Modeling and prediction of subject accrual and event times in clinical trials: a systematic review

Abstract

Background Modeling and prediction of subject accrual and event times in clinical trials has been a topic of considerable interest for important practical reasons. It has implications not only at the initial planning stage of a trial but also on its ongoing monitoring. Purpose To provide a systematic view of the recent research in the field of modeling and prediction of subject accrual and event times in clinical trials. Methods Two classes of methods for modeling and prediction of subject accrual are reviewed, namely, one that uses the Brownian motion and the other uses the Poisson process. Extensions of the accrual models in multicenter clinical trials are also discussed. Trials with survival endpoints require proper joint modeling of subject accrual and event/lost-to-follow-up (LTFU) times, the latter of which can be modeled either parametrically (e.g., exponential and Weibull) or nonparametrically. Results Flexible stochastic models are better suited when modeling real trials that does not follow constant underlying enrollment rate. The accrual model generally improves as center-specific information is accounted for in multicenter trials. The choice between parametric and nonparametric event models can depend on confidence on the underlying event rates. Limitations All methods reviewed in event modeling assume noninformative censoring, which cannot be tested. Conclusions We recommend using proper stochastic accrual models, in combination with flexible event time models when applicable, for modeling and prediction of subject enrollment and event times in clinical trials.