How should efficacy of new adjuvant therapies be evaluated in colorectal cancer? Marc Buyse, ScD IDDI, Brussels, Belgium Based on Daniel Sargent’s talks at ODAC in May 2004 and ASCO in June 2004
Dec 13, 2015
How should efficacy of new adjuvant therapies be
evaluated in colorectal cancer?
Marc Buyse, ScD
IDDI, Brussels, Belgium
Based on Daniel Sargent’s talks at ODAC in May 2004
and ASCO in June 2004
HypothesisHypothesis
Disease-free survival (DFS), assessed
after 3 years, is appropriate to replace
overall survival (OS) as an endpoint in
adjuvant colon trials
(i.e. 3-year DFS is a valid
“surrogate endpoint” for 5-year OS)
Surrogate EndpointsSurrogate Endpoints
• Multiple statistical methods proposed• Prentice’s definition and criteria1 • Freedman’s proportion explained2
• Begg and Leung’ concordance 3
• Buyse et al’s correlation 4
• No agreement about best practice
1 Stat Med, 1989. 2 Stat Med, 1992. 3 JRSSA, 2000. 4 Biometrics 1998, Biostatistics 2000, JRSSC, 2001.
Prentice criteriaPrentice criteria
An endpoint can be used as a surrogate if
• it predicts the final endpoint
• it fully captures the effect of treatment
upon the final endpoint
But, how is this verified?
Ref: Prentice, Stat Med, 1989.
Proportion explainedProportion explained
The proportion explained is defined as the proportion of treatment effect that is captured by a surrogate.
But, the associated mathematical construct (the change in a model parameter) is flawed.
Ref: Freedman et al, Stat Med, 1992.
Concordance of resultsConcordance of results
‘The validity of a surrogate endpoint should be judged by the probability that the trial results based on the surrogate endpoint alone are ‘concordant’ with the trial results that would be obtained if the true endpoint were observed and used for the analysis’
But, concordance of hypothesis tests is driven by their power
Ref: Begg and Leung, JRSSA, 2000.
Correlation approachCorrelation approach
An acceptable surrogate must satisfy two conditions:
1. The surrogate must predict the true endpoint
2. The effect of treatment on the surrogate must predict the effect of treatment on the true endpoint
Refs: Buyse and Molenberghs, Biometrics 1998; Buyse et al, Biostatistics 2000.
Trial characteristicsTrial characteristics
• 33 Arms• 9 no treatment control• 24 ‘Active’ rx
• Median follow-up 8 years• 5 year data on 93% of patients
• Due to inconsistent long-term follow-up all analyses censored at 8 years
Trial First Accrual Treatment Arm(s) N
NCCTG 784852 1978 Control vs 5-FU/lev 247 INT 0035 1985 Control vs 5-FU/lev 926
NCCTG 874651 1988 Control vs 5-FU/CF 408
Siena 1985 Control vs 5-FU/CF 239
NCIC 1987 Control vs. 5-FU/CF 359
FFCD 1982 Control vs. 5-FU/CF 259
NSABP C01 1977 Control vs. MOF 773
NSABP C02 1984 Control vs. PVI 5-FU 718
NSABP C03 1987 MOF vs 5FU/CF 1081 NSABP C04 1989 5FU/Lev/CF 2151 NSABP C05 1991 5FU/CF vs + IFN 2176 GIVIO 1989 Control vs 5-FU/CF 867 NCCTG 894651 1989 5FU/Lev/CF 915
NCCTG 914653 1993 5FU/Lev/CF 878
SWOG 9415 1994 5FU/LEV/CF 1078
Total 12915 Total: 33 treatment arms
Patient CharacteristicsPatient Characteristics
• Age• < 50: 2237 (17%)• 50-59: 3487 (27%)• 60-69: 5039 (39%)• > 70: 2071 (16%)
• Treatment• Control: 2454 (18%)• Active: 11610 (82%)
• Gender• M: 7568 (54%)• F: 6496 (46%)
• Stage• I: 210 (2%)• II: 5137 (36%)• III: 8714 (62%)
Recurrence rate by 6 mo Recurrence rate by 6 mo intervalsintervals
0
7.26.9
5.6
4
3.2
2.2 2
1.3 1.20.9 0.8
0.5 0.5 0.4 0.3
0
1
2
3
4
5
6
7
8
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 8
Years after randomisation
Rec
urr
ence
Rat
e (%
)
3.5
3 year DFS 3 year DFS vsvs 5 year OS 5 year OS
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0.5 0.55 0.6 0.65 0.7 0.75 0.8
Disease-Free Survival
Ove
rall
Su
rviv
al
R2= 0.86
r = 0.89
Parameter
Intercept
Slope
Estimate
0.03
0.94
P-value
0.048
<0.001
• Regression equation:
•5 year OS= 0.03+0.94*3 year DFS
•Correlation 0.89, R2 = 0.86
3 year DFS 3 year DFS vsvs 5 year OS 5 year OS
• On an arm-by-arm basis:• 3 year DFS excellent predictor of 5 year OS• Formal approaches suggest surrogacy
• Event rates virtually identical• No impact on sample size• Power for DFS will adequately power for OS
3 year DFS 3 year DFS vsvs 5 year OS 5 year OS
Hazard ratios: DFS Hazard ratios: DFS vsvs OS OS
Disease-Free Survival Hazard Ratio
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3
Ove
rall
Su
rviv
al H
azar
d R
atio
R2= 0.87
r = 0.89
Hazard ratios: DFS Hazard ratios: DFS vsvs OS OS
• Regression equation:
• OS HR = 0.09 + 0.93 * DFS HR
• Correlation 0.89, R2 = 0.87
Parameter
Intercept
Slope
Estimate
0.092
0.93
P-value
0.24
<0.001
0,2 0,4 0,6 0,8 1 1,2 1,4 1,6 1,8
Hazard Ratio
Disease-FreeSurvival
Overall Survival
C01
C02
C03
C04 c1
C04 c2
C05
FFCD
GIVIO
INT-0035
N-78
N-87
N-89 c1
N-89 c2
N-89 c3
N-91
NCIC
S9415
SIENA
OS HR attenuated fromDFS HR toward unity in
12 of 18 comparisons
Hazard ratios: DFS Hazard ratios: DFS vsvs OS OS
Hazard ratios: DFS Hazard ratios: DFS vsvs OS OS
• As an endpoint for comparison:• Hazard ratio for DFS an excellent predictor of HR for OSwith slight attenuation
• Marginally significant improvements in 3 year DFS may not translate into improvements in 5 year OS
Predicted and Actual OS Predicted and Actual OS Hazard RatiosHazard Ratios
0,2
0,4
0,6
0,8
1
1,2
1,4
1,6
FFCD
SIENA
INT-0
035
N-78
C03
GIVIO
NCICN-8
7C02
C04 c2
C04 c1 C01
N-91
N-89
c3C05
S9415
N-89
c2
N-89
c1
Haz
ard
Rat
io
Predicted Overall Survival Hazard RatioActual Overall Survival Hazard Ratio
• Disease-Free Survival an excellent predictor of Overall Survival
• Meets most formal definitions of surrogacy
• Modest attenuation of treatment effect between the two endpoints
• Model allows prediction of OS effect based on DFS effect
DiscussionDiscussion
DiscussionDiscussion
• Is Overall Survival the most desirable endpoint?• It may be the ultimate goal of any therapy for
life-threatening disease• But, it is highly insensitive• True treatment benefit may be confounded by successive lines of therapy
CollaboratorsCollaborators
•S Wieand, M O’Connell - NSABP•J Benedetti - SWOG•R Labianca - Ospedali Riuniti (Italy)•D Haller - ECOG•L Shepherd - NCIC •JF Seitz - University of the Mediterranean (France)•G Francini - University of Siena (Italy)•A de Gramont - Hospital Saint Antoine (France)•R Goldberg - NCCTG/UNC•M Buyse - IDDI (Belgium)•Acknowledgement: E Green (Mayo)