OBJECTIVES: Is it enough to check for statistical heterogeneity before plunging into a meta-analysis? For example, there are many clinical trials of rhythm control drugs for atrial fibrillation (AF), suggesting a meta-analysis is the way to go. Is it? METHODS: A MEDLINE search yielded 347 studies, of which 20 met strict inclusion criteria. Only randomized, blinded studies of adults with ECG-confirmed AF or atrial flutter recurrence were considered. Studies had to compare antiarrhythmic agents or antiarrhythmics with placebo. Crossover studies and studies of AF after cardiac surgery were excluded. RESULTS: After detailed review, we decided that, in contrast with a published systematic review, meta-analysis was not appropriate due to intractable heterogeneity in study designs. Although recurrence risk differs substantially between chronic and paroxysmal AF, trials mixed these types in different (and even unstated) proportions. Some (n=12) required sinus rhythm for various durations at randomization, increasing the chance of success; 4 ignored early recurrences or allowed for recardioversion, while 16 counted every recurrence. In addition, methods used to detect recurrence varied from ECG confirmation at regularly scheduled visits to much more sensitive trans-telephonic monitoring (n=11), which was employed in differing ways. Furthermore, efficacy endpoints were affected by biased termination of follow-up, as treatment withdrawal due to side effects leads to less detection of recurrence. Nevertheless, a published meta-analysis ignored these differences, citing similar recurrence rates in control groups and chi-squared tests that did not detect statistical heterogeneity. In our opinion, this does not adequately address heterogeneity in study designs, and the resulting estimates are unsuitable for economic modeling. CONCLUSION: Strict adherence to frequentist views can lead to inappropriate pooling of trial data and erroneous inputs for economic models. A Bayesian perspective provides a more correct view of the intractable study design differences.