Predictive biomarker enrichment designs in Phase II clinical trials

Transcription

1 Predictive biomarker enrichment designs in Phase II clinical trials Deepak Parashar and Nigel Stallard Statistics and Epidemiology Unit Warwick Medical School 05 June 2018 Deepak Parashar 05 June /16

2 Fixed Enrichment Designs Assess biomarker in F = (S, S) Biomarker + (S) Biomarker ( S) Trial Design (Single-arm/Randomise etc.) Off Study Goal: Hypothesis test in biomarker + subgroup S only. Issue: No comparator test in the complementary subgroup S. Is the biomarker predictive? Deepak Parashar 05 June /16

3 Adaptive Enrichment Designs Stage 1: Enrol F Interim Stage 2 STOP Enrich S Continue F Goal: Hypotheses test in S and F. Issue: While S is included in F, no explicit testing in S. Is the biomarker predictive? Deepak Parashar 05 June /16

4 Biomarker ve: To test or not to test? FDA Concerns 1 Design may not be efficient if drug has at least some activity in biomarker ve patients. 2 Effect in biomarker ve patients may never be known. 3 Study would provide no new clinical evidence w.r.t. biomarker ve patients. 4 Implications for Phase III Need for testing in biomarker + and biomarker ve subgroup Deepak Parashar 05 June /16

5 Example: Simon two-stage enrichment design Phase II targeted cancer therapy Determine whether drug has activity only in target population or the general population Outcome is (RECIST) tumour response Single-arm trial Enrichment adaptation (with testing in Biomarker ve) based on Simon two-stage design Jones et. al., Contemp. Clin. Trials 2007 Parashar et. al., Pharmaceut. Statistics 2016 Deepak Parashar 05 June /16

6 Hypotheses (Group-sequential) H 0 : p = p 0, H+ 0 : p+ = p + 0 H 1 : p = p 1, H+ 1 : p+ = p + 1 Assume p < p + Conclude efficacy in full population if we reject H 0 Conclude efficacy in biomarker positive if we reject H + 0 Deepak Parashar 05 June /16

7 Design Schematic Recruit Unselected (N 1, N 1 + ) Go decision Unselected (X 1 k ) Sufficient Response Unselected (X 1 k 1 ) Insufficient Response Unselected (X 1 < k 1 ) Route 1 Go decision Positives (X 1 + k e + ) Sufficient Response Positives (X 1 + k 1 + ) Insufficient Response Positives STOP Route 3 Stage 2 Stage 2 Enrichment Recruit Unselected (N 2, N 2 + ) Recruit Positives (N 2e + ) Go decision Unselected (X k ) Route 1 Cumulative Response Unselected (X < k ) Go decision Positives (X e + k e + ) Route 3 No-Go decision Route 2 Go decision Positives (X + k + ) No-Go decision Deepak Parashar 05 June /16

8 Question: Is the biomarker truly predictive? Suggestion: Randomised Clinical Trial testing in targeted and non-targeted subpopulations! Deepak Parashar 05 June /16

9 Notation h E h C θ S θ S HR S HR S hazard using the experimental drug hazard using the control drug log ( he S ) /hs C ) log (h S E /h S C hazard ratio of S hazard ratio of S HR S HR S θ < 0 experimental treatment more efficient than control θ 0 no improvement with experimental treatment Deepak Parashar 05 June /16

10 Randomised Enrichment Design for TTE Endpoints Enrol (S, S) Test H S : θ S 0 in S E better than C in S E not better than C in S Test H S : θ S 0 Stage 2:Enrol (S, S) in both arms Stop for futility E better than C in S E better than C in S Go Decision (S, S) E not better than C in S Stage2: Enrich S in both arms No further testing Go Decision S Go Decision S No further testing Deepak Parashar 05 June /16

11 Mehta et.al. (Statist. Med. 2014) Key points Based on CER approach (Müller and Schäfer) guarantees strong control of Type 1 error rate. Permits utilisation of all interim data. Hypotheses testing in subgroups S and S instead of F and S. Conditioning event at interim: pair of future logrank test statistic after observing events in S and S Irle and Schäfer: critical value for testing H S satisfies the CRP principle, and guarantees stochastic independence of the logrank test statistic. Triple (k, T, c) (Events, Logrank test statistic, Critical value for T to reject null) Deepak Parashar 05 June /16

12 Methodology Sketch (à la Mehta) Enrol (S, S) Interim: (k S I, T S I, c S I ), (k S I, T S I, c S I ) Stage 2: Continue (S, S) Stage 1 events continuing (incl. censored) Enrich S Planned / Final:(k S, T S, c S ) (k S I (k S I, T S I, T S I, c S I ), c S I ) (k S e, T S e, c S e ) Actual Analysis:( k S, T S, c S ) ( k I S, T I S, c I S) ( k S I, T S I, c S I ) ( k S e, T S e, c S e ) Go Decision (S, S) Go Decision S Go Decision S Deepak Parashar 05 June /16

13 CRP Principle Go Decision S Reject H S if T S e > c S e s.t. P( T e S > c e S T I S ) + P( }{{} T S > c S T I S ) }{{} enrichment non enrichment P(Te S > ce S T S I ) + P(T S > c S T S I ) }{{} fixeddesign Go Decision (S, S) Reject H S if T S > c S s.t. P( T S > c S T S I ) P(T S > c S T S I ) CER of 2-stage design bounded by the error rates of fixed design Deepak Parashar 05 June /16

14 Non-small cell lung cancer trial Clinical Setting: endpoint: Progression-free Survival sample size: 160 patients accrual rate: 15 patients/month interim analysis - after recruitment of 40 patients from each subgroup Target Hazard Ratio: HR S = 0.5 Biomarker prevalence rate: {0.30, 0.35, 0.40, 0.45} Familywise error rate: α = 0.05 Power: 1 β = 0.80 HR S HR S Deepak Parashar 05 June /16

15 Probability of concluding efficacy (HR S = 0.5) Close to the desired power in all cases a trial of 160 patients should provide sufficient evidence of efficacy in the biomarker-positive group. Deepak Parashar 05 June /16

16 Conclusions Efficient when HR S is small (0.5 to 0.6) and HR S is large (0.8 to 1): Predictive biomarker Obtain desired power for recommending an enriched Phase III trial. Copes well with slower recruitment rate as well as varying prevalence rates. Warwick - Roche MRC icase PhD Student starting Sept Deepak Parashar 05 June /16

17 Conclusions Efficient when HR S is small (0.5 to 0.6) and HR S is large (0.8 to 1): Predictive biomarker Obtain desired power for recommending an enriched Phase III trial. Copes well with slower recruitment rate as well as varying prevalence rates. Warwick - Roche MRC icase PhD Student starting Sept Deepak Parashar 05 June /16