Early Trial Discontinuation in Toxicity-Driven, Dose-Escalating, Phase I Cancer Trials: Occurrence, Outcomes and Predictive Factors

Introduction The selection of patients for phase I cancer trials remains challenging. Patients who dropout of the trial before completion need to be replaced and this can result in significant delays to trial completion. Objective The objective of this study was to identify patients enrolled in phase I oncology trials who were unable to complete the minimum evaluation period of the trial, and to use these data to develop a predictive model of risk factors for patient replacement. Patients and Methods We retrospectively reviewed all consecutive patients who were enrolled in dose-escalating phase I cancer trials at four medical centers in France between May 2003 and May 2013. Replacement was defined as trial discontinuation before 6 weeks, without the occurrence of dose-limiting toxicity. Using logistic regression and decision-tree analyses, we developed a predictive model to identify patients who were at high risk for replacement, and also their common risk factors. This model was designed to provide maximum specificity and a negative predictive value. Results Of 332 patients enrolled in the study, 16 had to be replaced (4.8 %). The median overall survival time was 45 days for the patients who were replaced versus 480 days for the patients who were not replaced (p < 0.0001). In the univariate analysis, the risk factors for replacement included Eastern Cooperative Oncology Group performance status (ECOG-PS) = 2 [odds ratio (OR) 11], Royal Marsden Score (RMS) = 3 (OR 29), and enrollment in a study investigating multiple agents (OR 7). Multivariate analysis retained PS >= 2 and RMS = 3 as independent predictive factors for replacement. The following two patient subgroups were identified: low risk of replacement (RMS <= 2 and PS <= 1) and high risk of replacement (RMS = 3 or PS >= 2). The rates of replacement were 8/312 (2.5 %) and 8/20 (40.0 %) for low- and high-risk patients, respectively, and the specificity and negative predictive values of our model were 96.2 and 97.5 %, respectively. Conclusions Approximately 5 % of enrolled patients were replaced, and these patients experienced very poor outcomes. To minimize trial delays, enrolling patients who have an RMS = 3 or PS >= 2 should be avoided.