Regulatory Challenges in Reviewing Oncology Product Applications

Size: px

Start display at page:

Download "Regulatory Challenges in Reviewing Oncology Product Applications"

Brook Austin
5 years ago
Views:

1 Regulatory Challenges in Reviewing Oncology Product Applications Kun He & Rajeshwari Sridhara Division of Biometrics V Office of Biostatistics Center for Drug Evaluation and Research U.S. Food and Drug Administration June 17,

2 Outline - Precision Medicine - Design - Treatment Switching - Non-proportional Hazards - Accuracy of Estimation - Borrowing Data - Alternative Endpoint 2

3 Precision Medicine - Numerous accelerated and regular approvals over the past several years - Challenges of rare disease drug development now confront the most common human cancers - Many unique characteristics - Innovative design and analysis method 3

4 Design: Basket Trial for Imatinib - Objective is to identify evidence of imatinib activity in any of the studied malignancies - Advanced, life-threatening malignancies expressing one or more imatinib-sensitive tyrosine kinases - Refractory to standard therapy or no proven conventional therapy existed - The number of indications analyzed was not predefined; the number of patients per indication was not prospectively stipulated patients with 40 different malignancies enrolled with 5 indications approved Ref: (1), (2), (3) 4

5 5

6 Design: Bayesian Hierarchical Model Ref: (4), (5), (6), (7), (8) 6

7 Design: Small Size Randomized Trial - Sample size can be small if effect size is large - Siltuximab (Sylvant) for Castleman s disease - Sample size: 53 vs Primary endpoint: durable tumor and symptomatic response - Results: 34% in Sylvant vs 0% in placebo (p=0.0012) Ref: (9) 7

8 Design: Master Protocol Characteristics of a Master Protocol One Protocol Central governance structure Central IRB Central DMC Central Independent review committee Central repository of data and specimens Ref: (4), (10) Study multiple drugs Targeting more than one marker More than one drug for one marker Study multiple markers Overlapping expression of markers Leverage common control group(s) Flexibility to add/remove 8 agents

9 Design: Master Protocol Streamlined enrollment procedures Established systems in place to improve trial processes Central randomization (e.g., via web portal) Central electronic data capture system In-network clinic personnel trained and experienced on existing systems Study start-up time reduced once established Common case report forms could help focus on critical data elements and minimize collection of less important items Efficiencies realized during study conduct; data quality improvements 9

10 Design: Master Protocol Common elements of trial protocols can improve processes Informed consent, clinical monitoring, data close-out, etc. In-network clinical monitors trained and experienced on common elements Efficiencies realized during study conduct; data quality improvements Centralized governance structure Use of central IRBs, a standing DMC, and/or other bodies Reduce study start-up time and provide ongoing efficiencies Central labs, reading centers, etc., with quality assurance oversight Some additional procedures required but improvements can be substantial in both data quality and reduced variability across clinics 10

11 Treatment Switching Definition: patients randomized to the control arm allowed to switch onto the new intervention during the trial after, for example, disease progression Problem: switching most likely leads to an underestimate of the treatment effect on OS Adjustment: a) take no action at all b) Inverse Probability of Censoring Weights (IPCW) c) Rank Preserving Structural Failure Time Models (RPSFTM) 11

12 Treatment Switching: IPCW Procedure - Censor switchers at point of switch - Weight remaining observations according to baseline and time-dependent characteristics - Derive adjusted estimate with new weights Key Assumptions - No unmeasured confounders assumption, require information on all baseline and time-varying characteristics that are prognostic for survival, which predict switching, and continued data collection over time is a problem - Need correctly specified models for survival and for switching Ref: (11) 12

13 Treatment Switching: RPSFTM Procedure - For each patient, split survival into time on treatment and time off treatment - Use g-estimation to identify treatment effect - Conduct standard survival analysis on counterfactual survival times Key Assumptions - Common treatment effect : treatment effect is the same regardless of when treatment is received Ref: (12) 13

14 Treatment Switching: RPSFTM GIST Median in months Hazard Ratio Sunitinib Placebo (95% CI) (n=243) (n=118) Cox Model (0.68, 1.13) a RPSFTM (0.26, 1.13) a: 103 (87.3%) placebo arm patients switched to receive sunitinib Ref: (13) 14

15 Treatment Switching: RPSFTM Thyroid Median in months Hazard Ratio P-value Cancer Lenvatinib (n=261) Placebo (n=131) (95% CI) Cox NA NA 0.73 (0.50, 1.07) a Model RPSFTM NA NA 0.62 (0.40, 1.00) a: 109 (83%) placebo arm patients switched to receive lenvatinib 15 Ref: (14)

16 Treatment Switching: RPSFTM Melanoma Median in months Hazard Ratio (95% CI) Pem 2 Pem 10 Chemo Pem 2 vs C Pem 10 vs C (n=180) (n=181) (n=179) Cox Model (0.65, 1.24) a RPSFTM (0.59, 1.12) a 0.71 a: 86 (48%) patients in control arm switched to receive pem Ref: (15) (0.51, 1.16) (0.43, 1.07) 16

17 Non-Proportional Hazards Pembrolizumab for the Treatment of Ipilimumab- Refractory Melanoma Ref: (15) 17

18 Non-Proportional Hazards Nivolumab for the Treatment of Metastatic Non- Small Cell Lung Cancer With Progression On or After Platinum-Based Chemotherapy Ref: (16) 18

19 Non-Proportional Hazards - For any comparison of two survival curves, there is no single metric or parameter that can capture the entire profile of their difference: HR and difference in median - Sample size calculation and testing procedure - Survival rate at a particular time: overall performance - Piece wise Cox model: partition of time axis - Mean survival time: infeasible due to censoring - Restricted mean: selection of follow-up time point Ref: (17), (18), (19), (20), (21) 19

20 Non-Proportional Hazards Pembrolizumab for the Treatment of Ipilimumab- Refractory Melanoma PFS Pem 2 mg/kg Pem 10 mg/kg Chemo (n=179) (n=180) (n=181) K-M Estimate RMST Ref: (15) 20

21 Accuracy of Estimation Ref: (22) 21

22 Accuracy of Estimation Sunitinib vs. placebo, PNET, in 171 patients DMC took several unplanned looks at the efficacy data and recommended closure after 73 PFS events, representing 28% of the 260 planned PFS events, and did not cross a prespecified efficacy boundary. Ref: (23) 22

23 Accuracy of Estimation Example 1: 40 patients enrolled in a 1:1 ratio, and primary PFS HR=0.97 (95% CI: 0.48, 1.97), and OS HR=0.30, (95% CI: 0.11, 0.86), medians are 21 vs. 15 months Example 2: 120 patients enrolled in a 1:1 ratio, and primary PFS HR=0.68 (95% CI: 0.50, 1.12); and OS HR=0.47 (95% CI: 0.32, 0.72), medians are 26 vs. 14 months 23

24 Borrowing Data Example 1: Objective: convert from AA to RA Issue: enroll pts in a confirmatory randomized trial when the single arm showed a high ORR Proposal: use the extension of the single arm trial to compare with the standard of care from the historical trial Method: propensity score 24

25 Borrowing Data Example 2: Suppose in designing a NI trial, 400 patients are needed in each arm. Can one use 250 patients from historical control trial and enroll 150 for control arm in the current proposed trial? Example 3: Suppose two trials are designed to test the experimental: one in all lines, and another in 1 st line. To increase power of OS analysis, can one pool those 1 st line patients from all lines trial If both primary analyses of PFS are significant? 25

26 Alternative Endpoints - Immune-related Response Criteria: Beaver et al discussed Nivolumab for Melanoma, and Kazandjian et al for NSCLC - Research ongoing - Anti-tumor activity duration of response - Measuring survival rates at one or several specific time points Ref: (24), (25) 26

27 Concluding Remarks During this exciting period of time for oncology drugs development with so many issues and challenges, innovative design, endpoint, analysis method, reporting statistics, and many other tools are needed! 27

28 Acknowledgement Thanks to all statistical reviewers in DBV and clinical reviewers in OHOP for their work done in research and review which directly or indirectly cited in this presentation. 28

29 Reference (1) MHeinrich, HJoensuu, GDemetri et al: Phase II, Open-Label Study Evaluating the Activity of Imatinib in Threating Life-Threatening Malignancies Known to Be Associated with Imatinib-Sensitive Tyrosine Kinases, Clinical Cancer Research 14: , 2008 (2) MCohen, KHe: Clinical and Statistical Review of NDA / S-011, 012, 013, 014, (3) MCohen, STang: Clinical and Statistical Review of NDA / S-017, (4) DBerry, The Brave New World of clinical cancer research: adaptive biomarker-driven trials integrating clinical practice with clinical research, Molecular Oncology 9: , 2015 (5) SBerry, KBroglio, SGroshen, DBerry: Bayesian hierarchical modeling of patient subpopulations: efficient designs of Phase II oncology clinical trials, Clinical Trials 10: , 2013 (6) MEscobar, MWest, Bayesian Density Estimation and Inference Using Mixtures, JASA 90: , 1995 (7) RNeal, Markov Chains Sampling Methods for Dirichlet Process Mixture Models, J of Compuational and Graphical Statistics 9: , 2012 (8) JSethuraman, A Constructive Definition of Dirichlet Priors, Statistica Sinica 4: , 1994 (9) CKo, Statistical Review and Evaluation of BLA / 0 for Sylvant, (10) RSridhara, KHe, LNie, YShen, and STang, Current Statistical Challenges in Oncology Clinical Trials in the Era of Targeted Therapy, Statistics in Biopharmaceutical Research 7: ,

30 Reference (11) MHeman, BBrumback, JRobins, Marginal Structural Models to Estimate the Joint Causal Effect of Nonrandomized Treatments, JASA 96: , 2001 (12) JRobins, ATsiatis, Correcting for Noncompliance in Randomized Trials Using Rank Preserving Structural Failure Time Models, Communication in Statistics, Theory and Methods 20: , 1991 (13) JZhang, Statistical Review and Evaluation of NDA /077, label at (14) XJiang, Statistical Review and Evaluation of NDA / 0, label at (15) HChen, Statistical Review and Evaluation of BLA , label at (16) DKazandjia, DSuzman, GBlumenthal, SMushti, KHe, MLibeg, PKeegan, RPazdur, FDA Approval Summary: Nivolumab for the Treatment of Metastatic Non-small Cell Lung Cancer with Progression on or after Platinum-based Chemotherapy, The Oncologist 21, 1-9, 2016 (17) TKarrison, Restricted Mean Life with Adjustment for Covariates, JASA 82: , 1987 (18) DZucker, Restricted Mean Life with Covariates: Modification and Extension of a Useful Survival Analysis Method, JASA 93: , 1998 (19) PChen, ATsiatis, Causal Inference on the Difference of the Restricted Mean Lifetime between Two Groups, Biometricks, 57: , 2001 (20) XZhang, Comparison of Restricted Mean Survival Times between Treatments based on a Stratified Cox Model, Bio-Algorithms and Med-Systems 9: ,

31 Reference (21) HUno, BClaggett, LTian et al, Moving Beyond the Hazard Ratio in Quantifying the Between-Group Difference in Survival Analysis, Journal of Clinical Oncology 32: , 2014 (22) JZhang, GBlumenthal, KHe, STang, PCortazar, RSridhara, Overestimation of the Effect Size in Group Sequential Trials, Clinical Cancer Research 18: , 2012 (23) ODAC Meeting on April 12, (24) JBeaver, MTheoret, SMushti, KHe, MLibeg, RSridhara, AMcKee, PKeegan, RPazdur, FDA Approval of Nivolumab for the First-Line Treatment of Patients with BRAF V600 Wild-Type Unresectable or Metastatic Melanoma, Clinical Cancer Research, submitted (25) DKazandjian, GBlumental, SKhozin, DSuzman, PKeegan, RPazdur, Characterizatino of Patients Treated with a Programmed Cell Death Protein 1 Inhibitor (anti-pd-1) Past RECIST Progression from a Metastatic Non-Small Cell Lung Cancer (mnsclc) Trial, ASCO Abstract 3000,

Adjusting the Crossover Effect in Overall Survival Analysis Using a Rank Preserving Structural Failure Time Model: The Case of Sunitinib GIST Trial

Adjusting the Crossover Effect in Overall Survival Analysis Using a Rank Preserving Structural Failure Time Model: The Case of Sunitinib GIST Trial Xin Huang 1 and Qiang (Casey) Xu 2 1 Pfizer Oncology