The practice of meta-analysis.

Abstract Meta-analysis seems to have a potentially useful role in carefully selected situations where the primary literature is of good quality, heterogeneity in the response to treatment of the tested population is small and well-understood, interest centers on estimation of a specific, critical parameter of outcome, and the meta-analyst is deeply expert in the subject matter. Other uses can produce, and have produced, results that may be seriously misleading. Five short case studies are presented (diethylstilbestrol and outcome of pregnancy, chlorination of drinking water and cancer, cisplatin and cancer of the ovary, antibiotic therapy for otitis media with effusion, and beta-agonists and asthma).