Predicting analysis time in event-driven clinical trials with event-reporting lag

For a clinical trial with a time-to-event primary endpoint, the rate of accrual of the event of interest determines the timing of the analysis, upon which significant resources and strategic planning depend. It is important to be able to predict the analysis time early and accurately. Currently available methods use either parametric or nonparametric models to predict the analysis time based on accumulating information about enrollment, event, and study withdrawal rates and implicitly assume that the available data are completely reported at the time of performing the prediction. This assumption, however, may not be true when it takes a certain amount of time (i.e., event-reporting lag) for an event to be reported, in which case, the data are incomplete for prediction. Ignoring the event-reporting lag could substantially impact the accuracy of the prediction. In this paper, we describe a general parametric model to incorporate event-reporting lag into analysis time prediction. We develop a prediction procedure using a Bayesian method and provide detailed implementations for exponential distributions. Some simulations were performed to evaluate the performance of the proposed method. An application to an on-going clinical trial is also described.