A Bayesian dose-finding design for phase I/II clinical trials with nonignorable dropouts

Area of Science:

• Clinical Trials
• Biostatistics
• Pharmacology

Background:

• Phase I/II trials combine toxicity and efficacy assessments for dose finding.
• Extended trial durations for efficacy evaluation increase susceptibility to patient dropout.
• Nonignorable dropouts, where patients lacking efficacy are more likely to withdraw, pose a significant challenge.

Purpose of the Study:

• To develop a Bayesian phase I/II trial design that effectively accommodates nonignorable patient dropouts.
• To improve the efficiency and accuracy of dose-finding in clinical trials despite missing efficacy data.

Main Methods:

• Treated toxicity as a binary outcome and efficacy as a time-to-event outcome.
• Modeled toxicity using logistic regression and jointly modeled efficacy and dropout times using proportional hazard models.
• Employed a shared frailty model to capture the correlation between efficacy and dropout times.
• Implemented a two-stage algorithm for adaptive patient allocation to optimal doses.

Main Results:

• The proposed Bayesian design demonstrated desirable operating characteristics in simulation studies.
• The design successfully selected the target dose with high probability.
• The algorithm effectively allocated the majority of patients to the identified target dose.

Conclusions:

• The novel Bayesian design provides a robust method for dose finding in phase I/II trials with nonignorable dropouts.
• This approach enhances the reliability of dose selection by accounting for informative missing efficacy data.
• The adaptive algorithm ensures efficient patient enrollment at the most promising doses.