Adaptive biomarker-driven designs in Phase III clinical trials Alex - - PowerPoint PPT Presentation

Adaptive biomarker-driven designs in Phase III clinical trials

Alex Dmitrienko Vice President Center for Statistics in Drug Development Quintiles Innovation

KU Medical Center Department of Biostatistics April 18, 2014

Outline

– Personalized medicine approach

• Development of tailored therapies

– Clinical trial designs

• Subpopulation-only designs versus multi-population designs

– Case study

• Oncology Phase III trial

– Adaptive biomarker-driven clinical trial designs

• Development and optimization of complex data-driven decision

rules

2

Personalized medicine approach

– Tailored therapies

• Novel therapies that target subgroups of patients with certain

characteristics

• Patient characteristics include demographic variables, clinical

variables, gene or protein expression markers, etc

– Regulatory guidance documents

• U.S. (Food and Drug Administration): Enrichment strategies for

clinical trials to support approval of human drugs and biological products (December 2012)

• Europe (Committee for Medicinal Products for Human Use:

Guideline on the investigation of subgroups in confirmatory clinical trials (January 2014)

3

Clinical trial designs

– Patient populations

• Overall population (OP)
• Marker-positive subgroup (M+)
• Marker-positive subgroup (M-)

– Therapeutic benefit

• Greater therapeutic benefit is expected in the marker-positive

subgroup

4

Clinical trial designs

– Subpopulation-only designs

• Enrollment is restricted to M+ (only marker-positive patients are

enrolled) and M- is not studied (Freidlin, McShane and Korn, 2010)

• Also known as biomarker-enriched designs

– Multi-population designs

• All patients are enrolled and treatment effect is examined in OP as

well as M+ (Millen et al., 2012)

• Also known as biomarker-stratified designs

5

Subpopulation-only designs

– Breast cancer example

• Herceptin (trastuzumab) for treatment of breast cancer (Romond et

al., 2005)

• Important marker: Amplified HER2 gene
• Efficacy in marker-positive patients (with HER2-positive tumors)

was established and marker-negative patients were not enrolled

• Marker-negative patients were denied access to potentially

beneficial treatments but Herceptin is likely to have a positive effect in classifier-negative patients (Paik et al., 2008)

– Regulatory position

• Desirable to evaluate the efficacy and safety of new treatments in

both marker-positive and marker-negative patients (FDA enrichment guidance)

6

Case study

– Phase III clinical trial

• Patients with a rare type of cancer

– Treatment comparison

• Experimental therapy plus chemotherapy versus placebo plus

chemotherapy

– Primary endpoint

• Overall survival (OS)

– Potential predictive biomarker

• Biomarker (protein expression marker) discovered in Phase II trial
• Believed to be predictive of treatment response

7

Adaptive biomarker-driven trial design

– Two-stage design

• Interim analysis to review the safety and efficacy data

– Stage I (before the interim analysis)

• Multi-population trial design (both marker-negative and marker-

positive patients are enrolled)

– Stage II (after the interim analysis)

• Trial design may be adaptively modified at the interim analysis to

improve the overall probability of success

8

Two-stage design

9

Futility stopping rules, population selection rules and sample size adjustment rules will be applied at the interim analysis Stage I Stage II Marker positive Marker negative ??? Interim analysis

Decision rules at the interim analysis

– Futility stopping

• Evaluate futility in the overall population and marker-positive

subgroup

– Population selection

• Select the overall population or marker-positive subgroup for

Stage II

– Sample size adjustment

• Increase the sample size in Stage II if necessary

10

Decision rules at the interim analysis

11

Summary of all decision rules (↑ increase the sample size) ↑ M+ Stop

Low

Probability of success in M+

High

Probability of success in OP

Low High

M+

Medium

↑ OP ↑ OP M+ OP

Medium

Futility stopping rules

– Futility stopping rules

• Futility stopping rules will be implemented based on predicted

probability of success at the interim analysis

– Commonly used approaches

• Frequentist (conditional power)
• Bayesian (predictive power)
• Bayesian (predictive probability)

12

Conditional power

13

Conditional power function CPn(d)=P(ZNza|Zn,d)

• Likelihood of a statistically significant result given the interim data
• Zn, Test statistic at the interim analysis
• ZN, Test statistic at the planned end
• d, Assumed treatment difference

Observed data Future data are generated from assumed

treatment difference

Interim look Start End

Statistically significant difference

Predictive power

14

Predictive power function PPn=CPn(d)f(d|Zn)dd

• Likelihood of a statistically significant result given the interim data

averaged over the posterior distribution of d

• f(d|Zn), Posterior density of treatment difference d given the interim

data

Observed data Future data are generated from posterior distribution

Interim look Start End

Prior distribution Posterior distribution Statistically significant difference

Predicted probability of success

– Outcome 1: Low probability of success

• Discontinue enrollment due to futility if the predicted probability of

success at the planned study end is low (e.g., less than 30%)

– Outcome 2: Medium probability of success

• Apply the population selection and sample size adjustment rules if

the predicted probability of success at the planned study end is medium (e.g., 30-70%)

– Outcome 3: High probability of success

• Apply the population selection and sample size adjustment rules if

the predicted probability of success at the planned study end is high (e.g., greater than 70%)

15

Population selection rules

16

Since greater benefit is expected in M+, predicted probability of success in OP is lower than in M+

Low

Probability of success in M+

High

Probability of success in OP

Low High Medium Medium

NA NA NA

Population selection rules

17

No group is selected for Stage II (futility rule is met in both OP and M+) because the success probability in M+ is low Stop

Low

Probability of success in M+

High

Probability of success in OP

Low High Medium Medium

Subgroup selection rules

18

OP is selected for Stage II because the success probability in OP is comparable to that in M+ (non-informative marker)

Low

Probability of success in M+

High

Probability of success in OP

Low High Medium

OP OP

Medium

Population selection rules

19

M+ is selected for Stage II because the success probability in M+ is medium or high and no effect in M- M+

Low

Probability of success in M+

High

Probability of success in OP

Low High

M+

Medium Medium

Population selection rules

20

M+ and OP are selected for Stage II because the success probability is high in M+ and medium in OP

Low

Probability of success in M+

High

Probability of success in OP

Low High Medium

OP M+

Medium

Population selection rules

21

Summary of population selection rules M+ Stop

Low

Probability of success in M+

High

Probability of success in OP

Low High

M+

Medium

OP OP M+ OP

Medium

Potential regulatory claims

– Broad claim

• Treatment effectiveness in the overall population only (OP only)

– Restricted claim

• Treatment effectiveness in the marker-positive subgroup only (M+
• nly)

– Enhanced claim

• Treatment effectiveness in the overall population as well as marker-

positive subgroup (OP and M+)

22

Sample size adjustment rules

– Promising zone approach

• Adjust the sample size in the overall population or marker-positive

subgroup in Stage II based on the predicted probability of success

– Outcome 1: Original sample size

• in M+ or OP if the predicted probability of success is high (e.g.,

greater than 70%)

– Outcome 2: Increase the sample size

• Increase the sample size in M+ or OP in Stage II if the predicted

probability of success is medium [promising zone] (e.g., 30-70%)

• The sample size will be adjusted to increase the predicted

probability of success to 70%

• Sample size increase will be capped (e.g., at 30%)

23

Sample size adjustment rules

24

Use the original sample size in M+ or OP since the predicted probability

• f success is high

M+ Stop

Low

Probability of success in M+

High

Probability of success in OP

Low High

M+

Medium

OP OP M+ OP

Medium

Sample size adjustment rules

25

Increase the sample size in M+ since the success probability is medium M+ Stop

Low

Probability of success in M+

High

Probability of success in OP

Low High

M+

Medium

OP OP M+ OP

Medium

Sample size adjustment rules

26

Increase the sample size in OP since the predicted probability of success is medium M+ Stop

Low

Probability of success in M+

High

Probability of success in OP

Low High

M+

Medium

OP OP M+ OP

Medium

Final decision rules

27

Summary of all decision rules at the interim analysis (↑ increase the sample size) ↑ M+ Stop

Low

Probability of success in M+

High

Probability of success in OP

Low High

M+

Medium

↑ OP ↑ OP M+ OP

Medium

Final analysis

– Impact of data-driven decisions

• Virtually all adaptive (data-driven) decisions at an interim analysis

change the distribution of the treatment effect test statistics in OP and M+ at the final analysis

• Overall Type I error rate is inflated if no adjustment is introduced

– Adjustment for data-driven changes

• Combination function approach (Brannath, Posch and Bauer, 2002)

will be applied to adjust the test statistics in OP and M+ and protect the overall Type I error rate

– Adjustment for multiple comparisons

• Fallback procedure (Dmitrienko and D’Agostino, 2013) will be

applied to control the Type I error rate if two population are selected for the final analysis (OP and M+)

28

Adaptive biomarker-driven clinical trial designs

– Design parameters

• Timing of the interim analysis
• Cutoff points used in the population selection and sample size

adjustment rules

• Cap used in the sample size adjustment rule

– Clinical trial optimization

• Comprehensive quantitative evaluation of candidate trial designs

was performed to optimally select individual designs parameters

• Operating characteristics of over 50 different trial designs were

evaluated to select the final design

29

Summary

– Adaptive designs

• Adaptive designs with complex decision rules are becoming more

widespread in Phase III trials

• Powerful statistical methods are available to support complex data-

driven decision making rules

– Operational issues

• Overall decision making process must be carefully planned, e.g.,

set up an independent data monitoring committee and data analysis group, set up a flexible randomization system, develop site expansion strategies, etc

30

References

Brannath, W., Posch, M., Bauer, P. (2002). Recursive combination tests. Journal of the American Statistical Association. 97, 236-244. Dmitrienko, A., D'Agostino, R.B. (2013). Tutorial in Biostatistics: Traditional multiplicity adjustment methods in clinical trials. Statistics in Medicine. 32, 5172-5218. Freidlin, B., McShane, L.M., Korn, E.L. (2010). Randomized clinical trials with biomarkers: Design issues. Journal of National Cancer Institute. 102, 152- 160. Millen, B., Dmitrienko, A., Ruberg, S., Shen, L. (2012). A statistical framework for decision making in confirmatory multipopulation tailoring clinical trials. Drug Information Journal. 46, 647-656. Paik, S et a. (2008). HER2 status and benefit from adjuvant trastuzumab in breast cancer. New England Journal of Medicine. 358, 1409-1411. Romond, E.H. et al. (2005). Trastuzumab plus adjuvant chemotherapy for

• perable HER2-positive breast cancer. New England Journal of Medicine.

353, 1673-1684.

31