OBJECTIVE: To assess how reliable UK routine data are for ascertaining major bleeding events compared with adjudicated follow-up. METHODS: The ASCEND (A Study of Cardiovascular Events iN Diabetes) primary prevention trial randomised 15 480 UK people with diabetes to aspirin versus matching placebo. The primary safety outcome was major bleeding (including intracranial haemorrhage, sight-threatening eye bleeding, serious gastrointestinal bleeding and other major bleeding (epistaxis, haemoptysis, haematuria, vaginal and other bleeding)) ascertained by direct-participant mail-based follow-up, with >90% of outcomes undergoing adjudication. Nearly all participants were linked to routinely collected hospitalisation and death data (ie, routine data). An algorithm categorised bleeding events from routine data as major/minor. Kappa statistics were used to assess agreement between data sources, and randomised comparisons were re-run using routine data. RESULTS: When adjudicated follow-up and routine data were compared, there was agreement for 318 major bleeding events, with routine data identifying 281 additional-potential events, and not identifying 241 participant-reported events (kappa 0.53, 95% CI 0.49 to 0.57). Repeating ASCEND's randomised comparisons using routine data only found estimated relative and absolute effects of allocation to aspirin versus placebo on major bleeding similar to adjudicated follow-up (adjudicated follow-up: aspirin 314 (4.1%) vs placebo 245 (3.2%); rate ratio (RR) 1.29, 95% CI 1.09 to 1.52; absolute excess +6.3/5000 person-years (mean SE±2.1); vs routine data: 327 (4.2%) vs 272 (3.5%); RR 1.21, 95% CI 1.03 to 1.41; absolute excess +5.0/5000 (±2.2)). CONCLUSIONS: Analyses of the ASCEND randomised trial found that major bleeding events ascertained via UK routine data sources provided relative and absolute treatment effects similar to adjudicated follow-up. TRIAL REGISTRATION NUMBER: ISRCTN60635500; NCT00135226.
atherosclerosis, electronic health records, outcome assessment, health care, research design