The Journal of Practical Medicine ›› 2026, Vol. 42 ›› Issue (17): 3119-3126.doi: 10.3969/j.issn.1006-5725.2026.17.010

• Treatise：Mechanism and Practice • Previous Articles     Next Articles

Prediction model for postoperative delirium in elderly patients undergoing spinal surgery based on various machine learning methods

Yunyun Yu1, Fang Zheng1, Weimin Ding1, Jiawei Fang2, Qing Yang3, Chuanguang Wang1()   

  1. 1. Department of Anesthesiology，Lishui Municipal Central Hospital，the Fifth Affiliated Hospital of Wenzhou Medical University，Lishui 323000，Zhejiang，Chin
    2. Department of Spine Surgery，Lishui Municipal Central Hospital，the Fifth Affiliated Hospital of Wenzhou Medical University，Lishui 323000，Zhejiang，Chin
    3. Department of Radiology，Lishui Municipal Central Hospital，the Fifth Affiliated Hospital of Wenzhou Medical University，Lishui 323000，Zhejiang，China
  • Received:2026-06-09 Online:2026-09-10 Published:2026-10-10
  • Contact: Chuanguang Wang
  • About author:

    Yu Yunyun performed the experiments and wrote the article. Zheng Fang performed the experiments. Ding Weimin， Fang Jiawei， and Yang Qing revised the article. Wang Chuanguang designed the study and reviewed the article. All authors read and approved the final manuscript as submitted.

PDF (PC)

43

Abstract:

Objective This study aims to develop machine learning-based prediction models for postoperative delirium （POD） in elderly patients undergoing spinal surgery， and to compare the predictive performance of different algorithms. Methods This retrospective study enrolled 809 elderly patients who underwent spinal surgery at our hospital between January 2018 and May 2025. All patients were randomly divided into a training set and a validation set at an 8：2 ratio. Least absolute shrinkage and selection operator （LASSO） regression analysis was applied to identify potential predictive factors for POD， the synthetic minority oversampling technique （SMOTE） algorithm was adopted to resolve the class imbalance issue， and nine machine learning models were subsequently constructed. Receiver operating characteristic （ROC） curves were generated to assess the predictive performance of the established models in the validation set. SHapley Additive exPlanations （SHAP） analysis was utilized to interpret the best-performing model and rank the importance of each feature variable. Results Statistically significant differences （P < 0.05） were observed between the study groups in age， prevalence of diabetes， prevalence of cerebrovascular disease， drinking history， albumin level， intraoperative hypotension， intraoperative blood loss， duration of surgery， duration of anesthesia， and intraoperative sevoflurane dosage. LASSO regression analysis identified age， diabetes， albumin level， intraoperative hypotension， intraoperative sevoflurane dosage， and duration of surgery as independent predictive factors for POD in elderly patients undergoing spinal surgery. Among the nine tested machine learning models， the Light Gradient Boosting Machine model achieved the optimal predictive performance， with an area under the receiver operating characteristic curve （AUC） of 0.928. Feature importance ranking indicated that albumin made the greatest contribution to POD prediction. Conclusion The prediction model for postoperative delirium in elderly patients undergoing spinal surgery based on Light Gradient Boosting Machine demonstrates good predictive performance and can provide a reference for the early identification of high-risk patients and the formulation of perioperative intervention strategies.

Key words: spinal surgery, delirium, machine learning, perioperative management

CLC Number: