Accuracy of healthcare systems data for identifying cardiovascular outcomes after stroke in the UK.

Background Healthcare systems data (HCSD) could improve the efficiency of clinical trials, but their accuracy and validity are uncertain. Our objective was to assess the accuracy of HCSD as the sole method of outcome detection in the REstart or STop Antithrombotics Randomised Trial (RESTART; ISRCTN71907627) compared with adjudicated questionnaire follow-up; and compare estimates of treatment effect. Methods RESTART was a prospective, open, assessor-blind, parallel-group randomized controlled trial (RCT) of antiplatelet therapy after intracerebral haemorrhage (ICH) in the UK. We included 496 (92%) of 537 RESTART participants, who were resident in England or Scotland at randomization. Computerized randomization incorporating minimization allocated participants (1:1) to start or avoid antiplatelet therapy. RESTART used annual questionnaires to detect its primary outcome (recurrent ICH) and secondary outcome (a composite of haemorrhagic or ischemic major adverse cardiovascular events [MACE]) over a median of 2.0 years; an independent adjudication committee verified outcomes using medical records and brain imaging. We obtained ICD10-coded HCSD on hospital admissions and deaths in England and Scotland to identify primary and secondary outcomes. We compared HCSD with a reference standard of adjudicated outcomes. We estimated the effects of antiplatelet therapy using HCSD alone in a Cox proportional hazards model adjusted for minimization variables. Results In the original RESTART trial, 31 people experienced a primary outcome event. HCSD had sensitivity of 84% (95% CI 66% to 95%) and positive predictive value of 68% (51% to 82%) for recurrent ICH. HCSD estimated an effect of antiplatelet therapy (adjusted hazard ratio [aHR] 0.51, 95% CI 0.27 to 0.98; p=0.044) that was almost identical to adjudicated outcomes (aHR 0.51, 95% CI 0.25 to 1.03; p=0.060). HCSD had sensitivity of 84% (76% to 91%) and positive predictive value of 78% (69% to 85%) for MACE, on which HCSD estimated an effect of antiplatelet therapy (aHR 0.81, 95% CI 0.56 to 1.16; p=0.247) that was similar to adjudicated outcomes (aHR 0.65, 95% CI 0.44 to 0.95; p=0.025). Conclusions In a RCT of antiplatelet therapy for people with ICH, HCSD was reasonably accurate and provided similar estimates of treatment effect compared with adjudicated outcomes. Trial registration: RESTART; ISRCTN71907627