口腔医学研究

口腔医学研究 ›› 2024, Vol. 40 ›› Issue (10): 867-872.DOI: 10.13701/j.cnki.kqyxyj.2024.10.004

• 口腔颌面外科学研究 • 上一篇    下一篇

基于机器学习的青年人群颞下颌关节紊乱病风险预测模型的构建

崔宇琛, 张晗, 胡志强, 张琦, 袁佳敏, 朱宪春*   

  1. 吉林大学口腔医院正畸科 吉林 长春 130021
  • 收稿日期:2024-05-06 出版日期:2024-10-28 发布日期:2024-10-24
  • 通讯作者: *朱宪春,E-mail:zhuxc@jlu.edu.cn
  • 作者简介:崔宇琛(1999~),女,山西人,硕士在读,研究方向:口腔正畸治疗的生物力学机制及临床优化应用。
  • 基金资助:
    吉林省科技厅自然科学基金项目(编号:YDZJ202201ZYTS057)

Construction of A Risk Prediction Model for Temporomandibular Disorders in Young People Based on Machine Learning

CUI Yuchen, ZHANG Han, HU Zhiqiang, ZHANG Qi, YUAN Jiamin, ZHU Xianchun*   

  1. Department of Orthodontics, Hospital of Stomatology, Jilin University, Changchun 130021, China
  • Received:2024-05-06 Online:2024-10-28 Published:2024-10-24

PDF (PC)

18

摘要： 目的: 本研究旨在探讨青年人群颞下颌关节紊乱病(temporomandibular disorders, TMD)的影响因素,并基于机器学习(machine learning, ML)方法构建一个针对青年人的TMD风险预测模型,以便为青年人提供更加准确和有效的TMD风险评估工具。方法: 共纳入960例符合条件的大学生做为研究对象。使用单因素分析和最小绝对值收缩和选择算子回归算法筛选TMD风险因素。运用6种不同的ML方法构建TMD风险预测模型,并采用Shapley加法解释算法对最终模型进行解释。结果: 共纳入12个预测因素进行模型的构建,随机森林模型在6种ML模型中表现最佳。该模型在外部测试集上的受试者工作特征曲线下面积为0.863(95%CI:0.812~0.915),准确度为0.732,灵敏度为0.898,特异度为0.728,阳性预测值为0.864,阴性预测值为0.703。校准曲线表明该模型预测效果和实际结果基本一致,决策曲线表明模型具有良好的临床适用性。结论: 基于ML方法构建的青年人群TMD风险预测模型具有良好的预测性能及临床适用性,可辅助临床进行更高效的疾病管理和更精准的医疗干预。

关键词: 颞下颌关节紊乱病, 机器学习, 预测模型, shapley加法解释算法

Abstract: Objective: To investigate the influencing factors of temporomandibular disorders (TMD) in young people, and to construct a TMD risk prediction model for young people based on machine learning (ML) methods in order to provide a more accurate and effective risk assessment tool. Methods: A total of 960 eligible college students were included in the study. Univariate analysis and the least absolute shrinkage and selection operator (LASSO) regression algorithm were used to screen for risk factors. Six ML methods were used to construct the TMD risk prediction model and the Shapley additive explanations (SHAP) algorithm was used to interpret the final model. Results: A total of 12 predictors were included in the model construction, and the random forest (RF) model performed best among the 6 ML models. The model had an area under the receiver operating characteristic (ROC) curve (AUC) of 0.863 (95%CI: 0.812-0.915), an accuracy of 0.732, a sensitivity of 0.898, a specificity of 0.728, a positive predictive value (PPV) of 0.864, and a negative predictive value (NPV) of 0.703 on the external test set. The calibration curve showed that the predictive performance of the model was basically consistent with the actual situation, and the decision curve analysis (DCA) showed that the model had good clinical applicability. Conclusion: The TMD risk prediction model for young people constructed based on the ML method has good prediction performance and clinical applicability, which can help the clinic to carry out more efficient disease management and more accurate medical interventions.

Key words: temporomandibular disorders, machine learning, prediction model, SHAP algorithm