Admission glucose-to-potassium ratio and risk of prolonged ICU stay in hemorrhagic stroke: a retrospective cohort study

Abstract Background Hemorrhagic stroke (HS) often necessitates intensive care unit (ICU) admission, yet simple early predictors of ICU length of stay (LOS) remain scarce. Although the glucose-to-potassium ratio (GPR) has shown prognostic value in acute brain injury, its relationship with ICU LOS in HS patients has not been examined. Methods We analyzed 2271 HS patients from the eICU Collaborative Research Database (eICU-CRD), a large multicenter critical care database in the United States. Admission GPR was calculated from first-day glucose and potassium and divided into quartiles. Associations with prolonged ICU LOS, defined as ICU LOS greater than the cohort median (3.6 days), were examined using multivariable logistic regression, Fine–Gray competing risk models, and restricted cubic spline analyses. Subgroup and sensitivity analyses were performed. Results Among 2271 HS patients (mean age 63.5 ± 15.4 years; 51.9% female), higher GPR quartiles were associated with longer ICU stay and higher in-hospital mortality (Q1: 8.3%, Q2: 12.9%, Q3: 23.1%, Q4: 30.6%; P < 0.001). Multivariable logistic regression demonstrated a graded increase in odds of prolonged ICU LOS across quartiles (Q2–Q4 ORs: 1.76, 2.18, and 2.07; P for trend < 0.001), while Fine–Gray competing risk models showed higher GPR was associated with lower subdistribution hazards for prolonged ICU stay (adjusted sHRs: 0.85, 0.76, 0.72; P for trend = 0.003). Restricted cubic spline analysis revealed a non-linear relationship with an inflection point at GPR = 2.208, indicating elevated risk below this threshold. Findings were consistent across subgroup and sensitivity analyses. Conclusions Admission GPR is independently associated with prolonged ICU LOS in HS, exhibiting a non-linear relationship. GPR may serve as a simple biomarker to identify patients at risk for extended ICU stay and support early risk stratification and clinical awareness.