Interrater reliability and accuracy in identifying ischemic strokes using computed tomography scans in people with dementia

OBJECTIVES: To assess the reliability and interobserver agreement of stroke identification on neuroimaging in patients presenting with dementia. DESIGN: Comparison study between neurologists, radiology reports, and autopsy. SETTING: Dementia registry within a health maintenance organization. PARTICIPANTS: Dementia patients with computed tomography (CT) scans obtained near the time of diagnosis and postmortem neuropathological examinations (N=99). MEASUREMENTS: Three neurologists independently read CT scans for the presence and locations of strokes. Radiology reports from these scans were reviewed. The results from neurologists, radiologists, and autopsies were compared. RESULTS: The positive predictive value for CT-observed strokes compared with their presence on autopsy was 0.44 to 0.49, regardless of the specialty of the observer. Strokes were present at autopsy in 46 of 99 cases. Agreement between neurologists on the presence of strokes was fair to moderate (kappa=0.27-0.56). Less agreement was found between neurologists and radiologists (kappa=0.00-0.11). Results improved slightly when each case was evaluated as any stroke present versus no stroke on imaging (kappa=0.34-0.75) or for the presence of multiple strokes (kappa=0.17-0.69). CONCLUSION: There is only fair to moderate agreement between observers regarding the identification of strokes on CT scans in patients presenting with dementia. Furthermore, strokes identified on imaging were present on pathology only half the time.