Evaluation of admission-level clinical, laboratory, and vascular imaging predictors of prolonged length of stay in acute ischemic stroke

Background: Prolonged hospital length of stay (LOS) in patients with acute ischemic stroke (AIS) adversely affects patient outcomes and strains healthcare resources. Previous studies have focused on clinical determinants but have rarely integrated admission-level laboratory and vascular imaging data with conventional variables.Methods: In this retrospective cohort of AIS patients admitted to a tertiary center between 2017 and 2024, prolonged LOS was defined as hospitalization ≥8 days. Multivariable logistic regression was used to evaluate admission-level clinical, laboratory, and vascular imaging variables and the Trial of Org 10172 in Acute Stroke Treatment (TOAST) classification. Discrimination was assessed using a random forest. Pre-specified sensitivity analyses tested LOS quartiles, a survivor-only model, and calendar-period and pandemic effects.Results: Among 8,448 patients with AIS, 2,696 (31.9%) had prolonged LOS. Twelve admission-level variables and four TOAST subtypes were independent predictors. Positive predictors included initial National Institutes of Health Stroke Scale (NIHSS) score (adjusted odds ratio [aOR], 1.10), coronary heart disease (1.36), previous stroke (1.21), premorbid modified Rankin Scale score (1.08), white blood cell count (1.06), systolic blood pressure (1.03), initial glucose level (1.01), and relevant large-artery steno-occlusion (1.29); in contrast, hemoglobin level (0.95), triglyceride level (0.99), intravenous thrombolysis (0.79), and atrial fibrillation (0.82) showed inverse associations. In comparison with small-vessel occlusion, the other four TOAST subtypes showed prolonged LOS (large-artery atherosclerosis, 1.85; cardioembolism, 2.22; other determined, 2.67; undetermined, 1.78; p<0.001). Logistic regression (area under the curve, 0.736) and random forest (0.723) analyses performed comparably well (p=0.288). The findings persisted across LOS quartiles and survivor-only analysis (n=8,114); the calendar year showed no association (p=0.41), although the pandemic period (2020–2022) reduced the LOS (aOR, 0.89; p=0.028) with other predictors unchanged.Conclusion: Admission-level variables spanning clinical severity, medical history, vascular imaging, laboratory values, and stroke subtype were independently associated with prolonged LOS in patients with AIS and may inform risk stratification in acute stroke care.