Meta-analyses involving cross-over trials: methodological issues.
Elbourne DR., Altman DG., Higgins JPT., Curtin F., Worthington HV., Vail A.
BACKGROUND: Meta-analysis of randomized controlled trials (RCTs) is usually based on trials where patients are randomized individually into two different, parallel, treatment groups. This paper concentrates on RCTs of a different design-two-period, two-treatment cross-over trials. METHODS: The characteristics of these trials are outlined, with detailed examples of methods for analysis for both continuous and binary data. These case studies are then extended into the context of a meta-analysis. The Cochrane Library was surveyed to assess current practice for synthesis. RESULTS: Methods are described for continuous and binary data for use both when the necessary paired data are given and also when they need to be calculated or imputed, and some suggestions are provided to help people wishing to synthesize data from cross-over trials into meta-analyses. The survey suggested that about 8% of the trials in the Cochrane library were cross-over trials and 18% of the reviews referred to such trials, although there was no consistent approach to their inclusion into the reviews. CONCLUSIONS: Methods do exist for including valuable information from two-period, two-treatment cross-over trials into quantitative reviews. However, poor reporting of cross-over trials will often impede attempts to perform a meta-analysis using the available methods.