Subgroup-specific dose finding using Bayesian clustering in phase I-II clinical trials

The objective of clinical trials is to determine whether a new medical treatment will be safe and effective for human patients, and which dose provides the best balance of safety and efficacy. Selection of the optimal dose is usually the goal of the first phase of clinical trials for a new treatment. Unfortunately, most statistical models for dose-finding clinical trials only recommend one dose for all patients who meet the trial inclusion criteria – not accommodating patient groups with different capacities for efficacy or different tolerances for side effects. The new statistical model we propose allows researchers to recruit patients from 2-5 different subgroups (defined at the beginning of the trial) who might benefit from the same treatment, but not at the same dose. For example, cancer patients whose tumors grow using the same biochemical pathway, but whose tumors started in different organs.

After establishing that our method out-performs existing statistical models for handling dose-finding in subgroups using simulation studies, we add practically relevant options to accommodate scenarios where some subgroups recruit patients more quickly than others, and we improve safety of the trial by slowing recruitment at the beginning of the trial and adding additional criteria for a subgroup to stop enrollment if the drug seems unsafe or ineffective.

Finally, we consider methods of allocating patients to the doses which aim to maximize the amount of information gained from each participant. We obtain a more precise estimate of the optimal dose while maintaining patient safety.