基于随机森林算法构建与验证缺血性脑卒中患者社会功能缺陷风险预测模型

Construction and Validation of a Risk Prediction Model for Social Functional Impairment in Ischemic Stroke Patients Based on the Random Forest Algorithm

摘要: 目的：探讨缺血性脑卒中患者社会功能缺陷的影响因素，构建并验证基于随机森林算法的社会功能缺陷预测模型。方法：采用病例对照研究选取2022年8月至2023年3月在华北理工大学附属医院诊治的存在社会功能缺陷的患者为病例组，不存在社会功能缺陷的患者为对照组。使用SPSS22.0进行单因素分析，使用二元logistic回归分析进行多因素分析。采用随机森林模型算法建模，通过受试者工作特征曲线下面积、准确率、灵敏度、特异度和F1分数等对模型性能进行综合评价。结果：多因素logistic回归分析显示，年龄 ≥ 60岁(OR = 3.856, 95% CI: 2.552~5.827)、文化程度低(OR = 2.300, 95% CI: 1.430~3.699)、认知功能障碍(OR = 1.633, 95% CI: 1.047~2.549)、伤残接受度低(OR = 2.387, 95% CI: 1.611~3.537)、健康自我管理能力低水平(OR = 1.697, 95% CI: 1.115~2.584)、存在卒中后疲劳(OR = 2.815, 95% CI: 1.927~4.112)与缺血性脑卒中患者社会功能缺陷高风险相关。构建的随机森林预测模型的AUC值、准确率、灵敏度、特异度和F1分数分别为0.785、0.721、0.744、0.698和0.727。引入SHAP解释工具对预测模型进行解释，变量重要性由高到低依次为年龄、卒中后疲劳、伤残接受度、文化程度、健康自我管理能力和认知功能。结论：本研究构建的随机森林模型预测效能良好，可以帮助临床医护人员对缺血性脑卒中社会功能缺陷高危人群进行筛查。

Abstract: Objective: To investigate the influencing factors of social functional impairment in patients with ischemic stroke, and to develop and validate a predictive model for social functional impairment based on the random forest algorithm. Methods: A case-control study was conducted, selecting patients with social functional impairment treated at the Affiliated Hospital of North China University of Science and Technology from August 2022 to March 2023 as the case group, and patients without social functional impairment as the control group. Univariate analysis was performed using SPSS 22.0, and multivariate analysis was conducted using binary logistic regression. A random forest model was developed, and its performance was comprehensively evaluated using metrics such as the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, and F1 score. Results: Multivariate logistic regression analysis showed that age ≥60 years (OR = 3.856, 95% CI: 2.552~5.827), low education level (OR = 2.300, 95% CI: 1.430~3.699), cognitive impairment (OR = 1.633, 95% CI: 1.047~2.549), low disability acceptance (OR = 2.387, 95% CI: 1.611~3.537), low health self-management ability (OR = 1.697, 95% CI: 1.115~2.584), and the presence of post-stroke fatigue (OR = 2.815, 95% CI: 1.927~4.112) were associated with a higher risk of social functional impairment in ischemic stroke patients. The random forest predictive model achieved an AUC of 0.785, accuracy of 0.721, sensitivity of 0.744, specificity of 0.698, and an F1 score of 0.727. Introducing the SHAP interpretation tool to explain the prediction model, the order of variable importance from high to low is age, post-stroke fatigue, disability acceptance, education level, health self-management ability, and cognitive function. Conclusion: The random forest model developed in this study demonstrates good predictive performance and can assist clinical healthcare providers in screening high-risk populations for social functional impairment in ischemic stroke patients.