Prediction of Event Time for A Time-to-Event Endpoint Under A Piecewise Exponential Model

• Activity Number: 573
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Liang Chen*+ and Xiaoyu Dong
• Companies: Pfizer Inc. and MedImmune
• Keywords: time-to-event ; predicting analysis time ; interim analysis ; piecewise exponential model

Abstract

Randomized clinical trials commonly include one or more planned interim analyses. With a time-to-event endpoint, timing of interim analysis or final analysis is usually event-driven. Because these analyses involve timeline and resource planning, it is worthwhile to predict timing of these analyses early and accurately. Parametric models on observed event data from on-going clinical trials were used in Bagiella and Heitjan (2001) and Ying and Heitjan (2008). However, parametric model may not be flexible enough to fit the real data well. We propose the piecewise exponential (PE) model to fit the observed event data and estimate the parameters of the model, then predict timing of interim analysis or final analysis. PE model is quite flexible, and can fit most of time-to-event data quite well. Accuracy of prediction is assessed by simulation studies by comparing the performance of exponentia