Construction of a risk prediction model for acquired intraoperative pressure injury in elderly orthopedic surgery: Based on machine learning methods
老年骨科术中获得性压力性损伤风险预测模型构建——基于机器学习方法

摘要

目的 借助基于机器学习方法构建老年骨科术中获得性压力性损伤（IAPI）的风险预测模型，为临床预测老年骨折患者IAPI提供参考依据。方法 挑选医院2023年11月至2025年11月收治的94例老年骨科手术患者当作对象进行回顾性分析，依照术中是否发生IAPI进行分组，分为对照组（术中未发生IAPI，n=65）、研究组（术中发生IAPI，n=29）。收集两组的病历资料，分析老年骨科患者术中IAPI的风险因素，基于机器学习方法构建预测模型，分析受试工作者特征（ROC）曲线对比不同预测模型的性能。结果 多因素Logistic回归模型分析发现，BMI≥24kg/m2、手术时间≥3h、术前无营养支持、未使用预防性敷料、局部皮肤潮湿、术中低体温是老年骨科手术患者发生IAPI的风险因素（P＜0.05）。决策树模型的准确率、敏感度、精确度、F1分数、AUC分别为0.853、0.856、0.841、0.836、0.62，朴素贝叶斯算法模型分别为0.826、0.829、0.765、0.791、0.78，随机森林算法模型分别为0.875、0.878、0.849、0.845、0.83，相较于决策树模型、朴素贝叶斯模型，随机森林算法模型的各项数值更高、整体预测效能更优。结论 基于机器学习方法所构建的随机森林模型，对于老年骨科术中IAPI具备较高的预测性能，可作为临床预测老年骨科术中IAPI发生的辅助手段。

Abstract

Objective To construct a risk prediction model for acquired intraoperative pressure injury (IAPI) in elderly orthopedic surgery using machine learning methods, and to provide a reference for clinical prediction of IAPI in elderly fracture patients.
Methods 94 elderly orthopedic surgery patients admitted to the hospital from November 2023 to November 2025 were selected for retrospective analysis. They were divided into the control group (no IAPI occurred during the operation, n=65) and the study group (IAPI occurred during the operation, n=29). The medical records of the two groups were collected, and the risk factors of IAPI in elderly orthopedic patients during the operation were analyzed. A prediction model was constructed based on machine learning methods, and the performance of different prediction models was compared using the receiver operating characteristic (ROC) curve of the test subjects’ characteristics.
Results The multivariate Logistic regression model analysis found that BMI≥24 kg/m2, operation time≥3 hours, no preoperative nutritional support, no use of preventive dressings, local skin moisture, and intraoperative hypothermia were risk factors for IAPI in elderly orthopedic surgery patients (P<0.05). The accuracy, sensitivity, specificity, F1 score, and AUC of the decision tree model were 0.853, 0.856, 0.841, 0.836, and 0.62, respectively. Those of the naive Bayes algorithm model were 0.826, 0.829, 0.765, 0.791, and 0.78, respectively. Those of the random forest algorithm model were 0.875, 0.878, 0.849, 0.845, and 0.83, respectively. Compared with the decision tree model and the naive Bayes model, the values of the random forest algorithm model were higher, and the overall predictive efficacy was better.
Conclusion   The random forest model constructed based on machine learning methods has a high predictive performance for IAPI during intraoperative orthopedic surgery in the elderly and can be used as an auxiliary means for clinical prediction of IAPI occurrence in elderly orthopedic surgery.