<i>BLRM-EffTox</i> : A seamless two-stage phase 1 Bayesian logistic regression design incorporating efficacy-toxicity trade-off for dose optimization

Traditionally, phase 1 oncology trial designs aim to find the maximum tolerated dose (MTD) based on dose-limiting toxicities (DLTs) observed in patients. The MTD is then further tested for potential anti-tumor activity in larger expansion cohorts, usually with 20-40 patients. However, recent research shows that higher doses are not necessarily associated with better efficacy, especially for novel cancer treatments based on oncogenic or immunological mechanisms of action. Consequently, an optimal dose of these treatments should consider the efficacy-toxicity trade-off, a fundamental motivation for the recent US Food and Drug Administration (FDA) initiative, Project Optimus. In this work, we propose a novel seamless two-stage phase 1 design called <i>BLRM-EffTox</i>, which extends the Bayesian logistic regression model (BLRM) to jointly model toxicity and efficacy events at the end of dose escalation (first stage). Based on a new "decision index", <i>BLRM-EffTox</i> can select more than one dose and then adaptively randomize patients to the selected doses in the subsequent expansion stage to find the optimal dose at the end of dose expansion (second stage). We compare the proposed <i>BLRM-EffTox</i> design to benchmark designs that either use the MTD (<i>BLRM-MTD</i>) or explore two dose levels (<i>BLRM-Eff2d</i>) in the expansion stage. Simulations show that the proposed design outperforms its counterparts. We conclude with two case studies to show the application of this novel design in real-world settings.