Abstract
OBJECTIVE: This study aims to evaluate and compare the predictive performance of various nutritional assessment tools. METHODS: This prospective observational study enrolled 315 elderly patients (≥65 years) scheduled for non-cardiac surgery at Shanxi Medical University First Hospital between March and May 2025. Preoperative data collected included demographics, laboratory indices, Geriatric Nutritional Risk Index (GNRI), Prognostic Nutritional Index (PNI), and Mini Nutritional Assessment (MNA). Postoperative delirium (POD) was diagnosed daily during the 7 days postoperatively using the 3-Minute Diagnostic Confusion Assessment Method (3D-CAM). Patients were stratified into Delirium (n = 54) and non-delirium (n = 249) groups. Logistic regression identified independent POD predictors. Subsequently, Receiver Operating Characteristic (ROC) curve analysis assessed predictive performance (AUC, sensitivity, specificity) of individual tools and combined models. RESULTS: MNA and PNI scores were significantly lower in the delirium group compared to the non-delirium group (p < 0.05), while GNRI scores showed no significant difference. Multivariate analysis identified older age (OR = 1.07, 95% CI: 1.02-1.12), elevated CRP (OR = 1.06, 95% CI: 1.03-1.10), and lower MNA score (OR = 0.79, 95% CI: 0.70-0.88) as independent predictors of POD. ROC analysis revealed the continuous variable of MNA score as the superior single predictor (AUC = 0.741, 95% CI: 0.67-0.81), significantly outperforming PNI (AUC = 0.603, p = 0.008) and GNRI (AUC = 0.442, p < 0.001). The combined model including age, C-reactive protein (CRP), and MNA achieved the highest predictive accuracy (AUC = 0.810, 95% CI: 0.75-0.87; sensitivity 71%, specificity 80%), significantly better than other combinations. Adding PNI or GNRI did not further improve model performance. CONCLUSION: MNA is the most effective standalone nutritional tool for predicting POD in elderly non-cardiac surgery patients. A combined model incorporating age, CRP, and MNA score (AUC = 0.810) shows higher accuracy and improved clinical usefulness for preoperative risk stratification. This allows targeted interventions for high-risk individuals.