Dr Chatterjee and colleagues performed a meta-analysis comparing thrombolysis with anticoagulation for pulmonary embolism using the Peto method to determine pooled odds ratios (ORs) because of low expected event rates among included trials. We suggest that this method is flawed for several reasons. First, although simulation suggests that the Peto method may outperform standard Mantel-Haenszel OR estimates when event rates are less than1%, the overall risk of mortality after pulmonary embolism is higher (approximately 3%). The Peto method may produce biased estimates with higher events rates. Second, the Peto and other methods ignore trials with zero events, creating another source of bias away from no effect. Third, the Peto fixed-effects model assumes that included studies measured a single underlying effect. It seems unlikely that thrombolytic trials spanning more than 30 years and using different selection criteria and outcome measures would generate 1 treatment effect.
展开▼
