Development and External Validation of a Propensity Score-Matched Predictive Model for 48-Hour Intensive Care Unit...

This study aimed to develop and validate a clinically applicable predictive model for estimating the probability of intensive care unit (ICU) readmission within 48 hours following ICU discharge in patients with sepsis. The model's predictive performance was evaluated across development and validation cohorts.

Clinical data from patients with sepsis-classified according to 48-hour ICU readmission status-were extracted from the Medical Information Mart for Intensive Care IV database using structured query language. Propensity score matching was applied to balance baseline covariates and reduce confounding between comparison groups. Candidate variables were identified through univariate analysis and refined using least absolute shrinkage and selection operator regression. A logistic regression-based predictive model was subsequently constructed and validated using an independent dataset.

The predictive model was developed using clinical data from 1002 patients and validated in an independent external cohort of 100 patients. The final model incorporated 6 predictors, including the 24-hour serum albumin level at ICU admission, 24-hour activated partial thromboplastin time, antibiotic use, mechanical ventilation, heart rate within 24 hours prior to ICU discharge, and the Acute Physiology Score III. The model demonstrated robust predictive performance, with C-index values of 0.82 in the development cohort, 0.81 in the internal validation cohort, and 0.76 in the external validation cohort.

The predictive model demonstrated reliable performance in estimating the probability of ICU readmission within 48 hours among patients with sepsis following ICU discharge. The variables incorporated into the model are routinely collected in clinical practice, supporting its feasibility for early risk stratification and targeted interventions aimed at reducing early ICU readmission rates.