A Bayesian pharmacokinetics integrated phase I–II design to optimize dose-schedule regimes

Abstract

Abstract The schedule of administering a drug has profound impact on the toxicity and efficacy profiles of the drug through changing its pharmacokinetics (PK). PK is an innate and indispensable component of the dose-schedule optimization. Motivated by this, we propose a Bayesian PK integrated dose-schedule finding (PKIDS) design to identify the optimal dose-schedule regime by integrating PK, toxicity, and efficacy data. Based on the causal pathway that dose and schedule affect PK, which in turn affects efficacy and toxicity, we jointly model the three endpoints by first specifying a Bayesian hierarchical model for the marginal distribution of the longitudinal dose-concentration process. Conditional on the drug concentration in plasma, we jointly model toxicity and efficacy as a function of the concentration. We quantify the risk-benefit of regimes using utility—continuously updating the estimates of PK, toxicity, and efficacy based on interim data—and make adaptive decisions to assign new patients to appropriate dose-schedule regimes via adaptive randomization. The simulation study shows that the PKIDS design has desirable operating characteristics.