An extended proportional hazards model for interval-censored data subject to instantaneous failures

Novel application of a discrete time-to-event model for randomized oral immunotherapy clinical trials with repeat food challenges

The evaluation of double-blind, placebo-controlled food challenges (DBPCFC) generally focuses on a participant passing a challenge at a predetermined dose, and does not consider the dose of reaction for those who fail or are censored due to study discontinuation. Further, a number of food allergy trials have incorporated multiple DBPCFCs throughout the duration of the study in order to evaluate changes in reaction over time including sustained unresponsiveness from treatment. Outcomes arising from these trials are commonly modeled using Chi-squared or Fisher's exact tests at each time point. We propose applying time-to-event methodology to food allergy trials in order to exploit the inherent granularity of challenge outcomes that additionally accommodates repeated DBPCFCs. Specifically, we consider dose-to-failure for each study challenge and extend the cumulative tolerated dose across challenges to result in a dose-time axis. A discrete time-to-event framework is applied to the dose-time outcome to assess the efficacy of treatment across the entire study period. We illustrate ideas with data from the Peanut Oral Immunotherapy Study: Safety, Efficacy and Discovery (POISED) trial, conducted at Stanford University, which evaluated the efficacy of oral immunotherapy on desensitization and sustained unresponsiveness in peanut allergic children and adults. We demonstrate the advantages of time-to-event approaches for assessing the efficacy of treatment over time and incorporating information for those who failed or were lost to follow up. Further, we introduce a dose-time outcome that is interpretable to clinicians and allows for examination of such outcomes over time.