A mismatch between methods and objectives in phase I drug development: Statistical limitations and personalized medicine—A California Cancer Consortium (CCC) study.

Abstract

2538 Background: Patient variation in drug response and toxicity impacts all phases of drug development. While detailed sample-size calculations and analysis based on statistical methodology are critical for addressing variation in late stage trials, phase I studies pose unique and largely unsolved challenges related to variability. Changes in patient selection during the study, small sample-sizes and large patient variation in toxicity are exactly the kind of problems that statistical methods cannot address in small, isolated phase I studies, regardless of apparent mathematical rigor. We have documented these problems via a physician survey. Methods: A 10-question anonymous survey was sent to 670 oncologists at National Comprehensive Cancer Network (NCCN) and CCC institutions via an online survey. 19 % (126/670) of the oncologists polled responded. 78/126 (62%) specialized in Medical Oncology. The number of years in practice varied from 2-45 yrs with a median of 17 yrs. Results: a) 66% of all respondents stated that non-DLT toxicities on one dose level would impact their patient selection on the following dose level, conflicting with the assumption of random patient selection implicit in simulations used in evaluating statistical designs. b) Only 13% stated a desire to target a non-heme toxicity as high as 20% grade 3, while 87% desired a 10% or less grade 3 rate; c) More than half the respondents would prefer not to escalate if 3/3 patients experienced grade 2 LFTs; d) 82% of the respondents thought the appropriate target toxicity differed for patients depending on their potential for becoming a surgical candidate, furthering the need for personalized dosing. Conclusions: Statistical methods in phase I trials are unable to address many of the salient features of phase I study conduct and investigator goals. We will present several approaches we have initiated to address these limitations, and present future plans to help produce a more reliable estimate of a recommended dose. Supported in part by NCI grants U01CA062505, N01-CM-62209, and NCCN data collection assistance.