Prediction model for postoperative delirium risk in elderly hypertensive patients: machine learning-based development...

Area of Science:

• Geriatric Medicine
• Cardiology
• Artificial Intelligence in Healthcare

Background:

• Postoperative delirium (POD) is a significant complication in elderly hypertensive patients, linked to adverse long-term outcomes.
• Current predictive models often require intraoperative data, hindering early preoperative risk assessment.
• This study addresses the need for non-invasive, preoperative risk stratification for POD.

Purpose of the Study:

• To develop and validate a machine learning model for predicting POD in elderly hypertensive patients using preoperative data.
• To identify key preoperative markers that predict POD occurrence.
• To investigate the impact of these preoperative markers on three-year mortality in patients who develop POD.

Main Methods:

• A cohort of 1,782 patients was analyzed, with preoperative variables selected via LASSO regression.
• Ten machine learning models were trained and validated, with performance assessed using AUC-ROC and DCA.
• The optimal model was interpreted using SHAP values, and long-term survival was analyzed using Kaplan-Meier curves and Cox regression.

Main Results:

• The incidence of POD was 10.9%.
• The Gradient Boosting Machine (GBM) model achieved the highest performance (AUC = 0.868).
• Key predictors identified by SHAP analysis included MMSE, HADS, CFS, frailty, and PSQI scores. Lower MMSE and higher HADS, CFS, frailty, and PSQI scores independently predicted increased three-year mortality in POD patients.

Conclusions:

• A reliable machine learning tool for individualized POD prediction was developed.
• Cognitive impairment, psychological distress, frailty, and poor sleep quality are crucial dual-prognostic markers for POD and long-term survival.
• Multidimensional preoperative assessment is essential for personalized interventions in vulnerable hypertensive populations.