Initial stroke severity is a significant predictor of short and longer-term outcomes, disease burden, and mortality. Stroke severity is commonly assessed using the National Institutes of Health Stroke Scale (NIHSS). Despite its known utility for predicting outcomes, NIHSS availability remains low. Thus, there is a significant need to develop predictive models of stroke severity using electronic health record (EHR) data. Funded by HSR&D, the purpose of this study was to develop and validate a predictive model of initial stroke severity using EHR data elements.
Investigators in this observational study used VA’s EHR data to identify 15, 346 Veterans who had been admitted to a VA medical center between October 1, 2018, and July 31, 2022, with a discharge diagnosis of ischemic stroke. They then extracted 65 independent predictors from the EHR data. The primary analysis modeled mild (NIHSS score 0-3) versus moderate/severe stroke (NIHSS score ≥4) using multiple logistic regression. Study findings showed that the primary model correctly classified 70% of patients.
In conclusion, EHR data can be used to create a proxy of initial stroke severity in the absence of a documented NIHSS score or ICD-CM code. Further, the predictive model had acceptable performance when discriminating mild vs moderate/severe stroke—and showed that receiving endovascular treatment, transferring to another hospital facility within two days of admission, and the presence of hemiplegia were the key variables in this discrimination. Further work is needed to examine how to better differentiate moderate and severe stroke severity using EHR data elements.
Waddell K, Myers L, Perkins A, et al. Development and validation f a model predicting mild stroke severity on admission using electronic health record data. Journal of Stroke and Cerebrovascular Diseases. September 2023;32(9):107255.
