statistical challenges in immuno-oncology · 2013. 11. 26. · moa: cytotoxic vs. cytostatic agents...

1

Statistical Challenges in Immuno-Oncology

EFSPI, Brussels, Belgium

November 7, 2013

Veerle De Pril, MSc, Associate Director, Oncology Biostatistics, BMS

Tai-Tsang Chen, PhD, Director, Oncology Biostatistics, BMS

2

Outline

Cytotoxic vs. cytostatic agents

Mechanism of action

Endpoints

Immunotherapies

Important issues to consider in study design and analysis

Efficacy

– Overall Survival

3

MOA: Cytotoxic vs. Cytostatic Agents

Cytotoxic agents

Dose-dependent rapid cell kill or tumor shrinkage

Lack of selectivity leads to undesired toxicity or side effects

Cytostatic agents

Inhibit or suppress cellular growth or division which leads to delayed progression

Minimal or less severe toxicity, prolonged duration of treatment at lower dose

4

Endpoints: Cytotoxic vs. Cytostatic

Cytotoxic

OS: Clinical benefit

BOR (WHO or RECIST): Direct cell kill action leads to tumor shrinkage

Cytostatic

OS: Clinical benefit

PFS/TTP: Stop or delay tumor growth

BOR: Some may shrink tumor

5

Immunotherapies

Stimulate the patient’s own immune system to fight cancer

Immune cell activation; change in tumor burden

Toxicity or side effects caused by the modulation of immune activity

Endpoints remain similar

OS: clinical benefit

BOR: tumor shrinkage

6

Important Issues in Design and Analysis in Immuno-Oncology

Sample size determination

Expected number of events

Timing of analysis

Efficacy analysis

Interim analysis strategy

Additional analysis considerations

7

Typical Survival Curve – Advanced Breast Cancer

8

Ipilimumab (Yervoy) in Metastatic Melanoma

G R O U P # D E A T H S / # R A N D O M IZ E D M E D IA N (95% C I)

IP I+ D T IC 198 /250 11 .17 (9 .40 - 13 .60 )D T IC 226 /252 9 .07 (7 .75 - 10 .51 )

S U B JE C T S A T R IS KIP I+ D T IC 2 50 1 82 1 15 8 6 6 9 5 7 5 0 4 6 4 4 1 4 1 0D T IC 2 52 1 60 9 0 6 4 4 4 3 7 3 0 2 6 2 2 8 1 0

IP I+ D T ICC E N S O R E D

D T ICC E N S O R E D

/w w b dm /c lin /p ro j/ca /184 /024 /va l/s ta ts /C S R /ta i/dm c/dm cs .sas R U N D A T E : 18 -Jan -2013 9 :11

C A 184-024 F irs t L ine M e lanom a S tudyD M C a n lays is based on da tabase

PR

OP

OR

TIO

N A

LIV

E

0 .0

0 .1

0 .2

0 .3

0 .4

0 .5

0 .6

0 .7

0 .8

0 .9

1 .0

M on ths

0 6 1 2 1 8 2 4 3 0 3 6 4 2 4 8 5 4 6 0 6 6

Yervoy + DTIC

DTIC alone

1 2 3 4 Years

Yervoy + gp100

Yervoy alone

gp100 alone

* Hodi, S. et al. NEJM , 363 (8): 2010

Months * Maio et al. presented at 37th ESMO, 2012

Previously Treated

Previously Untreated

9

Interferon alfa-2b (Intron-A) – Adjuvant Melanoma

Kirkwood et al., 2004, Clinical Cancer Research

10

Pegylated Interferon alfa-2b (Sylatron): Relapse-Free Survival – Adjuvant Melanoma

Eggermont, AMM, et al., 2012, Journal of Clinical Oncology

11

Study Design and Sample Size Determination

Standard study design

Assumes exponential distribution

Unconventional study design

Long-term survival (or “cure rate” or “functional cure”)

Delayed clinical effect

Does unconventional study design impact sample size / power calculation?

12

Exponential OS Study Design

* Proportional hazards model (exponential)

___ CONTROL ___ EXPERIMENTAL

13

Long-Term (LT) Survival

* Proportional hazards cure model

___ CONTROL ___ EXPERIMENTAL

14

Delayed Clinical Effect

* Non-proportional hazards model

___ CONTROL ___ EXPERIMENTAL

HR1 HR2

15

Delayed Clinical Effect with Long-Term Survival

* Non-proportional hazards cure model

___ CONTROL ___ EXPERIMENTAL

HR1 HR2

16

Example of a Standard Study Design

Consider the following standard study design

Exponential distribution

Median OS: 12 vs. 16 months (HR=0.75)

Power: 90%

Two-sided type I error rate: 5%

Accrual rate: 20 pts/month

No interim analysis

Required number of events: 512 events

Sample size: 680 subjects

Accrual duration: 34 months

Study duration: 48 months

17

Impact of LT Survival and Delayed Clinical Effect on Study Duration and Power

Standard

(exponential)

LT survival --

Delayed effect --

Sample size 680

# events 512

Hazard Ratio 0.75

Power 0.90

Study duration 48

* Based on 10000 simulations

LT Survival

0.10/0.18

--

680

512

0.75

0.90

55

Delay

--

3 m

680

512

1/0.75

0.70

47

LT Survival /

Delay

0.10/0.17

3 m

680

512

1/0.75

0.70

54

18

Impact of LT Survival and Delayed Clinical Effect on Study Duration and Power

Long-term survival

Results in prolonged study duration

Higher LT survival results in longer study duration

Delayed clinical effect

Reduces statistical power

Longer delay results in more power loss

Expected number of events

Can the number of events be achieved?

Follow-up duration

Is the study designed to allow sufficient follow-up for all patients?

19

Interim Analysis Strategy

Necessity of interim analysis

Interim analysis vs. final analysis only

Timing of interim analyses

Early vs. late interim analysis

Type of interim analysis

Superiority vs. futility

20

Probabilities for Stopping at Interim Analysis

Standard

(exponential)

Interim sample size 520

# events 256

PETa (superiority) 0.25

PETa (futility) 0.01

* Based on 10000 simulations

PETa = Probability of Early Termination when agent is active

Using O’Brien-Fleming boundaries

LT Survival

540

256

0.25

0.01

Delay

480

256

0.06

0.08

LT Survival /

Delay

500

256

0.06

0.08

21

Interim Analysis Strategy - Conclusion

Delayed clinical effect and LT survival

Careful consideration warranted:

– Necessity of interim analysis

– Timing of interim analysis

– Type of interim analysis

22

Additional Analysis Considerations

Prediction of timing of analyses

Does the long-term survival alter projected study duration?

23

Additional Analysis Considerations Summary Measures

G R O U P # D E A T H S / # R A N D O M IZ E D M E D IA N (95% C I)

IP I+ D T IC 198 /250 11 .17 (9 .40 - 13 .60 )D T IC 226 /252 9 .0 7 (7 .75 - 10 .51 )

S U B JE C T S A T R IS KIP I+ D T IC 2 50 1 82 1 15 8 6 6 9 5 7 5 0 4 6 4 4 1 4 1 0D T IC 2 52 1 60 9 0 6 4 4 4 3 7 3 0 2 6 2 2 8 1 0

IP I+ D T ICC E N S O R E D

D T ICC E N S O R E D

/w w bdm /c lin /p ro j/ca /184 /024 /va l/s ta ts /C S R /ta i/dm c/dm cs .sas R U N D A T E : 18 -Jan -2013 9 :11

C A 184-024 F irs t L ine M e lanom a S tudyD M C an lays is based on da tabase

PR

OP

OR

TIO

N A

LIV

E

0 .0

0 .1

0 .2

0 .3

0 .4

0 .5

0 .6

0 .7

0 .8

0 .9

1 .0

M o n ths

0 6 1 2 1 8 2 4 3 0 3 6 4 2 4 8 5 4 6 0 6 6

Median

Hazard Ratio

Mean (Area Under the Curve)

Yearly OS rates

24

Statistical Analysis Considerations

Primary analysis

Remains log-rank test and Cox model?

Long-term survival

Regulatory: Median vs. OS rates

Market access: Mean

Cure rate models

Delayed clinical effect

Fleming-Harrington weighted log-rank test

25

Summary

Understand disease characteristics and MOA of therapy

– Delayed clinical effect

– Long-term survival

Implications on study design and analyses

26

Reference

Statistical issues and challenges in immuno-oncology, Tai-Tsang Chen, Journal for ImmunoTherapy of Cancer 2013, 1:18