# Development and validation of a postoperative delirium risk prediction model for non-cardiac surgery in elderly patients

## TL;DR

This study developed a model to predict the risk of postoperative delirium in elderly patients undergoing non-cardiac surgery, using factors like age, sleep quality, and medical history.

## Contribution

The study introduces a novel prediction model incorporating sleep quality and postoperative pain as risk factors for delirium.

## Key findings

- 131 out of 663 patients (19.76%) developed postoperative delirium.
- The model achieved high predictive accuracy with an AUC of 0.939, sensitivity of 94.44%, and specificity of 85.09%.
- Key predictors included age, MMSE score, diabetes history, sleep quality, and anesthesia duration.

## Abstract

Postoperative delirium (POD) is one of the common central nervous system complications in elderly patients after non-cardiac surgery. Therefore, it is necessary to develop and validate a preoperative model for POD risk prediction.

This study selected 663 elderly patients undergoing non-cardiac elective surgery under general anesthesia for tracheal intubation in general surgery, from September 1st, 2020 to June 1st, 2022. Simple random sampling method was used according to 7: 3. The occurrence of POD within 1 to 7 days after the operation (or before discharge) was followed up by the confusion assessment method (CAM). This study innovatively included the pittsburgh sleep quality index (PSQI) and the numerical pain score (NRS) for clinical work, to explore the relationship between sleep quality and postoperative pain and POD. Univariate and Multivariable Logistic regression analysis was used to analyze stepwise regression to screen independent risk factors for POD. The creation of prediction models involved the integration of outcomes through the implementation of logistic regression analysis. In addition, internal validation is employed to ensure the reproducibility of the model.

A total of 663 elderly patients were enrolled in this study, and 131 (19.76%) patients developed POD. The incidence of POD in each department was not statistically significant. The predictors in the POD column line graph included age, Mini Mental State Examination (MMSE) score, history of diabetes, years of education, sleep quality index, ASA classification, duration of anesthesia and NRS score. The formula Z= 8.293 + 0.102 × age - 1.214 × MMSE + 1.285 × diabetesHistory - 0.304 × yearsOfEducation + 0.602 × PSQI + 1.893 × ASA + 0.027 × anesthesiaTime + 1.297 × NRS. Conducive to the validation group to evaluate the prediction model, the validation group AUC is 0.939 (95% CI 0.894-0.969), the sensitivity is 94.44%, and the specificity is 85.09%. The calibration curves show a good fit between the clinically predicted situation and the actual situation.

The clinical prediction model constructed based on these independent risk factors has a good predictive performance, which can provide reference for the early screening and prevention of POD in clinical work.

ChiCTR2000033639 Retrospectively registered (date of registration: 06/07/2020).