Meta-analysis of NSAIDs: contribution of drugs, doses, trial designs, and meta-analytic techniques.

To elucidate potential bias sources in meta-analyses, I studied whether the effect of nonsteroidal, anti-inflammatory drugs on joint count in patients with rheumatoid arthritis was related to trial design (active or placebo control), sample size, duration of treatment, drop-out rate, the existence of a wash-in period, drug, dose, and meta-analytic technique. In short-term trials, none of the covariates was related to the effect size. No differences between drugs or doses were found in usual gold standard meta-analyses, comparing drugs within trials before pooling. In a meta-analysis of treatment arms, however, four drugs were either significantly more or less effective than average, but these deviations were spurious. In meta-analyses of NSAIDs, indirect comparisons may allow a preliminary comparison of drugs that have not been compared directly. Treatment arms should be compared within trials before pooling, however.