# Comparative predictive value of preoperative GNRI, PNI, and CONUT for postoperative delirium in geriatric abdominal surgery patients admitted to the ICU

## TL;DR

This study compares three nutritional scores to predict postoperative delirium in elderly ICU patients after abdominal surgery, finding that the CONUT score performs best.

## Contribution

The study directly compares GNRI, PNI, and CONUT for predicting postoperative delirium and identifies CONUT as the superior preoperative predictor.

## Key findings

- CONUT score outperformed PNI and GNRI in predicting postoperative delirium (AUC 0.751 vs. 0.673 and 0.666).
- A clinical prediction nomogram incorporating diabetes, hypoalbuminemia, and low cholesterol showed good discrimination (AUC 0.769).
- Adding CONUT to the nomogram did not significantly improve predictive performance.

## Abstract

Postoperative delirium (POD) is a serious complication in geriatric patients admitted to the ICU following abdominal surgery. Malnutrition is a significant modifiable risk factor for POD, yet the comparative predictive value of established nutritional indices—Geriatric Nutritional Risk Index (GNRI), Prognostic Nutritional Index (PNI), and Controlling Nutritional Status (CONUT)—remains unclear in this high-risk population. This study aimed to directly compare these indices to identify the optimal preoperative predictor for POD.

This single-center retrospective study analyzed 333 patients (≥65 years) admitted post-abdominal surgery to the ICU (from October 2021 to December 2024). POD was diagnosed using CAM-ICU. A clinical prediction nomogram was developed based on significant predictors from the multivariate model. The discriminative ability of preoperative GNRI, PNI, and CONUT scores was compared using receiver operating characteristic (ROC) curves, DeLong’s test for the area under the ROC curve (AUC) differences, along with net reclassification improvement (NRI) and integrated discrimination improvement (IDI) to assess model performance enhancements. Optimal cut-off values were determined by maximizing the Youden index, and corresponding sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and kappa statistics were reported. The study was approved by the Institutional Ethics Committee of Jinling Hospital (Approval No. 2024NZKY-038-02).

Factors identified from multivariable analysis (diabetes mellitus, hypoalbuminemia, reduced total cholesterol) were incorporated into a clinical prediction nomogram, which demonstrated good discrimination (AUC = 0.769, 95%CI: 0.707–0.832, p<0.001) and calibration (Hosmer-Lemeshow test p = 0.444; Brier score = 0.137). Decision curve analysis confirmed its clinical utility. Among the nutritional indices, the CONUT score demonstrated superior predictive performance (AUC = 0.751, 95% CI: 0.686–0.816, p<0.001), significantly outperforming PNI (AUC = 0.673, p<0.001) and GNRI (AUC = 0.666, p<0.001). At an optimal cutoff of 7.5, CONUT achieved 60.9% sensitivity and 81.1% specificity. However, adding CONUT to the clinical nomogram did not significantly improve the predictive performance compared to the clinical model alone (p > 0.05).

We developed a practical nomogram and identified the CONUT score as a valuable preoperative predictor for POD—both demonstrating comparable predictive utility. The CONUT score outperformed PNI and GNRI by integrating key biomarkers (albumin, cholesterol, lymphocytes) into a single metric. Although its components overlap with the clinical model, CONUT offers high specificity and simplicity, making it an efficient tool for rapid preoperative risk stratification.