Co-occurring leukocytosis and hyperglycemia identify patients at high risk of subsequent mortality after intracerebral hemorrhage: a 24-h landmark analysis in MIMIC-IV and eICU

Background White blood cell (WBC) count and glucose levels may jointly reflect physiological stress after intracerebral hemorrhage (ICH). We examined their associations with subsequent in-hospital mortality and whether they improved prediction beyond established clinical information. Methods This retrospective study used the Medical Information Mart for Intensive Care IV (MIMIC-IV) and the eICU Collaborative Research Database. Adults with ICH identified by diagnosis codes who were alive and hospitalized 24 h after intensive care unit admission were included. WBC > 10 × 10 9 /L and glucose >140 mg/dL defined four groups: neither elevated, WBC elevation only, glucose elevation only, and both elevated. Logistic regression estimated adjusted associations, with bias reduction in the eICU cohort. Restricted cubic splines assessed nonlinear associations. Prediction models used the minimum first-day Glasgow Coma Scale score alone or a clinical baseline model that also included age, sex, blood pressure, and other laboratory measurements. Models were developed on MIMIC-IV and evaluated without updating in the eICU database. Results The MIMIC-IV and eICU landmark cohorts included 1,606 and 264 patients, with 332 and 42 subsequent deaths, respectively. Adjusted odds ratios for both markers elevated versus neither elevated were 2.98 (95% confidence interval [CI], 2.04–4.36) in MIMIC-IV and 3.34 (95% CI, 0.97–11.54) in eICU. The glucose association in MIMIC-IV was nonlinear ( p = 0.005), whereas the WBC curve varied with the spline specification. Among 229 eICU patients with both markers measured, adding the joint category to the clinical baseline model changed the area under the receiver operating characteristic curve from 0.796 to 0.811 (difference, 0.014; 95% CI, −0.012 to 0.039) and the Brier score, a measure of prediction error, by −0.0017 (95% CI, −0.0075 to 0.0041). Conclusion Joint WBC–glucose elevation was associated with subsequent in-hospital mortality in MIMIC-IV, but estimates in the eICU cohort were imprecise. Additional predictive value remained uncertain, and continuous analyses did not establish a stable U-shaped association.