Abstract TP69: Prediction Performance Of Socio-economic And Clinical Factors Of Prolonged Acute Ischemic Stroke Hospitalization

Introduction: Patients with acute ischemic stroke (IS) experience varying length of stay (LOS), which is a key hospital outcome linked with resource allocation and cost. We investigated predictive performance of socio-economic and clinical factors of prolonged LOS in the National Inpatient Sample (NIS) data set. Methods: We performed a retrospective analysis on the NIS data set containing patients with an IS diagnosis (I63*) from years 2016 to 2019. We investigated differences between patients with a LOS 14 ≥ vs < 14 days. We utilized a LASSO regression to select predictors of LOS 14 ≥ vs < 14 prior to a ROC analysis. We included: age, sex, race/ethnicity, hypertension, Diabetes II, Atrial fibrillation, congestive heart failure, NIHSS categories, insurance payer, income, teaching hospital and bed size, thrombolytic therapy, mechanical thrombectomy (MT), intubation, tracheostomy, and percutaneous endoscopic gastrostomy (PEG) tube. Results: The cohort (n=137,370; 51% male) had a mean (SD) age of 68.39(± 14.01) years and admitted NIHSS of 6.32 (± 6.81). All differences between LOS 14 ≥ vs < 14 were significant (p ≤ 0.021). Groups were similar in age (66.31 ± 13.91 vs 69.50 ± 13.99), but differed in metrics of stroke severity including NIHSS > 15 (41.61% vs 10.55%), MT (21.04% vs 6.40%), intubation (23.4% vs 1.60%), tracheostomy (11.54% vs 0.11%), and PEG tube (34.06% vs 1.89%). They were more likely to have Medicaid (19.74% vs 8.78%) and identify as Black (27.86% vs 16.93%). LASSO regression identified 13 predictive variables (Pseudo-R2 0.264) with an ROC AUC of 0.84 discriminating between LOS ≥ 14 vs $lt 14 ( Figure) . Conclusion: Independent of stroke severity and procedure, Medicaid insurance and identifying as black were significant predictors of prolonged acute stroke hospitalization. Although the ROC curve was highly accurate in identifying prolonged LOS, the LASSO only accounted for 26% of the variance. Further research into novel predictors of LOS is warranted.