Creating a delirium risk assessment model for geriatric patients with femoral neck fractures through artificial intelligence techniques: a theoretical exploration.

BACKGROUND

Postoperative delirium (POD) is a common occurrence following orthopedic surgery, particularly in the older population. However, there is a relative scarcity of research on the use of intelligent algorithms to predict POD in older patients after orthopedic surgery. Therefore, the objective of this study was to evaluate the efficacy of ten distinct intelligent algorithms in predicting POD in older patients undergoing femoral neck fracture surgery.

METHODS

This study selected ten advanced artificial intelligence algorithms to predict the occurrence of postoperative delirium by analyzing patient data.

RESULTS

A total of 1889 patients were included in this study. The dataset for this study was divided into a training dataset, which consisted of synthetic data, and a testing dataset, representing real-world clinical scenarios. In the training dataset, we identified 267 cases of POD, accounting for 26.70% of the group. In the testing dataset, 172 cases of POD were confirmed, representing 19.35% of the group. Analysis using the Gradient Boosting Decision Tree (GBDT) algorithm revealed that age, preoperative hemoglobin levels, duration of anesthesia, and intraoperative blood loss are key predictive factors for POD in older patients with femoral neck fractures. Among the intelligent algorithms tested for predicting POD in the testing group, logistic regression, random forest, and the Multilayer Perceptron Classifier (MLPC) performed best with accuracy rates of 0.810, 0.810, and 0.808, respectively. In terms of precision, MLPC led with a score of 1.000, followed by random forest (0.714) and logistic regression (0.548). The highest recall rates were achieved by Gaussian Naive Bayes (gnb, 0.337) and AdaBoost (adab, 0.198). Gaussian Naive Bayes also performed best in F1 score (0.244). In the evaluation of the Area Under the Curve (AUC), logistic regression, MLPC, and XGBoost (XGB) demonstrated the best performance, with values of 0.669, 0.669, and 0.652, respectively.

CONCLUSIONS

The results of this study indicate that the Multilayer Perceptron Classifier (MLPC) algorithm performed the most excellently in predicting POD after femoral neck fracture surgery in older adults, with an accuracy rate reaching 80.8%. These findings suggest that machine learning algorithms, particularly MLPC, have significant potential and practical effectiveness in predicting POD in specific older patient populations.

SUPPLEMENTARY INFORMATION

The online version contains supplementary material available at 10.1186/s12883-025-04389-w.