Review of literature on dose-volume
dependence of endpoints
– Level of evidence
1
Organ specific QUANTEC articles
• Patient related risk factors
• Models of the data
– Limitations
• Caveats concerning models
• Dose-volume and model dependent limits for
clinical use
• Future studies –
• Additional knowledge required to improve toxicity prediction
• Endpoint scoring and data capture in future studies
Overview of Results for Specific
Organs
• Great variety in quality and quantity of data
• Consistency issues:
– Endpoints studied
– Physical issues (organ motion, contouring issues,
dosimetry)
– Models/D-V constraints reported
• Some organs are more equal than others
– Lung: ~70 studies
– Ear: ~4 studies of dose response of late hearing loss
Organs Included
• Brain, Brainstem, Cord, Optic Structures, Ear
• Salivary Glands, Larynx & Pharynx
• Lung, Heart, Esophagus
• Liver, Kidney, Small Bowel,
• Rectum, Bladder, Penile Bulb
Mean dose response of pneumonitis
(A. Jackson with L. Marks/S. Kong/J.Deasy/J.Bradley/M. Martel/S. Bentzen)
• Patients treated for NSCLC
– Data from 9 institutions, 10 separate studies
• 1,167 patients with 222 cases of
pneumonitis
• Grade 3 RTOG ~ Grade 2 SWOG
– (requiring steroids)
– accepted grade 1 definition if few grade 1
cases
2
Pneumonitis, mean dose response - whole lung
Mean dose response of
pneumonitis
– Numbers of pts w./w.o. pneumonitis
– Bin locations on quartile plots
• Fit of logistic function:
– D50 = 30.75 [29.9 – 31.7] Gy
– 50 = 0.907 [0.836 – 0.987]
Probability of pneumonitis
• Reporting rate (and S.D.) of pneumonitis
as function of mean dose to total lung
1.0
MSKCC (10/78)
Duke (39/201)
Michigan (17/109)
MD Anderson (~49?/223)
NKI (17/106)
WU (52/219)
Martel et al. (9/42)
Oeztel et al. (10/66)
Rancati et al. (7/55)
Kim et al. (12/68)
logistic fit
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0
10
20
30
Mean dose (Gy)
Probability of aspiration as function of mean dose to Glottic and Supra-Glottic Larynx
Rectal dose volume limits
Jackson/Deasy/Gay/Michalski/Tucker/Zelefsky
– Published limits having sig. correlation with
grade 2 rectal bleeding
– Color coded to indicate prescription dose
• Blue = 66-70 Gy
• Red = 83 Gy (LQ equivalent dose in 2 Gy fr)
– Thickness of line indicates overall
complication rate in study
(T. Rancati, M. Schwartz, A. Allen, F. Feng, A. Popovtzer, B. Mittal, and A. Eisbruch)
3
Dose-volume limits
with LQ corrected doses (a/b = 3 Gy)
Jackson 70.2
Jackson 75.6
Akimoto
Wachter
Koper
Cozarrino
Zapatero
70-75.6 Gy: 7%
Huang
Hartford
75.6 Gy: 34%
Grade 1
Fiorino
Boersma
100
66 Gy: 33%
80
66 Gy: 14%
75.6 Gy: 19%
60
70-76 Gy: 9%
69 Gy: 25%
3 Gy/fr
40
– 4 published studies fitting LKB model to rectal
complication data
• Forrest plot of “n” values (n=1/a)
• 1 S.D. indicated
• Average value of “n” for all studies indicated
– One further study* of RTOG 94-06 included
• n=0.08 (95% conf: 0.04-0.26)
20
74-78 Gy: 23%
70.2 Gy: 6%
70-78 Gy: 3%Grade 3
0
10
20
30
40
50
60
70
80
LQ equivalent dose in 2 Gy fractions (Gy)
estimates of LKB volume effect parameter n
for rectal complications
90
* Tucker ASTRO 2007, IJROBP
Kidney
(L. Dawson, A. Paulino, B. Kavanagh, C. Pan, S. K. Das,
M. Miften, X. A. Li, R. K. Ten Haken, T. E. Schultheiss)
RTOG 94-06
Tucker
Study
% volume
66.2-70.2Gy: 11%
Rectal lyman model fits
Rancati
Peeters
Sohn
0.01
0.1
volume effect parameter, n
1
From: Cassady JR. Clinical radiation nephropathy. IJROBP 1995;31:1249-1256.
4
Bilateral Whole Kidney RT - TBI
0.5
Bilateral Partial Kidney RT
minimal risk
C he ng e t a l . Fi g ur e 2
low risk
moderate-high risk
high risk
undefined
Model 3
peptic ulcer
gastric cancer
seminoma
lymphoma
Model 4
0.4
Adults + Mixed
Adults + Mixed: no NT drugs
estimated
threshold
for injury
Percent Volume
Incidence of Late Renal Toxicity
(% x 100)
Data from 916 patients
0.3
0.2
0.1
gastric cancer
lymphoma
0
4
6
8
10
12
14
gynecological cancer
16
Equivalent Total Dose (Gy)
Cheng J, Schultheiss T, Wong J. Impact of Drug Therapy, Radiation
Dose and Dose Rate on Renal Toxicity following Bone Marrow Transplantation.
International Journal of Radiation Biology 2008; In press.
Dose, Gy, over 5 weeks
Mean dose response for SNHL at 4 kHz
Late Hearing Loss
1.0
(A. Jackson, N. Bandare, W. Mendenhall)
Chen et al.
Oh et al.
Honore et al.
Pan et al.
Kwong et al.
– (post-treatment vs pre-treatment)
• Differences in way endpoint is defined
– Ispi- relative to contra-lateral hearing loss vs hearing
loss
• Dose reconstruction
– 1 study, doses reconstructed with surrogate CT scans
– 1 study, ipsi- doses relative to contra-lateral
Incidence of SNHL at 4 kHz
0.9
• Hearing loss tests from 3 studies as function of
mean cochlea dose
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0
20
40
60
80
Mean cochlea dose (Gy)
5
Necessity of combining data sets
• Number of complications in any given treatment
series is usually low
– False negatives
– No statistical power to determine model parameters
• Dose-volume exposures correlated in individual
series
– Introduces phony correlations with complications
(False positives)
– Insufficient range of dose-volume combinations to
determine model parameters
Problems in synthesizing data
Problems in synthesizing data
• Endpoint definitions:
– Need to be clinically relevant
– Need to be specific
• Rectal bleeding or incontinence vs grade 2 RTOG
toxicity
– Different comps. have different dose-volume effects
– Need to be standardized
Problems in synthesizing data
• Standard of reporting is POOR
• Variety of dose volume limits proposed
– These cannot be combined
• Variety of models may be fit
– Responses cannot be combined
• gEUD responses with different “a” values cannot
be combined
– Lack of basic statistics (numbers not stated!)
• Schultheiss 1994: “The information in this report
would be of greater clinical use if some
indication had been provided of the total
number of patients from which the myelopathy
cases were drawn”
– Locations of bins in e.g. quartile plots not
given
– Model parameters (and errors) not be stated
6
Example: EUD Atlas of
Pneumonitis
In other words:
• Report the numbers of patients with
complications and the number treated
• Report the number of NSCLC patients
whose gEUD exceeds a given level
– Elementary statistics increase clinical utility
• Be comprehensive
– Both with and without complication
– Report as much about the data as possible
• Be comprehensive:
• How far can we take this?
– Do this for each gEUD value
– Vary the n=1/a parameter
EUD Atla s o f severe p ne umoni ti s a t 6 mo nths
0
1
0
0 .8
0
Lower 68% confidence limit on
complication probability
1 0/78
10 /60
7/31
4/14
3/6
1 /2
1 /1
0 /0
0 /0
0 /0
0 /0
0 /0
0 /0
0 /0
0/0
0/0
0/0
1 0/78
10 /62
7/33
4/17
3/7
1 /2
1 /1
0 /0
0 /0
0 /0
0 /0
0 /0
0 /0
0 /0
0/0
0/0
0/0
1 0/78
10 /62
7/39
4/17
3/7
1 /2
1 /1
0 /0
0 /0
0 /0
0 /0
0 /0
0 /0
0 /0
0/0
0/0
0/0
1 0/78
10 /64
9/43
5/21
3/9
1 /3
1 /1
1 /1
0 /0
0 /0
0 /0
0 /0
0 /0
0 /0
0/0
0/0
0/0
1 0/78
10 /68
9/49
6/26
3/11
2 /4
1 /2
1 /1
0 /0
0 /0
0 /0
0 /0
0 /0
0 /0
0/0
0/0
0/0
1 0/78
10 /69
10 /5 5
6/31
4/15
3 /7
1 /2
1 /1
1 /1
0 /0
0 /0
0 /0
0 /0
0 /0
0/0
0/0
0/0
1 0/78
10 /74
10 /5 9
9/42
5/22
3 /9
1 /3
1 /1
1 /1
0 /0
0 /0
0 /0
0 /0
0 /0
0/0
0/0
0/0
0.7
1 0/78
10 /76
10 /6 2
9/50
6/36
4 /16
2 /5
1 /1
1 /1
1 /1
0 /0
0 /0
0 /0
0 /0
0/0
0/0
0/0
0.6
1 0/78
10 /76
10 /7 0
10 /5 9
9/45
6 /33
3 /1 5
1 /4
1 /1
1 /1
1 /1
0 /0
0 /0
0 /0
0/0
0/0
0/0
0.5
1 0/78
10 /76
10 /6 8
9/57
9/47
7 /35
4 /1 8
1 /3
1 /1
1 /1
1 /1
0 /0
0 /0
0 /0
0/0
0/0
0/0
0.4
1 0/78
10 /76
10 /7 6
10 /7 1
9/59
9 /49
9 /4 0
5 /2 6
1 /6
1 /2
1 /1
1 /1
0 /0
0 /0
0/0
0/0
0/0
0.3
1 0/78
10 /77
10 /7 6
10 /7 4
10/6 7
9 /57
9 /4 8
7 /3 6
3 /2 1
1 /5
1 /3
1 /2
1 /1
0 /0
0/0
0/0
0/0
0.2
1 0/78
10 /76
10 /7 5
10 /7 4
9/68
9 /59
9 /4 8
7 /3 5
3 /2 1
1 /8
1 /5
1 /2
1 /2
0 /0
0/0
0/0
0/0
1 0/78
10 /77
10 /7 6
10 /7 5
9/70
9 /63
9 /5 0
7 /4 0
4 /2 5
3 /15
1 /6
1 /5
1 /2
0 /1
0/0
0/0
0/0
1 0/78
10 /77
10 /7 6
10 /7 5
9/74
9 /68
9 /5 8
7 /4 4
5 /3 1
3 /23
2 /1 4
1 /7
1 /5
0 /1
0/1
0/0
0/0
1 0/78
10 /77
10 /7 7
10 /7 6
9/73
9 /71
9 /6 6
7 /5 2
6 /3 9
3 /28
3 /2 4
2 /14
1 /7
0 /4
0/2
0/1
0/0
1 0/78
10 /77
10 /7 7
10 /7 4
9/71
9 /70
7 /6 1
6 /5 0
5 /3 7
3 /28
2 /2 4
1 /14
0 /8
0 /4
0/2
0/1
0/0
1 0/78
10 /76
10 /7 5
9/73
9/71
7 /64
6 /5 6
5 /4 1
3 /3 4
2 /30
2 /2 4
0 /16
0 /8
0 /4
0/2
0/0
0/0
1 0/78
10 /77
10 /7 5
9/73
9/73
8 /67
7 /5 9
5 /4 8
5 /3 9
2 /32
1 /2 8
1 /21
0 /1 3
0 /8
0/2
0/2
0/0
1 0/78
10 /77
10 /7 7
10 /7 4
9/73
9 /72
8 /6 7
7 /6 0
5 /4 9
5 /43
2 /3 4
1 /30
1 /2 6
0 /1 9
0/1 2
0/4
0/2
-0.5
1 0/78
10 /77
10 /7 6
10 /7 5
9/72
9 /72
8 /6 8
7 /5 9
5 /4 9
3 /41
2 /3 7
1 /30
1 /2 7
0 /2 2
0/1 6
0/8
0/1
-0.6
Lower 68% confidence limit on complication probability
1.0
0
0 .6
0
0
0 .4
0
0
2
4
log10(n)
2
0 .2
8
0
-0 .2
12
16
-0 .4
20
26
-0 .6
32
36
-0 .8
38
42
-1
0.9
0.8
log10 n
0
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
0.1
0.0
-0.1
-0.2
-0.3
-0.4
-0.7
dEUD (Gy)
0
4
8
12
16
E UD - dEUD (Gy)
20
24
30
-0.8
-0.9
-1.0
0
10
20
30
40
50
60
70
80
EUD (Gy)
7
Comparison of results from atlases with 2-0.5 Gy
grid spacings with those from exact EUD
For range of significant p-value,
2 Gy grid spacing in EUD is adequate
p value vs n, 2 Gy binning and no binning in EUD
Conclusions
• QUANTEC is:
For lower 95% and 68% conf. int. on n
2 and 1 Gy spacing resp. are adequate
– Updating our clinical understanding of normal tissue
tolerances
– Providing clinical guidelines where possible
Residual Deviance for EUD from Atlas (comparing various spacings)
1
p value
anti-correlation
0.01
p value vs n (eud bins-2Gy)
p=0.05
p value vs n (exact EUD per patient)
0.001
-1.0-0.9-0.8-0.7-0.6-0.5-0.4-0.3-0.2-0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
log10 n
60
EUD in 0.5 Gy bins
Exact EUD
EUD in 2 Gy Bins
EUD in 1 Gy bins
59
• With appropriate caveats
58
– Defining areas of our ignorance
57
56
• recommend studies to remedy this
55
54
– Investigating future directions:
53
• Reporting standards
• Clinically relevant but specific endpoint definitions
• Inter-institutional data synthesis (atlases or pooling)
52
51
-1.0-0.9-0.8-0.7-0.6-0.5-0.4-0.3-0.2-0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
log 10 n
Best fit value of n sensitive to grid noise
- likelihood function is flat for n > 0.1
EUD atlas for severe pneumonitis at 6 months
1
0.8
0.6
0.4
0.2
log10 (n)
correlation
0.1
residual deviance (=-2*ln(likelihood))
61
0
-0.2
-0.4
-0.6
-0.8
-1
10/78 10/60 7/31 4/14 3/6
1/2
1/1
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
10/78 10/62 7/33 4/17 3/7
1/2
1/1
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
10/78 10/62 7/39 4/17 3/7
1/2
1/1
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
10/78 10/64 9/43 5/21 3/9
1/3
1/1
1/1
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
10/78 10/68 9/49 6/26 3/11 2/4
1/2
1/1
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
10/78 10/69 10/55 6/31 4/15 3/7
1/2
1/1
1/1
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
10/78 10/74 10/59 9/42 5/22 3/9
1/3
1/1
1/1
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
10/78 10/76 10/62 9/50 6/36 4/16 2/5
1/1
1/1
1/1
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
10/78 10/76 10/70 10/59 9/45 6/33 3/15 1/4
1/1
1/1
1/1
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
10/78 10/78 10/76 10/68 9/57 9/47 7/35 4/18 1/3
1/1
1/1
1/1
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
10/78 10/78 10/76 10/76 10/71 9/59 9/49 9/40 5/26 1/6
1/2
1/1
1/1
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
10/78 10/78 10/78 10/77 10/76 10/74 10/67 9/57 9/48 7/36 3/21 1/5
1/3
1/2
1/1
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
10/78 10/78 10/78 10/78 10/78 10/76 10/75 10/74 9/68 9/59 9/48 7/35 3/21 1/8
1/5
1/2
1/2
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/77 10/76 10/75 9/70 9/63 9/50 7/40 4/25 3/15 1/6
1/5
1/2
0/1
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/77 10/76 10/75 9/74 9/68 9/58 7/44 5/31 3/23 2/14 1/7
1/5
0/1
0/1
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/77 10/77 10/76 9/73 9/71 9/66 7/52 6/39 3/28 3/24 2/14 1/7
0/4
0/2
0/1
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/77 10/77 10/74 9/71 9/70 7/61 6/50 5/37 3/28 2/24 1/14 0/8
0/4
0/2
0/1
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/76 10/75 9/73 9/71 7/64 6/56 5/41 3/34 2/30 2/24 0/16 0/8
0/4
0/2
0/0
0/0
0/0
0/0
0/0
0/0
0/0
10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/77 10/75 9/73 9/73 8/67 7/59 5/48 5/39 2/32 1/28 1/21 0/13 0/8
0/2
0/2
0/0
0/0
0/0
0/0
10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/77 10/77 10/74 9/73 9/72 8/67 7/60 5/49 5/43 2/34 1/30 1/26 0/19 0/12 0/4
0/2
0/0
0/0
10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/78 10/77 10/76 10/75 9/72 9/72 8/68 7/59 5/49 3/41 2/37 1/30 1/27 0/22 0/16 0/8
0/1
0
10
20
30
40
50
60
70
EUD (Gy)
8
EUD AND DOSE-VOLUME
BASED ATLASES OF
COMPLICATION INCIDENCE
- A PROPOSAL FOR NEW STANDARDS IN
1) MSKCC, Yorke et al. IJROBP 63 2005: 672-682, from Fig 4a) ( RTOG grade 3, 6
months)
2) Duke, Hernando et al. IJROBP 51 2001: 650-659, from Table 4 ( CTC grade 1, 6
months)
3) Michigan, Kong et al. IJROBP 65 2006: 1075-1086, from Table 4 and Fig 2a)
( SWOG grade 2, 6 months) – bin location and time from authors.
4) MD Anderson, Wang et al. IJROBP 66 2006: 1399-1407, from Fig 2 ( CTC grade 3, 1
year actuarial – includes concurrent chemo patients)
5) NKI, Seppenwoolde et al. IJROBP 60 2004: 748-78, from Fig 3a) ( SWOG grade 2, 6
months)
6) WU, Hope et al. IJROBP 65 2006: 112-124, from Fig 9c) ( SWOG grade 2 – no time
limit) with bin locations from authors, increased by 11% to ~account for inhomogeneity
corrections.
7) Michigan, Martel et al. IJROBP 28 1994: 575-581, from Table 1 ( SWOG grade 1)
with mean doses calculated from relationship between EUD (n=0.87) and mean dose
from Kwa et al., Radiotherapy and Oncology 48 1998: 61-69 Fig 2a).
8) Heidelberg, Oeztel et al. IJROBP 33 1995: 455-460, from Fig 2 and text ( RTOG
acute grade 1).
9) Milan, Rancati et al. Radiother. and Oncol. 67 2003: 275-283, from Fig 3 ( SWOG
Grade 2 – no time limit, patients without COPD – includes induction chemo patients).
10) Gyeonggi, Kim et al. Radiology 235 2005: 208-215, from Table 5 ( RTOG grade 3,
6 months – includes concurrent chemo patients) – median values of mean dose in each
bin provided by the authors.
11) Logistic fit: data fit to the form (f/(1+f)), where f=exp(b0+b1*dmean). Best fit values
[95% confidence intervals] are b0 = -3.63 [-3.53--3.74], b1 = 0.118 [0.109-0.128],
corresponding to D50 = 30.75 [29.9 – 31.7] Gy and 50 = 0.907 [0.836 – 0.987].
REPORTING RESULTS OF TREAMENT
PROTOCOLS.
Andrew Jackson, Ellen D. Yorke,
Kenneth E. Rosenzweig,
Ennapadam Venkatraman,
and C. Clifton Ling
Memorial Sloan-Kettering Cancer Center
Dose-volume limits
with LQ corrected doses (a/b = 3 Gy)
estimates of LKB volume effect parameter n
100
66 Gy: 33%
66 Gy: 14%
80
% volume
66.2-70.2Gy: 11%
75.6 Gy: 19%
60
70-76 Gy: 9%
69 Gy: 25%
3 Gy/fr
40
RTOG 94-06
Tucker
Study
Jackson 70.2
Jackson 75.6
Akimoto
Wachter
Koper
Cozarrino
Zapatero
70-75.6 Gy: 7%
Huang
Hartford
75.6 Gy: 34%
Grade 1
Fiorino
Boersma
Rancati
Peeters
Sohn
20
74-78 Gy: 23%
70.2 Gy: 6%
70-78 Gy: 3%Grade 3
1
0
10
20
30
40
50
60
70
80
90
2
3
4
5
volume effect parameter, n
LQ equivalent dose in 2 Gy fractions (Gy)
9
Data Set
Dose-Volume and EUD Atlases
• 78 patients from a Phase I dose escalation study
of 3D-CRT of non-small cell lung cancer*
– Treated doses 57.6-90 Gy
• Endpoint: RTOG Grade 3 radiation
pneumonitis (G3RP)
– G3RP = steroids and/or oxygen
– Pts either developed G3RP within 6 months following
treatment or survived that time without it
– There were 10 instances of G3RP
* Rosenzweig et al, Cancer 103:2118-27, 2005; Yorke et al, IJROBP 63:672-682, 2005
Calculation of EUD
• EUD values were generated for total lung
– Doses biologically corrected to l.q. equivalent
doses delivered in 2 Gy fractions using
/ =3Gy
– EUD calculated for each of a grid of n values
(log10(n) varying from -1 to + 1 steps of 0.1)
EUD(n j ) =
v k (d k )
k
1
• We previously introduced the dose-volume
atlas of complication incidence*
– analyzed data from the lung protocol
– does not account for overall shape of DVH
• Here we introduce the EUD atlas
– Accounts for overall shape of the DVH
– Atlases can be based on other models
* Jackson et al, Sem in Rad Onc 16:260-268, 2006
Additive property of the atlas
• Provided dosimetric and endpoint
information is compatible
– Doses defined in same way
– Endpoint defined in same way
• Data from different atlases (A and B) can
be added, grid point by grid point:
nj
nj
Nc(A,B) / Np(A,B) = [Nc(A) + Nc(B)] / [Np(A) + Np(B)]
10
Additive property of the atlas
• Facilitates meta-analysis of data from
different institutions
– Low numbers of complications from individual
protocols can be combined
– Potentially synergistic effect if adopted as a
standard of reporting
Calculations based on Atlas data
• At each grid point, can calculate e.g.:
Inefficient use of space
• Many repeated numbers where no patient
treatments occur
• Solution:
– Shift the range to eliminate this dead space
– Record to shift so that correct position can be
reconstructed
Calculations based on Atlas data
• For each value of n, can calculate e.g.:
– Logistic regression of EUD with endpoint
– Confidence limits on complication probability
for patients with EUD > grid value
– Probability that complication rate in patients
with EUD > grid value is greater than
tolerance
• p-value for correlation of EUD with endpoint
• range of n values for significant correlation with
endpoint
• Likelihood of fit of EUD response
• Find max likelihood value and confidence intervals
for n
11
Probability that G3PR rate is > 20%
Probability that observed complications
arise from true rate > 20%
1.0
• Need to determine effect of grid spacing
noise on
0.9
0.8
0.7
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
0.6
0.5
log10 n
0.4
0.3
0.2
0.1
0.0
Atlas grid spacing introduces
statistical noise
– Range of significant correlation
-0.1
-0.2
– Best fit value of n parameter
-0.3
-0.4
-0.5
-0.6
– confidence interval on fitted value of n
-0.7
-0.8
-0.9
-1.0
0
10
20
30
40
50
60
70
80
EUD (Gy)
Conclusions
• Demonstrated utility of EUD and dose volume
atlases using our lung data
• Fitting n values from EUD atlas data
–
–
–
–
At n 1 use 0.5 Gy grid
At n < 1 use 2 Gy grid
Noise should decrease as # pts and comps increase
Reformatting atlas (shifting range) for each value of n:
• better use of atlas space
• better grid resolution
Conclusions
• Publication of atlases facilitates in depth
meta-analysis of dependence of outcome
data on atlas variables
• Atlases allow for useful publication of
treatment series with few to no comps
• Atlases facilitate the quantitative
assessment of potential risks and benefits
of future treatments
12
P value, correlation with severe
pneumonitis – exact vs atlas (2Gy bins)
p value vs n, 2 Gy binning and no binning in EUD
Residual deviance
- exact and atlas values (2-0.5 Gy Bins)
Residual Deviance for EUD from Atlas (comparing various spacings)
61
correlation
0.1
p value
anti-correlation
0.01
p value vs n (eud bins-2Gy)
p=0.05
p value vs n (exact EUD per patient)
0.001
-1.0-0.9-0.8-0.7-0.6-0.5-0.4-0.3-0.2-0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
log10 n
residual deviance (=-2*ln(likelihood))
1
60
EUD in 0.5 Gy bins
Exact EUD
EUD in 2 Gy Bins
EUD in 1 Gy bins
59
58
57
56
55
54
53
52
51
-1.0-0.9-0.8-0.7-0.6-0.5-0.4-0.3-0.2-0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
log10 n
Determining grid spacing
• Noise increases at n>1
Basic Atlas Data
• 2 dimensional grid in EUD and log10(n)
– Range of EUD values , grid size gets cruder
• For p value range, and 95% confidence interval
on best fit value of n,
– 2 Gy grid spacing in EUD is adequate
• For lower limit of 68% confidence interval on n,
– 1 Gy grid spacing in EUD is adequate
• At each grid point (EUDi, log10(nj)), record
– # patients with G3RP, (Nci,j), with EUD(nj) > EUDi
– total # patients, (Npi,j), with EUD(nj) > EUDi.
• Best fit value of n sensitive to grid noise
– likelihood is a flat function of n
– low statistics
• Display as explicit ratio:
Ncij/Npij
13
Residual deviance
- exact values
Best fit EUD response
Best fit EUD response (n = 2), for >= grade 3 pneumonitis
Residual deviance for exact EUD per patient
1.0
60
exact EUD per patient
max likelihood, n = 2.0
lower 68.3% conf limit: n = 1.16
lower 95.4% conf limit: n = 0.710
59
58
57
56
55
54
53
52
51
-1.0-0.9-0.8-0.7-0.6-0.5-0.4-0.3-0.2-0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
log10 n
Probability of >= grade 3 pneumonitis
residual deviance (=-2*ln(likelihood))
61
best fit (n=2), EUD50 = 15.3 Gy γ50 = 1.08
0.9
>= grade 3 pneumonitis data (quartiles)
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0
5
10
15
20
25
30
EUD (Gy), n=2
Upper 68% confidence limit on
complication probability
Data Set
Upper 68% confidence limit on complication probability
1.0
0.9
0.8
0.7
0.5
0.4
0.3
log10 n
• treatments based on a single
CT data set
• treatment plans restorable
from electronic archive
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.2
0.6
0.1
0.7
0.8
0.9
1.0
0.0
-0.1
-0.2
-0.3
-0.4
-0.5
-0.6
-0.7
-0.8
-0.9
-1.0
0
10
20
30
40
50
60
70
80
EUD (Gy)
14
Liver
• Charlie Pan, Brian D Kavanagh, Laura A.
Dawson, X. Allen Li, Shiva K Das, Moyed
Miften, Randall K Ten Haken
15