Background: Methods for the meta-analysis of time-to-event data using summary statistics have been available since the late 1990s and their use is becoming more common in systematic reviews. However, several facets of time-to-event data present problems to the reviewer, particularly in older, more poorly reported controlled trials.

Objectives: To summarise in lay terms: (1) the different categories of analysis and reporting errors identified by the Breast Cancer Group's statisticians in a Cochrane review; and (2) the pragmatic solutions on which reviewers and peer reviewers agreed.

Methods: Peer review and discussion between reviewers, peer reviewers and other statisticians took place between May 2004 and October 2005. Errors in the review were identified and solutions negotiated.

Results: A wide variety of errors in analysis and presentation were made by reviewers. Most were easily understood and corrected. Others were less easy to address and, in the main, fell into three categories: (1) 'competing risks' in outcomes where death was not an event; (2) 'informative censoring' where censoring is related to the outcome; and (3) heterogeneity in the definitions of composite outcomes. The first and second are specific to time-to-event analysis; the third is a wider problem but prevalent in cancer studies.

Conclusions: There are limits to the use of summary data in the analysis of time-to-event outcomes. The support of a biostatistician with an interest in time-to-event analysis is invaluable.