Introduction: In randomised controlled trials, where the outcome of interest is dichotomous, event rates in control and treatment groups may vary according to the length of follow up (LFU). LFU has a direct influence on baseline rates and also may affect statistics used to summarise treatment differences, such as relative risks, odds ratios and risk differences. LFU may explain increased heterogeneity in meta-analyses (MA) which combine trials of mixed follow-up, and introduces problems in the calculation of the number needed to treat (NNT), which is specific to a stated LFU.

Objective: To assess the quality of information on LFU in Cochrane Reviews and provide a description of impact of LFU on the calculation of NNTs.

Methods: Systematic reviews published on the Cochrane Database of Systematic Reviews (1998 Issue I) were considered eligible for inclusion if they reported mortality as the main outcome. MAs of mortality related to childbirth were excluded. The relationship between follow-up, baseline rate and relative risk will be investigated using meta-regression and weighted least squares regression models.

Results: 45 out of 326 systematic reviews, were considered eligible for our investigation. 18 MAs (40%) reported no information on follow-up at all, a further 18 (40%) had incomplete information, whilst 9 (10%) specified a summary statistic (median, mean, range, other) for the follow-up in each trial. Only one provided complete information for all studies. Various models of the relationships between LFU, baseline rate and treatment effect are proposed and assessed in a sample of reviews.

Discussion: Few reviews pay adequate attention to this variable: reviewers should be more aware of its importance. Correct assessment of the relationship between LFU, baseline rates and treatment effects is crucial in the calculation of NNTs.