Why does primary angioplasty not work in registries? Quantifying the susceptibility of real-world comparative effectiveness data to allocation bias.

BACKGROUND: Meta-analysis of registries (comparative effectiveness research) shows that primary angioplasty and fibrinolysis have equivalent real-world survival. Yet, randomized, controlled trials consistently find primary angioplasty superior. Can unequal allocation of higher-risk patients in registries have masked primary angioplasty benefit? METHODS AND RESULTS: First, we constructed a model to demonstrate the potential effect of allocation bias. We then analyzed published registries (55022 patients) for allocation of higher-risk patients (Killip class ≥1) to determine whether the choice of reperfusion therapy was affected by the risk level of the patient. Meta-regression was used to examine the relationship between differences in allocation of high-risk patient to primary angioplasty or fibrinolysis and mortality. Initial modeling suggested that registry outcomes are sensitive to allocation bias of high-risk patients. Across the registries, the therapy receiving excess high-risk patients had worse mortality. Unequal distribution of high-risk status accounted for most of the between-registry variance (adjusted R(2)(meta)=83.1%). Accounting for differential allocation of higher-risk patients, primary angioplasty gave 22% lower mortality (odds ratio, 0.78; 95% confidence interval, 0.64-0.97; P=0.029). We derive a formula, called the number needed to abolish, highlighting situations in which comparative effectiveness studies are particularly vulnerable to this bias. CONCLUSIONS: In ST-segment elevation myocardial infarction, clinicians' preference for management of a few high-risk patients can shift mortality substantially. Comparative effectiveness research in any disease is vulnerable to this, especially diseases with an immediately identifiable high-risk subgroup that clinicians prefer to allocate to 1 therapy. For this reason, preliminary indications from registry-based comparative effectiveness research should be definitively tested by randomized, controlled trials.