Validating national registry data for randomized trials: comparison of registry-derived cardiovascular events with and without adjudication in the BETAMI-DANBLOCK trial

BACKGROUND: Registry-based randomised clinical trials are increasingly applied in clinical research, providing advantages in feasibility and data capture. However, registry-based data may introduce bias through misclassification or missing information. METHODS: In the BETAMI-DANBLOCK trial, 5,574 patients with myocardial infarction (MI) and no heart failure were randomised to beta-blocker or no beta-blocker therapy. The primary endpoint events (all-cause mortality, MI, ischemic stroke, heart failure, unplanned coronary revascularisation, and malignant ventricular arrhythmias) were identified from the Danish/Norwegian national patient registries, self-reported questionnaires, and medical records. All registry-identified events, except all-cause mortality, underwent blinded adjudication. We compared non-adjudicated and adjudicated events by calculating incidence rates and hazard ratios (HRs) for beta-blocker therapy versus controls. RESULTS: National registries captured 99.7% of primary endpoint events registered by self-report and medical records. Of the primary endpoint events identified through the registries, 75% were confirmed by adjudication (incidence rate 6.23 versus 4.43/100 person-years). The confirmation rate was lower during the first six months and varied by event type (from 92% for ischemic stroke to 45% for unplanned coronary revascularisations and heart failure). Estimated treatment effects were consistent for the primary endpoint with non-adjudicated and adjudicated events (HR 0.88, 95% confidence interval (CI): 0.78-0.98 and HR 0.85, 95% CI: 0.75-0.98, respectively), with no apparent differences across event types except for heart failure and revascularisation. CONCLUSION: Registry data provided effect estimates of beta-blockers comparable to adjudicated data. However, adjudication substantially reduced the number of cardiovascular endpoint events and event-specific misclassification appeared, particularly during the first months after the index-event.