目的 旨在构建高性能煤工尘肺(coal workers'pneumoconiosis,CWP)风险预测模型,促进CWP的早期预防.方法 基于医院的病例对照研究,收集2017-2022年山西省某职业病医院的CWP患者和同期矿工非CWP患者病例资料,建立CWP数据库,采用随机森林筛选特征变量,基于反向传播(back propagation,BP)神经网络和logistic回归分析模型分别构建CWP预测模型,并利用受试者工作特征(receiver operating characteristic,ROC)曲线评价2个模型的CWP预测能力.结果BP神经网络模型灵敏度为88.6％,特异度为87.6％,准确率为87.12％;变量正态化重要性结果显示,影响煤矿工人发生CWP最重要的因素有1秒通气率(forceful expiratory volume in 1 second/force-ful vital capacity,FEV1/FVC)、工龄、工种.Logistic回归分析模型结果显示灵敏度80.7％,特异度84.1％,准确率 82.7％.BP 神经网络模型 ROC 曲线下面积(area under the curve,AUC)(AUC=0.918,95％CI:0.903～0.964)高于 l ogistic 回归分析模型(AUC=0.802,95％CI:0.750～0.850),BP神经网络模型的预测性能优于logistic回归分析模型.结论BP神经网络的预测性能高于logistic回归分析模型,将BP神经网络应用在CWP预测上有更高的准确性.FEV1/FVC、工龄、工种是影响煤矿工人发生CWP的重要因素.
A neural network risk prediction model of coal workers'pneumoconiosis-a hospital-based case-control study
Objective This study aims to construct a high-efficiency coal workers'pneumoconiosis(CWP)risk prediction model to promote early prevention of CWP.Methods We conducted a case-control study based on hospital records,collected case data of coal workers diagnosed with CWP and non-CWP in an occupational disease hospital in Shanxi Province from 2017 to 2022 and established a database of CWP.Random forest method was used to screen the characteristic variables.The CWP prediction model was constructed based on back propagation(BP)neural network and Logistic regression respectively,and the CWP prediction ability of the two models was evaluated by receiver operating characteristic(ROC).Results The BP neural network model demonstrated a sensitivity of 88.6％,a specificity of 87.6％,and an accuracy rate of 87.12％.Based on variable normalization importance analysis,the most influential factors for CWP prevalence in coal workers were forceful expiratory volume in 1 second/forceful vital capacity(FEV1/FVC),working age and work type.The logistic regression model showed a sensitivity of 80.7％,a specificity of 84.1％,and an accuracy rate of 82.7％.The BP neural network model exhibited a higher area under the curve(AUC)value(AUC=0.918,95％CI:0.903-0.964)compared to the logistic regression model(AUC=0.802,95％CI:0.750-0.850),indicating superior predictive performance.Conclusions The BP neural network model provides better predictive performance compared to the logistic regression model,and applying the BP neural network to CWP prediction has higher accuracy.FEV1/FVC,working age and work type are identified as significant factors influencing the occurrence of CWP in coal workers.
Back propagation neural networkCoal workers'pneumoconiosisLogistic regres-sionPrediction model
万方数据
维普
