Abstract:

Objective To compare the performance of 3 risk prediction models based on machine learning in predicting the risk of compassion fatigue among obstetrics and gynaecology nurses. Methods Using the convenience sampling method，1 323 obstetrics and gynaecology nurses from 11 tertiary hospitals in 9 cities were selected from December 2022 to March 2023，and were randomly divided into a training set and a test set according to 7∶3 ratio. Compassion Fatigue Scale，Five Facet Mindfulness Questionnaire and Emotional Intelligence Scale were used for the survey. A total of 3 types of risk prediction models were constructed for the factors affecting the compassion fatigue of obstetrics and gynaecology nurses，namely，Logistic Regression，Decision Tree，and Random Forest. The accuracy，precision，specificity，sensitivity，F1 score and AUC were used to compare the predictive performance of the 3 models. Results Finally 1 276 maternity nurses completed the survey. All 3 models showed that nature of employment，years of experience，mindfulness and emotional intelligence were independent risk factors for compassion fatigue in obstetrics and gynaecology nurses （P<0.05）. The results of model comparison showed that the accuracy of Logistic regression，decision tree and random forest were 0.804，0.806，0.796；the precision was 0.821，0.827， 0.823；the sensitivity was 0.956，0.949，0.939；the F1 score was 0.883，0.884，0.877；the AUC was 0.704（95%CI：0.701~0.713），0.760（95%CI：0.751~0.771），0.742（95%CI：0.723~0.762） respectively. Conclusion The risk prediction model of factors affecting compassion fatigue among obstetrics and gynaecology nurses constructed by decision tree performed the best，and the predictive performance was better than that of the random forest and logistic regression models. The multi-model effectively predicts the risk of compassion fatigue in obstetrics and gynecology nurses，explores the interaction of influencing factors in multiple dimensions，and it can inform the early identification and prevention of compassion fatigue and the development of related interventions.

Key words: Machine Learning, Obstetrics and Gynecology Nurses, Compassion Fatigue, Root Cause Analysis, Nursing Administration Research

摘要：

目的 基于机器学习构建3种妇产科护士共情疲劳的风险预测模型，比较不同模型的预测性能。方法 采用便利抽样法，于2022年12月—2023年3月选取9个城市11所三级甲等医院的1 323名妇产科护士作为调查对象，按照7∶3的比例随机分为训练集和测试集，采用同情疲劳量表、五因素正念度量表及情绪智力量表进行调查。基于机器学习构建妇产科护士共情疲劳的Logistic回归模型、决策树模型和随机森林模型3种风险预测模型，比较各模型的准确率、精确率、灵敏度、F1指数和受试者工作特征（receiver operating characteristic，ROC）曲线下面积（area under the curve，AUC），评价模型的预测性能。结果 最终1 276名妇产科护士完成调查。3种模型均显示，正念水平、情绪智力、用工性质和工作年限是妇产科护士共情疲劳的影响因素（P<0.05）。Logistic回归模型、决策树模型和随机森林模型的准确率分别为0.804、0.806、0.796，精确率分别为0.821、0.827、0.823，灵敏度分别为0.956、0.949、0.939，F1指数分别为0.883、0.884、0.877，AUC分别为0.704（95%CI为0.701~0.713）、0.760（95%CI为0.751~0.771）、0.742（95%CI为0.723~0.762）。结论 通过决策树构建的妇产科护士共情疲劳风险预测模型性能优于随机森林模型和Logistic回归模型，多模型有效结合预测妇产科护士共情疲劳的发生风险、多维度探索影响因素交互作用，可为共情疲劳的早期识别和预防、相关干预措施的制订提供参考。

关键词: 机器学习, 妇产科护士, 共情疲劳, 影响因素分析, 护理管理研究