Machine learning-based prediction of delirium in older patients with chronic kidney disease requiring intensive care: A hospital-based retrospective cohort study.

OBJECTIVES

Delirium is a common complication in intensive care units (ICUs), especially among older adults with chronic kidney disease (CKD). It is associated with increased mortality and prolonged hospitalization. Machine learning (ML)-based models can help predict delirium. In this study, we developed an ML-based delirium prediction model for critically ill older patients with CKD.

METHODS

This retrospective cohort study included patients aged ≥65 years admitted to the ICU between January 2021 and November 2023. Delirium was assessed every 8 h throughout the ICU stay using the Intensive Care Delirium Screening Checklist (ICDSC) and defined as a score of ≥4. Eight ML models were compared in terms of the area under the receiver operating characteristic curve (AUROC), accuracy, F1-score, specificity, and recall.

RESULTS

This study included 895 patients, of whom 55.3 % developed delirium. The random forest model outperformed others (F1-score: 0.891; specificity: 0.911; recall: 0.864; accuracy: 0.885; precision: 0.923; AUROC: 0.950). Backward feature selection achieved an F1-score of 0.892 and an AUROC of 0.953, identifying 14 key predictors of delirium: physical restraint use, low Glasgow Coma Scale score, high white blood cell count, hypernatremia, fever, advanced age, hypoalbuminemia, hypercalcemia, low hemoglobin, high Acute Physiology and Chronic Health Evaluation II (APACHE II) score, hyperkalemia, mechanical ventilation, sedation, and elevated C-reactive protein.

CONCLUSION

Our ML model demonstrated good performance in predicting delirium in critically ill older adults with CKD, suggesting potential value for early identification. Future studies may explore integration into hospital systems to support delirium prevention strategies.