Abstract Body: Background: The vascular region of a stroke is reported in International Classification of Diseases-10 (ICD10) subcodes as either a specific or nonspecific/unspecified location. The coding has been shown to be accurate for specific ischemic stroke regions but has not been assessed for nonspecific/unspecified coding. The overall accuracy has implications for research. We aimed to describe the imaging characteristics in strokes coded with nonspecific/unspecified region subcodes and to assess for associations with their use.
Methods: From a single comprehensive stroke center, we randomly selected 200 stroke hospitalizations with a primary discharge I63 ICD10 code; 100 had non-specific/unspecified vascular location subcodes, and 100 had specific vascular location subcodes oversampled for posterior circulation strokes. Gold-standard scoring of the vascular region was performed using the imaging studies and reports blinded to the subcode. The NIHSS was categorized using R29 codes (i.e. missing, mild, moderate, severe). We used logistic regression to assess the association of potential predictors with nonspecific/unspecified ICD10 vascular region subcodes weighted for the sampling.
Results: In the 100 strokes with nonspecific/unspecified region subcodes, mean NIHSS was 4.0 (SD 4.2), 88% had a head CT and 93% had a MRI. The most commonly used non-specific/unspecified codes were I63.9 (48%), I63.89 (20%), and I63.81 (11%). The gold standard classification of vascular region identified an acute infarct in a defined location in 85 of the 100 cases that had non-specific/unspecified subcodes: 45 anterior circulation strokes (27 MCA, 9 ACA , 9 carotid), 40 posterior circulation strokes (17 PCA, 11 basilar, 6 cerebellar, 6 vertebral), and 15 with no infarct on imaging. In the model (c-statistic 0.63), NIHSS category (severe stroke vs missing: OR 0.08; 95% CI:0.01-0.49), no gold standard acute lesion (OR 10.8; 95% CI:1.0-115.9), and MRI performed (OR 3.25; 95% CI:1.0-10.4) were associated with nonspecific/unspecified ICD10 subcodes but posterior versus anterior circulation location of infarct was not (OR 0.88; 95% CI: 0.41-1.8).
Conclusions: We found that most stroke hospitalizations with a nonspecific/unspecified vascular region subcode have acute lesions of a specific region on imaging and were more likely to have low NIHSS. These findings suggest that selecting for specific stroke locations based on ICD10 codes could bias studies toward more severe strokes in that region.