ABSTRACT
In some phase I–II trials, Toxicity can be evaluated quickly but Efficacy requires a longer time period to be evaluated. For example, in oncology, particularly for cytotoxic agents, Toxicity is typically observable shortly after the treatment is administered; whereas evaluating tumor response may take many weeks or months. To account for potentially non-ignorable dropout, Efficacy is treated as a time-to-event outcome, with a joint model for the Efficacy and dropout processes. While the primary interest is Efficacy, and not the dropout process, the reason for modeling them jointly is to account for nonignorable missing data caused by dropout. The dropout process can be viewed as an informative censoring process for the time to Efficacy. To examine the design's robustness, conducted sensitivity analyses by simulating the times to Efficacy from accelerated failure time models with a log-logistic error. Toxicity, and dropout were matched to the scenarios with dropout rate 30%.
