摘 要: 目的 应用Logistic回归和BP神经网络模型对口腔医生工作相关肌肉骨骼疾患(WMSDs)进行预测研究,评估两种模型的预测效果。方法 采用网络问卷方式调查福州市口腔医生WMSDs的患病情况及影响因素,建立Logistic回归和BP神经网络两种预测模型,绘制ROC曲线比较两者的预测效能。结果 回收有效问卷627份,口腔医生WMSDs发生率前四位的部位分别为颈部(43.1%)、肩部(32.2%)、下背/腰部(28.4%)、手/腕部(20.1%)。综合两种预测模型分析结果,年龄、职称、长时间保持侧头姿势、长时间(≥2h)坐在同一位置是颈部WMSDs的影响因素;性别、年龄、肘部长时间保持弯曲、长时间(≥2h)坐在同一位置是肩部WMSDs的影响因素;年龄、长时间保持弯腰姿势是下背/腰部WMSDs的影响因素;性别、年龄、手腕需要保持向上/下弯曲的状态、经常伸长手臂取物品、长时间(≥2h)坐在同一位置、休息时间充足是手/腕部WMSDs的影响因素。除下背/腰部Logistic回归模型外,其他部位两种模型的AUC值均>0.7,各部位BP神经网络模型的AUC值、灵敏度、特异性、约登指数均大于Logistic回归模型,两种模型的预测效果均较好,BP神经网络模型预测能力更佳。结论 Logistic回归和BP神经网络模型可应用于口腔医生WMSDs的预测,BP神经网络模型的预测效能更好。 |
关键词: 工作相关肌肉骨骼疾患(WMSDs) 口腔医生 神经网络 Logistic回归 |
中图分类号: R135;R68
文献标识码: A
|
基金项目: 原国家卫生计生委工作项目(131031109000160004);福建省自然科学基金项目(面上项目:2020J01915);福建省医学创新课题(2020CXB039);福州市卫生健康科研创新团队培育项目(2020-S-wt9) |
|
Comparison of two models for predicting work-related musculoskeletal disorders in dentists |
LIU Pei-fang,SHEN Bo,XU Xu-yan,WANG Zhong-xu,JIA Ning,LIN Si-hao,SHE Yuan-yu
|
Fuzhou Municipal Center for Disease Control and Prevention,Fuzhou 350004,China
|
Abstract: Objective The Logistic regression and BP neural network model were used to predict the work-related musculoskeletal disorders(WMSDs)of dentists and evaluate the prediction effect of the two models. Methods The prevalence and influencing factors of WMSDs among dentists in Fuzhou were investigated by network questionnaire. Two prediction models of Logistic regression and BP neural network prediction model were established,and R0C curve was drawn to compare the prediction efficiency of the two models. Results Six hundred and twenty seven valid questionnaires were collected. The top four parts of the prevalence of WMSDs were neck(43.1%),shoulder(32.2%),lower back/waist(28.4%)and hand/wrist(20.1%). Age,professional title,long term lateral head posture and sitting in the same position for long time(≥2 hours)were the influencing factors of neck WMSD;gender,age,elbow bending for a long time and sitting in the same position for a long time(≥2 hours)were the risk factors of shoulder WMSDs;age and long-term bending posture were the influencing factors of lower back/waist WMSDs;while gender,age,the need to keep the wrist bent up and down,often extend the arm to reach items,sitting in the same position for a long time(≥2 hours),and sufficient rest time were the influencing factors of hand/wrist WMSDs. In addition to the lower back/waist Logistic regression model,the AUC of other models were more than 0.7,furthermore,the AUC,sensitivity,specificity and yoden's index of neural network model at all parts were better than those of Logistic regression model. The prediction effects of the two models were acceptable,but the prediction ability of BP neural network model were better. Conclusion The results suggested that both Logistic regression and BP neural network model can be applied to predict the WMSDs of dentists,BP neural network model has better prediction efficiency. |
Keywords: work-related musculoskeletal disorders(WMSDs) dentists neural network Logistic regression |