Codevelopment of Multiple Drug Combinations in Oncology A

Co-development of Multiple Drug Combinations in Oncology: A Bayesian Design for Dose Finding Caimiao Wei, Satrajit Roychoudhury Pfizer Inc. New York, United States 1. Background Figure 1. DLT Rate Probability Intervals Combination therapy in oncology • Combination therapies have potential to improve efficacy and minimize resistance • Co-development of multiple combinations is often necessary to find the optimal therapy • Single agent dose-toxicity data is often available Key development challenges • Limited sample size • Multidimensional escalation, drug-drug interaction, possibility of multiple MTDs • Difficult to prespecify escalation/de-escalation rule Objective • Efficient design to identify RP 2 D for dual and triple combination therapies in a seamless fashion • Allows on-trial adaptation based on evolving data • Leverage all information for better precision • Data scenarios help to understand the on-trial dose recommendations of the proposed design • Dose finding for dual combo can skip dose • Triple combo starting dose adaptively use historical data and on-study dual combo data. DLT rate intervals Figure 3: Hypothetical Scenarios for the Dual Combo and Starting Dose for the Triple Combo 5. 4 Operating Characteristics • Operating characteristics are evaluated under different scenarios (some examples for dual combo are below) via simulations • % of patients treated at overly toxic doses are well controlled: 0 -15% at over-toxic doses • % of trials with a correctly identified MTD is reasonable : 80 -98% at target doses Figure 5: OC for the Dual Combo True DLT rate scenarios for the dual combo Drug A, B (mg) 20, 35 20, 50 20, 70 20, 90 Prior means 0. 13 0. 17 0. 24 0. 31 75% more toxic 0. 24 0. 29 0. 41 0. 54 Lowest dose under dose, higher doses overly toxic 0. 10 0. 25 0. 40 0. 50 Proportion of patients treated at different toxicity intervals We propose an efficient model based approach to facilitate dose finding while co-developing dual and triple combinations 2. Study Design Compounds of interest • Drug A: fixed dose 20 mg • Drug B: 35, 50, 70, 90 mg • Drug C: 5, 7. 5, 10 mg Endpoint • Dose limiting toxicity (DLT) in cycle 1 Framework: sequential dose finding 5. Design Characteristics • Stage 1: Dual combo (A+B) 5. 1 Starting Dose for the Dual Combo • Stage 2: Triple combo (A+B+C), • Feasible to start dual combo at (Drugs A, B) =(20, 70) • Dose-DLT data available for single agent trial of drugs A, B, and C Figure 2: Prior Distribution of DLT rate for • No significant DDI expected but with considerable the Dual Combo uncertainty • Starting dose of triple combo depends on dual data Feasible starting dose for the triple combo Drug A, B, C (mg) 20, 35, 5 20, 5 Proportion of Trials with MTD at different toxicity intervals 20, 35, 7. 5 20, 50, 7. 5 Drug A, B, C (mg) Scenario #1: Triple combo MTD can be established at (20, 50, 7. 5), or trial could further explore (20, 50, 10) and (20, 70, 5) 20, 7. 5 1/3 20, 50, 7. 5 0/3 20, 70, 0 20, 35, 0 1/3 1/4 Scenario 2: Triple combo MTD can be established at (20, 35, 7. 5), or further explore (20, 5) 1/3 Feasible starting dose levels (0. 33 -1) Failed EWOC Criteria 1/4 0/3 20, 50, 0 (0. 16 -0. 33) 0/3 Dual data SC 1 0/3 20, 35, 5. 0 20, 35, 0 2/6 1/4 20, 35, 7. 5 (0 -0. 16) No feasible dose level Figure 4: Hypothetical Scenarios for the Triple Combo Drug A, B, C (mg) 3. Dosing Criteria • Escalation with overdose control (EWOC): dose combinations with P(true DLT rate ≥ 0. 33|data) ≥ 25% is not recommended for next cohort • Escalation to any dose levels ≤ 100% from current dose is allowed; including skipping • Flexible cohort size: 3 -6 • Final decision based on additional data: PK, other safety than DLT, efficacy. 5. 2 Data Scenarios and Starting Dose for the Triple Combo 1/3 1/3 6. Conclusions/Discussions • Proposed approach allows seamless transition from dual to triple combinations • Leverage all available data to elicit the starting dose for the triple combination • Dosing decision guided by sound statistical inference and clinical relevance • Flexible for incorporating intermediate doses • Communication with non-statistician is the key for successful practical implementation • Software available 7. References 1/4 Dual data SC 2 Based on Figure 3 and Figure 4 for dual and triple combo the dose recommendations are clinically meaningful. 1. Neuenschwander B, Matano A, Tang Z, et al. A Bayesian Industry Approach to Phase I Combination Trials in Oncology. In Statistical Methods in Drug Combination Studies. Zhao W and Yang H (eds); Chapman & Hall/CRC; 2014.