A machine learning-based prediction model for delirium risk in malnourished elderly ICU patients with SHAP interpretability.

OBJECTIVE

To identify risk factors associated with in-hospital delirium among malnourished elderly patients in the intensive care unit (ICU) and to develop and validate a machine learning-based prediction model for early risk stratification.

METHODS

Using data from a large single-center ICU database (MIMIC-IV) and a multicenter ICU database (eICU-CRD), elderly patients with malnutrition who met predefined inclusion criteria were enrolled. Multiple machine learning models were developed and systematically compared. Model performance was assessed using the area under the receiver operating characteristic curve, calibration curves, decision curve analysis, precision-recall curves, and additional performance metrics. External validation was conducted in an independent cohort to evaluate model generalizability. The final model was further interpreted using SHapley Additive exPlanations (SHAP), and a corresponding prediction tool was constructed.

RESULTS

In total, 6,449 malnourished elderly ICU patients were included. Patients who developed delirium showed significantly higher disease severity, greater physiological instability, and worse clinical outcomes than those without delirium. Among the evaluated models, the eXtreme Gradient Boosting (XGBoost) model achieved the best overall performance in terms of discrimination, calibration, and net clinical benefit, and demonstrated stable predictive ability in the external validation cohort. The final model included seven predictors: Sequential Organ Failure Assessment (SOFA) score, Glasgow Coma Scale (GCS) score, body temperature, peripheral oxygen saturation (SpO2), Geriatric Nutritional Risk Index (GNRI), pH value, and mechanical ventilation. SHAP analysis indicated that disease severity, nutritional risk, and respiratory function-related factors were key contributors to delirium risk.

CONCLUSION

This study developed and externally validated a machine learning-based prediction model for delirium risk in malnourished elderly ICU patients, with good predictive performance and interpretability. The model may aid in early identification of high-risk individuals and support targeted prevention and individualized management of delirium in clinical settings.