A new approach to integrate repeated toxicity data in the analysis of dose-finding trials in oncology

Abstract

Introduction The optimal dose of targeted treatment in oncology may not be the maximal tolerated dose. Evaluating jointly the toxicity and the activity is then desirable. Furthermore, while the toxicity is usually measured at all cycles of treatment, the MTD is commonly defined based on toxicity measure at cycle 1 only. In this communication, we propose a novel adaptive dose-finding approach to identify a dose based on repeated binary toxicity and continuous biomarker outcomes from the 2 first cycles. Methods Probit models were used for the toxicity at each cycle. Linear Gaussian models were used for the activity. The correlation between the binary toxicity and continuous activity outcome is modeled via a latent Gaussian variable. Maximum likelihood estimators were used. Two steps in this design were defined: – the dose-escalation where the decision rules are based only on toxicity observed at the first cycle. At the end of the dose-escalation step, a retrospective analysis using joint modeling of toxicity and efficacy observed at cycles 1 and 2 were proposed; – the expansion cohort where the decision rules are based on both repeated toxicity and efficacy outcomes using joint model. We performed a series of simulation studies to assess the operating characteristics and the robustness of the estimators of our design. Five scenarios with increasing toxicity and activity relationships were explored. Forty patients were enrolled in the trial. We have assessed the performance of this design in the case where we have missing data of toxicity and efficacy at cycle 2. Results The bias of all estimators of different models converges to 0 when the number of subject was more than 50 subjects for each dose level. This leads to a robustness of estimation using maximum likelihood framework. The design presented a good performance for different scenarios. The percentage of correct selection dose varied from 54% to 84%. There is no impact in the estimation parameters with missing data. The design presented then a similar performance where no missing data were presented. Conclusions Using repeated toxicity and efficacy data in dose-finding trials provide more reliable information to estimate the optimal dose for further trials.