Development and Validation of an Interpretable Machine Learning Model for Predicting ICU ‐Acquired Weakness in Postoperative Patients

Vollständiger Abstract

Worum geht es in dieser Arbeit?

ABSTRACT Background ICU‐acquired weakness (ICU‐AW) is a common and debilitating complication among critically ill patients, particularly those undergoing major surgery. Early identification of patients at high risk of ICU‐AW may facilitate timely preventive strategies and targeted rehabilitation interventions. Aim To develop and internally validate an interpretable machine learning model for predicting ICU‐AW in postoperative patients admitted to the ICU. Study Design This retrospective study collected data from patients who had previously been admitted to the surgical ICU after surgery. ICU‐AW was defined as a Medical Research Council sum score ≤ 48. Candidate predictors were selected using least absolute shrinkage and selection operator (LASSO) regression. Eight machine learning algorithms were developed and internally validated using a 7:3 random split and 10‐fold cross‐validation. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration curves and decision curve analysis. SHapley Additive exPlanations (SHAP) was applied to interpret model predictions. Results In total, 967 postoperative patients admitted to the surgical ICU were included in the analysis. The overall incidence of ICU‐AW was 24.4%. Five predictors were selected by LASSO: postoperative delirium, interleukin‐6, Barthel index, sepsis and Sequential Organ Failure Assessment (SOFA) score. Among the eight algorithms, the random forest model achieved the best discriminative performance in the validation cohort (AUC 0.788, 95% CI: 0.727–0.849) and demonstrated good calibration and clinical utility. SHAP analysis identified Barthel index, SOFA score and postoperative delirium as the top contributors to the model's predictions and provided individualised explanations of risk estimates. Conclusions This study developed an interpretable model using five variables to predict ICU‐AW in surgical ICU patients, which demonstrated good performance, predictive value and clinical utility. A user‐friendly web‐based tool was developed to support individualised risk assessment and enhance clinical applicability. Relevance to Clinical Practice The proposed model may assist clinicians in early risk stratification of postoperative ICU patients and facilitate targeted surveillance and preventive care for those at high risk of ICU‐AW.