Advanced models for predicting event occurrence in event-driven clinical trials accounting for patient dropout, cure and ongoing recruitment

TL;DR

This paper introduces advanced statistical models for predicting event occurrence in clinical trials, accounting for patient dropout, cure, and ongoing recruitment, supported by software and validated through simulations and real data.

Contribution

It develops new analytic models incorporating cure, dropout, and recruitment dynamics in event-driven clinical trials, with implementation in software.

Findings

Models effectively predict event counts and timing.

Software implementation supports practical application.

Validated with simulation and real clinical trial data.

Abstract

We consider event-driven clinical trials, where the analysis is performed once a pre-determined number of clinical events has been reached. For example, these events could be progression in oncology or a stroke in cardiovascular trials. At the interim stage, one of the main tasks is predicting the number of events over time and the time to reach specific milestones, where we need to account for events that may occur not only in patients already recruited and are followed-up but also in patients yet to be recruited. Therefore, in such trials we need to model patient recruitment and event counts together. In the paper we develop a new analytic approach which accounts for the opportunity of patients to be cured, as well as for them to dropout and be lost to follow-up. Recruitment is modelled using a Poisson-gamma model developed in previous publications. When considering the occurrence of events, we assume that the time to the main event and the time to dropout are independent random variables, and we have developed a few advanced models with cure using exponential, Weibull and log-normal distributions. This technique is supported by well developed, tested and documented software. The results are illustrated using simulation and a real dataset with reference to the developed software.