Epidemiological estimates regarding stroke prevalence and mortality are often based on ICD-9-CM codes from hospital discharge. The accuracy of such statements is dependent on the codes actually corresponding to the labelled diagnosis. The authors sought to investigate the sensitivity and positive predictive values of ICD-9-CM codes for stroke and intracranial hemorrhage using diagnoses from the Atherosclerosis Risk in Communities (ARIC) study as the gold standard.


The ARIC study is comprised of nearly 16,000 patients in four communities in the US, and the database was searched for hospitalizations for ischemic strokes and intracranial hemorrhage. Strokes were identified by use of a stroke/hemorrhage related ICD-9-CM code, keywords in the discharge summary, or cerebral radiographic findings and validated by both computer algorithm and physician reviewer. Using this group of validated stroke and hemorrhage diagnoses the ICD-9-CM codes were compared.

Looking at AHA/ASA code groupings for ischemic stroke, intracerebral hemorrhage, and subarachnoid hemorrhage they found the positive predictive value to be 76% and sensitivity of 68%. An alternative grouping using more ICD codes increased sensitivity to 83%. These numbers are lower than previously published values and suggest stroke incidence may be underestimated based on ICD-9-CM codes.