Sensitivity Analysis for Withdrawals in Grouped Time-to-Event Data

Abstract

This article presents an approach for sensitivity analysis to address the implications of possibly informative censoring for randomized clinical trials with grouped time-to-event data. Various post-withdrawal specifications are applied through a conditional odds ratio (OR) of failure for the discontinued versus retained patients, so that the counts of withdrawals are probabilistically redistributed to the failure outcomes in the time intervals for or after withdrawal or to the censored outcomes at the end of study, as if all the withdrawals completed their follow-up. The hypothetical failure probability estimates and the inferences for treatment effects (i.e., the OR and the Mann–Whitney probability) are produced by matrix operations with the covariance estimators obtained using the linear Taylor series approximations. An illustrative example is provided.

Key Words:

• Informative censoring for life table data
• Intention-to-treat
• Conditional odds ratio sensitivity index
• Probabilistic redistribution for censored outcomes
• Mann-Whitney treatment comparison for life table data
• Odds ratio treatment comparison for life table data