Aggregate Data Meta-Analysis With Time-To Event Outcomes: A Tutorial

Article type
Year
Authors
Tudur C, Williamson P, Khan S, Sutton R
Abstract
Introduction: Collecting individual patient data (IPD) has been described as the 'gold standard' for undertaking meta-analysis. This involves collecting data on each individual patient entered into each trial from relevant authors, in contrast to an aggregate data meta-analysis (AD) which uses the reported results directly. One advantage of using IPD compared to AD is the ability to undertake time-to-event analyses, however the process can be extremely expensive and time consuming. In certain circumstances undertaking such a large project may not always be a viable option due to unavailability of data or resource constraints, and an AD meta-analysis may be the only practical alternative. The question of the reliability of such an approach is therefore paramount. The process of extracting summary data to use in the meta-analysis is relatively straightforward for studies involving dichotomous or continuous outcomes. However, if studies involve time-to-event outcomes, the process of extracting summary data can be problematic, largely due to inadequate reporting of results. Using several previously reported methods (Parmar et al 1998) to obtain summary statistic estimates, we have undertaken a meta-analysis of nine randomised controlled trials. We illustrate the practicality and problems encountered in undertaking a meta-analysis based on aggregate data with time-to-event outcomes. Various sensitivity analyses are proposed to assess the reliability of the results from and AD meta-analysis.

Objectives:
Methods:
Results:
Discussion: