|
| 基于机器学习方法的脑卒中患者口腔衰弱风险预测模型构建与验证 |
| Construction and validation of an oral frailty risk prediction model for stroke patients based on multiple machine learning methods |
| 投稿时间：2025-07-17 |
| DOI：10.3969/j.issn.1000-0399.2026.03.015 |
| 中文关键词: 机器学习 脑卒中 口腔衰弱 风险预测 影响因素 |
| 英文关键词: Machine learning Stroke Oral frailty Risk prediction Influencing factors |
| 基金项目:安徽省高校科研项目(编号:2024AH0509832024) |
|
| 摘要点击次数: 498 |
| 全文下载次数: 96 |
| 中文摘要: |
| 目的 基于机器学习方法分析脑卒中患者并发口腔衰弱的影响因素，并构建风险预测模型。方法 选取2024年9月至2025年2月在安徽中医药大学第二附属医院住院治疗的301例脑卒中患者为研究对象。采用多种调查工具收集患者的基本资料和专科指标，包括口腔衰弱指标筛查-8(OFI-8)、身体衰弱量表（FRAIL）、口腔健康评估工具（OHAT）等。通过单因素后进行多因素分析筛选出显著影响因素，使用逻辑回归（Logistic）、随机森林（Random Forest）、决策树（Decision Tree）、支持向量机（SVM）和K近邻分类（KNN）等5种机器学习方法构建预测模型，并进行验证分析。结果 多因素logistic回归分析结果显示，OHAT得分低（OR=1.588,95%CI:1.355～1.861）、身体衰弱（OR=2.642,95%CI:1.475～4.731）是脑卒中患者并发口腔衰弱的独立危险因素（P<0.05）；日常生活活动能力强（OR=0.408,95%CI:0.227～0.736）为其独立保护因素（P<0.05）。受试者工作特征（ROC）曲线分析结果显示，5种模型中，决策树模型在训练集中的表现最佳，其曲线下面积（AUC）为0.907(95%CI:0.844～0.970)；但在验证集中，逻辑回归模型的表现优于其他模型，其AUC为0.713(95%CI:0.415～0.981)，灵敏度为0.636(0.473～0.799)，特异度为0.653(0.547～0.760)。结论 本研究基于机器学习方法成功构建了脑卒中患者并发口腔衰弱的风险预测模型。验证集结果表明，逻辑回归模型具有更好的泛化能力，更适合用于临床预测。 |
| 英文摘要: |
| Objective To analyze the influencing factors of oral frailty in stroke patients using machine learning methods and to construct a risk prediction model. Methods A total of 301 stroke patients hospitalized at the Second Affiliated Hospital of Anhui University of Chinese Medicine from September 2024 to February 2025 were enrolled. Comprehensive data were collected using various assessment tools, including the Oral Frailty Index-8(OFI-8), FRAIL scale, and Oral Health Assessment Tool(OHAT). Significant predictors were identified through univariate and multivariate analyses. Five machine learning algorithms—Logistic Regression, Random Forest, Decision Tree, Support Vector Machine(SVM), and K-Nearest Neighbors(KNN)—were employed to develop prediction models, followed by validation analyses. Results Multivariate logistic regression analysis revealed that lower OHAT scores(OR=1.588, 95%CI:1.355～1.861), and physical frailty(OR=2.642, 95%CI:1.475～4.731) were independent risk factors for oral frailty in stroke patients, whereas strong ability in activities of daily living(OR=0.408, 95% CI: 0.227～0.736) was an independent protective factor(P<0.05). Receiver operating characteristic(ROC) curve analysis demonstrated that among the five models, the Decision Tree model achieved the best performance in the training set with an area under the curve(AUC) of 0.907(95%CI: 0.844～0.970). However, in the validation set, the Logistic Regression model outperformed others with an AUC of 0.713(95%CI: 0.415-0.981), sensitivity of 0.636(95%CI: 0.473～0.799), and specificity of 0.653(95%CI: 0.547～0.760). Conclusion This study successfully develops risk prediction models for oral frailty in stroke patients using machine learning approaches. Validation results indicate that the Logistic Regression model exhibits superior generalizability and is more suitable for clinical prediction. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
