Accuracy of Identifying Acute Stroke Admissions in a Michigan Stroke Registry

INTRODUCTION The accurate identification of acute stroke cases is an essential requirement of hospital-based stroke registries. We determined the accuracy of acute stroke diagnoses in Michigan hospitals participating in a prototype of the Paul Coverdell National Acute Stroke Registry. METHODS From May through November 2002, registry teams (ie, nurse and physician) from 15 Michigan hospitals prospectively identified all suspect acute stroke admissions and classified them as stroke or nonstroke. Medical chart data were abstracted for a random sample of 120 stroke and 120 nonstroke admissions. A blinded independent physician panel then classified each admission as stroke, nonstroke, or unclassifiable, and the overall accuracy of the registry was determined. RESULTS The physician panel reached consensus on 219 (91.3%) of 240 admissions. The panel identified 105 stroke admissions, 93 of which had been identified by the registry teams (sensitivity = 88.6%). The panel identified 114 nonstroke admissions, all of which had been identified as nonstrokes by the registry teams (specificity = 100%). The positive and negative predictive value of the registry teams' designation was 100% and 90.5%, respectively. The registry teams' assessment of stroke subtype agreed with that of the panel in 78.5% of cases. Most discrepancies were related to the distinction between ischemic stroke and transient ischemic attack. CONCLUSION The accuracy of hospitals participating in a hospital-based stroke registry to identify acute stroke admissions was very good; hospitals tended to underreport rather than to overreport stroke admissions. Stroke registries should periodically conduct studies to ensure that the accuracy of case ascertainment is maintained.