Objective: In UK Biobank (UKB), a large population-based prospective study, cases of many diseases are ascertained through linkage to routinely collected, coded national health datasets. We assessed the accuracy of these for identifying incident strokes.

Methods: In a regional UKB sub-population (n=17,249), we identified all participants with ≥1 code signifying a first stroke after recruitment (incident stroke-coded cases) in linked hospital admission, primary care or death record data. Stroke physicians reviewed their full electronic patient records (EPRs) and generated reference standard diagnoses. We evaluated the number and proportion of cases that were true positives (i.e. positive predictive value, PPV) for all codes combined and by code source and type.

Results: Of 232 incident stroke-coded cases, 97% had EPR information available. Data sources were: 30% hospital admission only; 39% primary care only; 28% hospital and primary care; 3% death records only. While 42% of cases were coded as unspecified stroke type, review of EPRs enabled a pathological type to be assigned in >99%. PPVs (95% confidence intervals) were: 79% (73%-84%) for any stroke (89% for hospital admission codes, 80% for primary care codes) and 83% (74%-90%) for ischemic stroke. PPVs for small numbers of death record and hemorrhagic stroke codes were low but imprecise.

Conclusions: Stroke and ischemic stroke cases in UKB can be ascertained through linked health datasets with sufficient accuracy for many research studies. Further work is needed to understand the accuracy of death record and hemorrhagic stroke codes and to develop scalable approaches for better identifying stroke types.

ObjectiveLong-term follow-up of population-based prospective studies is often achieved through linkages to coded regional or national health care data. Our knowledge of the accuracy of such data is incomplete. To inform methods for identifying stroke cases in UK Biobank (a prospective study of 503,000 UK adults recruited in middle-age), we systematically evaluated the accuracy of these data for stroke and its main pathological types (ischaemic stroke, intracerebral haemorrhage, subarachnoid haemorrhage), determining the optimum codes for case identification.MethodsWe sought studies published from 1990-November 2013, which compared coded data from death certificates, hospital admissions or primary care with a reference standard for stroke or its pathological types. We extracted information on a range of study characteristics and assessed study quality with the Quality Assessment of Diagnostic Studies tool (QUADAS-2). To assess accuracy, we extracted data on positive predictive values (PPV) and—where available—on sensitivity, specificity, and negative predictive values (NPV).Results37 of 39 eligible studies assessed accuracy of International Classification of Diseases (ICD)-coded hospital or death certificate data. They varied widely in their settings, methods, reporting, quality, and in the choice and accuracy of codes. Although PPVs for stroke and its pathological types ranged from 6–97%, appropriately selected, stroke-specific codes (rather than broad cerebrovascular codes) consistently produced PPVs >70%, and in several studies >90%. The few studies with data on sensitivity, specificity and NPV showed higher sensitivity of hospital versus death certificate data for stroke, with specificity and NPV consistently >96%. Few studies assessed either primary care data or combinations of data sources.ConclusionsParticular stroke-specific codes can yield high PPVs (>90%) for stroke/stroke types. Inclusion of primary care data and combining data sources should improve accuracy in large epidemiological studies, but there is limited published information about these strategies.

Population characteristics and distribution of symptoms, blood tests and primary care consultation patterns in CPRD and UK Biobank.

Details of the different outcome measures considered.

Categorical Features from Question Answers in UK Biobank data fields.

Continuous Features from Sensor Records via Participant Accelerometers.