基于机器学习算法的护理人员心理健康状况预测模型研究  

Machine learning-based models for prediction of nursing staff mental health status

作　　者：王沛如 罗泽槟 郭植君 李丹丹[3] 王逸如[3] Wang Peiru;Luo Zebing;Guo Zhijun;Li Dandan;Wang Yiru(Department of Nursing,Shantou Central Hospital,Shantou 515031,China;Department of Nursing,Shantou University Medical College,Shantou 515041,China;Department of Nursing,Cancer Hospital of Shantou University Medical College,Shantou 515031,China)

机构地区：[1]汕头市中心医院护理部,515031 [2]汕头大学医学院护理系,515041 [3]汕头大学医学院附属肿瘤医院护理部,515031

出　　处：《中国实用护理杂志》2021年第35期2721-2728,共8页Chinese Journal of Practical Nursing

基　　金：2019年广东省科技专项资金(大专项+任务清单)项目(2019-132)。

摘　　要：目的建立基于机器学习算法的护理人员心理健康状况预测模型。方法采用便利抽样法,应用一般资料调查表、症状自评量表(SCL-90)、应对方式问卷、社会支持评定量表和工作适应障碍量表于2020年2月对汕头市中心医院和汕头大学医学院附属肿瘤医院的护理人员进行调查。心理健康状况作为二分类变量处理,采用单因素和多因素Logistic回归分析筛选出候选预测因子。研究对象按照8∶2比例随机分为训练集和测试集。应用机器学习算法的Logistic回归、人工神经网络、C5.0决策树、贝叶斯网络和支持向量机建立护理人员心理健康状况预测模型,并对5个模型进行验证及对比分析,筛选出最高预测效能模型。结果本研究共纳入415名护士,心理健康症状阳性检出率为20.48%。根据单因素和多因素Logistic回归分析筛选出的候选预测因子分别为工作适应障碍(OR值为1.098,95%CI 1.028~1.174)、自责(OR值为7.703,95%CI 2.014~29.468)、解决问题(OR值为0.131,95%CI 0.025~0.686)、每月的夜班数(OR值为0.204,95%CI 0.073~0.573)和支持利用度(OR值为0.830,95%CI 0.701~0.984)。Logistic回归、人工神经网络、C5.0决策树、贝叶斯网络和支持向量机5种模型预测准确率分别为84.21%、85.53%、82.89%、78.95%、84.21%;受试者工作特征曲线下面积(AUC)分别为0.801、0.825、0.777、0.583、0.774。人工神经网络模型预测效能高于Logistic回归、C5.0决策树、贝叶斯网络和支持向量机模型(DeLong Test,P<0.05)。结论基于机器学习算法建立的护理人员心理健康状况预测模型有较高的预测价值,工作适应障碍、自责和解决问题应对方式、每月的夜班数、支持利用度为模型的预测因子,可将模型纳入护理人员心理健康状况筛查决策,以精准掌握其动态变化,早期识别心理健康异常高危人员,早期干预。Objective To establish a model for predicting the mental health status of nurses based on machine learning algorithm.Methods In February 2020,the nurses from Shantou Central Hospital and Cancer Hospital of Shantou University Medical College were recruited by convenience sampling,investigated using the Self-reporting Inventory,Coping Style Questionnaire,Social Support Rating Scale and Work Attitude Scale.Mental health status was treated as a dichotomous variable,and candidate predictors were screened out by univariate and multivariate Logistic regression analysis.The subjects were randomly divided into a training set(80%)and a test set(20%).Then five prediction models of nursing staff mental health status were constructed using the five machine learning methods(Logistic Regression,Artificial Neural Network,C5.0 Decision Tree,Bayesian Network and Support Vector Machine),verified and compared to screen out the model with the highest predictive efficiency.Results A total of 415 nurses were enrolled,and the positive detection rate of mental health symptoms was 20.48%.According to univariate and multiple Logistic regression analysis,candidate predictors were work attitude(OR=1.098,95%CI 1.028-1.174),self-accusation(OR=7.703,95%CI 2.014-29.468),problem-solving(OR=0.131,95%CI 0.025-0.686),the number of night shifts per month(OR=0.204,95%CI 0.073-0.573)and support availability(OR=0.830,95%CI 0.701-0.984).The accuracy of prediction of Logistic Regression,Artificial Neural Network,C5.0 Decision Tree,Bayesian Network and Support Vector Machine were 84.21%,85.53%,82.89%,78.95%,84.21%.The area under the ROC curve was 0.801,0.825,0.777,0.583,0.774.Artificial Neural Network was significantly more effective than Logistic regression,C5.0 Decision Tree,Bayesian Network and Support Vector Machine(DeLong test,P<0.05).Conclusions The machine learning based predictive models for nursing staff mental health status has higher predictive value,which can be applied into nursing staff mental health screening decisions to accurately grasp it

关 键 词：护士 心理健康 机器学习 预测模型 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] R192.6[自动化与计算机技术—控制科学与工程]