【摘要】 目的 了解桂林市医务人员数字素养现状，比较二分类logistic回归模型与决策树模型在医务人员数字素养影响因素分析中的价值。方法 于2023年9-12月采用分层抽样的方法，在广西壮族自治区桂林市的医院、疾病预防控制中心和社区卫生服务中心等医疗机构抽取在职医务人员作为调查对象，包括医师、医技、医药和护士4个岗位。通过在线平台发放《医务人员数字素养及影响因素调查问卷》,收集基本信息，并对医务人员数字素养水平进行评分，得分>60分为数字素养良好，≤60分为数字素养不良。用二分类logistic回归模型RL向前筛选法、决策树模型分类χ~2自动交互检测法(CHAID)分析医务人员数字素养的影响因素。绘制受试者工作特征(ROC)曲线，并计算曲线下面积(AUC),评估2种模型的价值。结果 共发放858份问卷，回收有效问卷677份。医务人员数字素养不良243名，良好434名。二分类logistic回归模型分析结果显示，数字素养不良概率女性高于男性(OR=1.542,95%CI:1.034～2.298,P<0.05),等级较低医院医务人员高于三级甲等医院医务人员(OR值分别为6.178、8.278,95%CI分别为3.478～10.973、4.136～16.569,P<0.05),没有学习过数字素养的医务人员高于学习过数字素养的医务人员(OR=8.352,95%CI:4.911～14.205,P<0.05)。决策树模型分析结果显示，学习数字素养方式和工作医院等级是医务人员数字素养不良的影响因素(χ~2值分别为101.992、10.853,P分别为<0.001、0.002),学习数字素养方式是医务人员数字素养不良的主要影响因素。二分类logistic回归模型ROC曲线下面积稍大于决策树模型(AUC分别为0.782、0.758,Z=2.767,P<0.001)。结论 二分类logistic回归模型对医务人员数字素养影响因素的分析优于决策树模型，但决策树模型可以分析影响因素间的层次关系和相互作用，两种模型可结合应用。更多还原

【Abstract】 Objective To investigate the current status of digital literacy among medical staff in Guilin city and to compare the value of binary logistic regression models and decision tree models in analyzing the influencing factors of digital literacy among medical staff. Methods From September to December 2023,a stratified sampling method was used to select employed medical staff from hospitals, disease control centers, community health service centers, and other medical institutions in Guilin city, Guangxi Zhuang Autonomous Region, including physicians, medical technicians, pharmacists, and nurses.The “Questionnaire on Digital Literacy and Influencing Factors of Medical Staff” was distributed through an online platform to collect basic information.The digital literacy level of medical staff was scored, with a score >60 considered good digital literacy and ≤60 considered poor digital literacy.Binary logistic regression models with RL forward selection and decision tree models using the Chi-squared Automatic Interaction Detector(CHAID) method were applied to analyze the influencing factors of digital literacy among medical staff.Receiver operating characteristic(ROC) curves were plotted, and the area under the curve(AUC) was calculated to evaluate the value of the two models.A stratified sampling method was used to analyze the factors related to digital literacy of medical staff in Guilin, Guangxi, from September to December 2023,and the effects of two predictive models were evaluated using the subject work curve(ROC). Results A total of 858 questionnaires were distributed, and 677 valid responses were collected.Among the medical staff, 243 had poor digital literacy, while 434 had good digital literacy.The binary logistic regression analysis showed that the probability of poor digital literacy was higher among females than males(OR=1.542,95%CI:1.034-2.298,P<0.05),higher among medical staff from lower-level hospitals than those from tertiary Grade A hospitals(OR=6.178,8.278,95%CI:3.478-10.973 and 4.136-16.569,P<0.05),and higher among medical staff who had not received digital literacy training than those who had(OR=8.352,95%CI:4.911-14.205,P<0.05).The decision tree model analysis revealed that the method of digital literacy training and the grade of the hospital were influencing factors for poor digital literacy among medical staff(χ~2=101.992,10.853;P<0.001,P=0.002,respectively),with the method of digital literacy training being the primary influencing factor.The AUC of the ROC curve for the binary logistic regression model was slightly larger than that of the decision tree model(AUC=0.782,0.758;Z=2.767,P<0.001). Conclusion The binary logistic regression model outperforms the decision tree model in analyzing the influencing factors of digital literacy among medical staff.However, the decision tree model can reveal hierarchical relationships and interactions among influencing factors.The two models can be used in combination.更多还原