A Dose-Finding Design for Drug Combinations Using a Bayesian 4 Parameter Logistic Model With Penalised D-Optimality

Drug combinations, where two drugs are co-administered, are being explored widely within Phase I trials in oncology. These trials aim to identify the maximum tolerated dose combination (MTDC) by adapting the dose of both treatments based on current trial data. Various methods have been proposed to inform how escalation decisions are made within a trial. We propose a model-based design which aims to adapt the dose combination that participants receive to a level that maximises the information gained from each move across the toxicity surface, subject to admissibility and escalation constraints (penalised D-Optimality strategy). We simulated multiple different scenarios, and the penalised D-Optimality strategy is shown to have similar operating characteristics to the Interval approach, which is widely used within Phase I oncology trials. Furthermore, the penalised D-Optimality strategy shows further exploration across the toxicity surface, providing the potential to identify multiple MTDCs, further characterise the risk-benefit profile and provide flexibility for dose selection in future trials. We also show how priors of the two drugs can be specified to show a favoured escalation route if pre-clinical data suggests this.