Common statistical errors in the design and analysis of subfertility trials

BACKGROUND: The quality of clinical trials has received increasing attention with the growth of evidence-based medicine and systematic reviews. We aimed to identify whether errors and omissions commonly encountered when undertaking Cochrane reviews in this field are still passing peer review. METHODS: We undertook a review of trials published in 2001 by two major journals. We selected from Medline only trials in which authors compared pregnancy rates under two interventions by allocating women to different groups. RESULTS: We identified 39 trials meeting our criteria. Six trials were fatally flawed by design, either by inappropriate use of a cross-over design or by systematic allocation described by the authors as 'random'. Only six reports claimed to apply the intention-to-treat principle, and the principle was misunderstood by four of these. Only five trials reported live birth rates sufficiently to allow valid meta-analysis. Most trials (82%) included at least one 'unit of analysis' error. CONCLUSIONS: We selected simple trials from respected journals, assuming that our sample would represent trials of highest methodological quality in the field. Nevertheless, the standards of design, analysis and reporting of many subfertility trials are not sufficient to allow reliable interpretation of results, or inclusion in meta-analyses.