Abstract Body: Introduction: ICD-10 codes are often used for stroke research in administrative databases. Few studies have assessed their reliability in large, representative populations of the United States. We validated ICD-10 codes in a large population-based study of stroke in the Greater Cincinnati/Northern Kentucky (GCNK) region.

Methods: We ascertained all acute strokes in the GCNK region during 2020 using validated methodology. All hospitalizations were screened using a comprehensive list of ICD codes in any diagnosis position (G45-46/H34/I60-69). Additional cases were captured through cold pursuit. Each case was adjudicated by a stroke-trained physician. Only acute ischemic stroke (AIS) and hemorrhagic strokes (HS, defined as intracerebral hemorrhage or subarachnoid hemorrhage) were included in this analysis. We examined how many AIS cases were identified using the standard codes used for administrative studies (G46/I63), assessing their sensitivity and positive predictive value (PPV) in the primary or secondary positions. A similar analysis was conducted using standard HS codes (I60/I61). Differences in the demographics of cases identified with standard codes in the primary position, standard codes in the secondary position, or through other methods were then evaluated with Chi-squared and Kruskal-Wallis tests.

Results: We identified 3,522 AIS/HS events in 2020. For AIS (Table 1), we found that standard codes in the primary position had a sensitivity of 72.8% and a PPV of 89.1%. When the standard AIS codes were considered in any diagnosis position, sensitivity improved to 89.3%, but PPV decreased to 82.3%. For HS, standard codes in the primary position had a sensitivity of 80.0% and a PPV of 81.0%. When the standard HS codes were considered in any position, the sensitivity improved to 93.4% but the PPV decreased to 54.4%. When looking at baseline characteristics (Table 2), patients identified through a standard code in the primary position had a lower baseline mRS (overall P<0.01); nursing home residence was also more common in patients identified through a standard code in the secondary position (7.8%, overall P<0.01).

Conclusion: For AIS, standard codes in the primary position showed only moderate sensitivity but reasonable PPV. For HS, standard codes in the primary position had moderate sensitivity and moderate PPV. Restricting analyses to patients with a standard code in the primary position likely selects against patients with more disability at baseline.