Identification of stroke in the community: A comparison of three methods

Background: Evidence concerning secondary prevention of cerebrovascular disease is not optimally used in clinical practice. A necessary first step is to identify those eligible for treatment. In primary care, this equates to setting up a register of prevalent stroke. Aim: To compare three different methods for identifying prevalent cases of cerebrovascular disease in the community: general practice-based computer systems; population surveys; and hospital-based routine information systems. Design of study: Comparison of results of each method applied to a defined population and then assessed against reference criteria for cerebrovascular disease. Setting: A total of 5801 people aged 65 years or over, resident in seven practices situated within the South Birmingham Primary Care Trust area. Method: The sensitivity, specificity, and predictive value of each method of identification were calculated against reference criteria applied by two investigators independently of each other. Results: The prevalence of reference criteria-validated cerebrovascular disease in patients aged 65 years or over was 8.2%. Overall, general practice-based computer systems had a sensitivity of 81.0%, a specificity of 97.2% and a positive predictive value (PPV) of 71.8%, but there was a wide range of sensitivity (33% to 90%) and PPV (42% to 92%) between practices. Patient survey and hospital information systems were less sensitive (75.7% and 28.4%, respectively) but had higher PPVs (77.5% and 89.2%, respectively). Thirty-nine per cent of patients with a history of cerebrovascular disease had not been admitted to hospital. Conclusion: General practice-based computer systems can produce reasonably accurate prevalent stroke registers. In areas where these are poorly developed, patient survey is an alternative.