Objective: Assess how standardization of relative risks (RRs) and standard errors (SEs), according to blood pressure differences within trials, affects heterogeneity, overall effect estimates and study weights in meta-analyses of antihypertensive treatment.
Method: Data from a previous systematic review were used. Three sets of analyses were performed, using both random-effects and fixed-effects model for meta-analyses. First, we used raw data from the included trials. Second, we standardized RRs as if SBP was reduced by 10 mmHg in all trials. Third, we standardized both RRs and SEs.
Results: When RRs were standardized according to blood pressure lowering, heterogeneity between trials increased (I(2) = 36 vs. 93% for mortality). This conferred large differences in treatment effect estimates using random-effects and fixed-effects model (RR 0.79, 95% confidence interval 0.70-0.89, respectively, 0.97, 0.94-0.99). When SEs were standardized, confidence intervals for individual trials widened, resulting in lower power to detect heterogeneity across trials. Study weights were dissociated from number of events in trials (P < 0.0001, R(2) = 0.99 before standardization vs. P = 0.063, R(2) = 0.05 after standardization). This induced a secondary shift in weight from trials with lower baseline SBP to trials with higher baseline SBP, resulting in exaggerated overall effect estimates.
Conclusion: Standardization of RRs exaggerates differences between trials and makes meta-analyses highly sensitive to choice of statistical method. Standardization of SEs masks heterogeneity and results in biased effect estimates.