Group Sequential Design: Uses and Abuses Susan Halabi Department of Biostatistics and Bioinformatics, Duke University October 23, 2015 susan.halabi@duke.edu
What Does Interim Data Say? 2
Group Sequential Design (GSD) Monitoring review toxicity data and analyze response data during the trial Early Termination If intervention harmful If intervention clearly more beneficial than other treatments Unimpressive results associated with intervention 3
GSD as an Aid to Monitoring GSD are an aid to monitoring, but should not be the sole tool used to decide whether to continue or stop the trial. GSD help Data Safety Monitoring Boards decide on early stopping for efficacy or futility. 4 4
Interim Analysis: PREVAIL 100 90 80 Hazard Ratio: 0.706 (95% CI: 0.60,0.84) P<0.0001 70 Survival (%) 60 50 40 30 20 10 0 Enzalutamide Placebo Patients still alive at data cut off Enzalutamide: 72%; Placebo: 63% 0 3 6 9 12 15 18 21 24 27 30 33 36 Duration of Overall Survival (Months) Patients at Risk Enzalutamide 872 863 850 824 797 745 566 395 244 128 Estimated Placebomedian 845OS, months 835 (95% 781CI): Enzalutamide: 744 701 32.4644 (30.1, NYR); 484 Placebo: 328 30.2213 (28.0, NYR) 102 33 27 2 0 2NYR = Not 0Yet Reached 5
Design of COU-302 Patients Progressive chemonaïve mcrpc patients (Planned N = 1088) Asymptomatic or mildly symptomatic R A N D O M I Z E D AA 1000 mg daily Prednisone 5 mg BID Placebo daily Prednisone 5 mg BID Efficacy end points Primary: rpfs by central review OS Secondary: Time to opiate use (cancerrelated pain) 1:1 Time to initiation of chemotherapy Time to ECOG-PS deterioration TTPP metastatic castrate-resistant prostate cancer; OS, overall survival; rpfs, radiographic progression-free survival. 66
COU-AA-302 Design Overall Assumption rpfs OS Type I error rate 0.01 0.04 Power 91% 85% HR Expected events 0.67 (median 6 mo vs 4 mo) 378 (single analysis) 0.80 (median 27.5 mos vs 22 mo) 773 (three interim, one final analysis) Planned OS Analysis Interim 1* (~15% of Total Events) Interim 2 (43% of Total Events) Interim 3 (55% of Total Events) Final Projected Observed OS Events 116 333 425 773 Stopping Probability under H0 <0.0001 0.0008 0.0034 0.04 7
Co-Primary Endpoint: rpfs 100 HR (95% CI): 0.43 (0.35-0.52) P value: < 0.0001 Progression-Free Survival or Survival (%) 80 60 40 20 AA Placebo 0 0 3 6 9 12 15 18 Time to Progression or Death (Months) AA Placebo 546 542 489 400 340 204 164 90 46 30 12 3 0 0 Data cutoff 12/20/2010 8
Primary End Point: OS* 100 80 Survival (%) 60 40 20 0 0 AA+P Placebo+P 3 6 9 12 15 18 21 24 27 30 Time to Death (Months) AA + P (median): NR PL + P (median): 27.2 HR (95% CI): 0.75 (0.61-0.93) P value: 0.0097 33 AA Placebo 546 542 538 534 524 509 503 493 482 465 452 437 412 387 258 237 120 106 27 25 0 2 0 0 9
O Brien-Fleming Boundary Drug is effective * Drug is effective Drug is ineffective * Drug is ineffective 10 10
All Secondary End Points Favor Abiraterone AA + P (n = 546) Placebo + P (n = 542) Median (months) Median (months) HR (95% CI) P Value* Time to opiate use (cancer related pain) Time to chemotherapy initiation Time to ECOG PS deterioration NR 23.7 0.69 (0.57, 0.83) 0.0001 25.2 16.8 0.58 (0.49, 0.69) <0.0001 12.3 10.9 0.82 (0.71, 0.94) 0.0053 Time to PSA Progression 11.1 5.6 0.49 (0.42, 0.57) <0.0001 PSA Decline 50% 62% 24% NA <0.0001 *Adjusted type I error rate for multiplicity 11
Subsequent Therapy AA + P (n = 546) Placebo + P (n = 542) No. with selected subsequent therapy for mpc 242 (44.3%) 327 (60.3%) Docetaxel 207 (37.9%) 287 (53.0%) Cabazitaxel 45 (8.2%) 52 (9.6%) Ketoconazole 39 (7.1%) 63 (11.6%) Sipuleucel-T 27 (4.9%) 24 (4.4%) Abiraterone Acetate 26 (4.8%) 54 (10.0%) 12 12
FDA Assessment Kluetz, CCR 2013 13
Price You Pay: Biased Estimate of Clinical Benefit Analysis Observed Hazard Ratio (95%CI) Interim Analysis Log-Rank p-value 0.75 (0.61-0.93) 0.0097 Final Analysis Log Rank p-value 0.81 (0.70-0.93) 0.0033 14
Final Results: COU302 Ryan et al. Lancet Oncology, 2015 15
IMPACT Trial: Press Release Dendreon Corporation (Nasdaq: DNDN) announced today that it has completed the planned interim analysis of the Phase 3, randomized, double blind, placebo controlled IMPACT (IMmunotherapy for Prostate AdenoCarcinoma Treatment, also known as D9902B) clinical trial designed to assess the safety and efficacy of the investigational active cellular immunotherapy PROVENGE (sipuleucel T) in men with metastatic androgen independent prostate cancer. While Dendreon remains blinded to the data, the independent data monitoring committee (IDMC) reported to Dendreon a 20 percent reduction in the risk of death in the PROVENGE arm relative to placebo (Hazard Ratio= 0.80; 95% Confidence Interval [0.610 1.051]). The IDMC observed no safety concerns and recommended that the study continue to its final analysis. 16
IMPACT Trial: Final Analysis Kantoff et al. NEJM 2010 17
Futility: MainSail Trial Petrylak et al. Lancet Oncology 2015 18
Enthuse Trial 33 Fizazi et al. Journal of Clinical Oncology 2012 19
Conclusion There are a variety of approaches for interim monitoring of clinical trial data. The appropriate monitoring plan depends on the goals of the trial. GSD offer a mechanism to consider terminating a trial early (either for inferiority or superiority) while maintaining the overall type I error rate for a twosided test and the overall type I and type II error rates for a one-sided test. 20
Summary The decision to stop a trial is complex and requires both statistical and clinical judgment GSD should not be used as a sole tool in the decision to stop or continue a trial Totality of evidence for primary, secondary and safety issues also external evidence. 21