决策树在贫困农村老人就诊影响因素中的应用

Decision tree analysis in determinants of elderly visits in poor rural areas

目的:探究贫困地区农村老人就诊行为的决策模式,为精准医疗扶贫提供科学依据。方法:采用分阶段抽样的方法,对内蒙古自治区乌兰察布市下属察哈尔右翼前旗、察哈尔右翼中旗、察哈尔右翼后旗、凉城县共4个贫困旗县的1 271名老人进行横断面问卷调查,内容主要包括社会经济人口学特征、日常消费水平、欧洲五维健康量表(Euro Qol five dimensions questionnaire,EQ-5D)和直观式健康量表(visual analogue scale,VAS)、社会支持量表,以及卫生服务需要和利用情况,遴选其中在过去30 d内存在身体不舒适情况的1 039名老人为研究对象。采用卡方检验分析不同群体之间的差异,并利用Logistic回归和决策树两种方法对贫困农村老人在身体不舒适时的就诊行为进行多因素分析,探索老人就诊决策的影响因素。结果:研究对象平均年龄(71.8±7.1)岁,52.2%为文盲,85.8%具有中等社会支持,58.5%与配偶同住,多居于15 min医疗圈内,不能从子女处获得经济支持。30 d内身体不舒适时的就诊率为31.0%。卡方检验显示,就诊率在年龄、民族、居住模式、日常消费指数、住房类型、社会支持得分、有无子女补贴、前往医疗机构交通时间、健康自评得分水平间的差异均具有统计学意义。在影响老人就诊行为的因素判别上,决策树模型比Logistic回归模型有更低的分类错误率。Logistic回归模型错误分类率为31.4%,显示年龄、民族、居住模式、日常消费指数、社会支持总得分、前往医疗机构交通时间和健康自评得分的差异对就诊决策有统计学意义。决策树模型错误分类率为28.6%,显示前往医疗机构交通时间、居住模式、文化程度、年龄、是否有子女补贴、社会支持总得分依次构成老人就诊决策的主要影响因素,预测变量重要性分别为0.42、0.21、0.13、0.11、0.07和0.06。结论:在贫困农村地区,医疗资源、经济承受能力、家人,以及个人的社会人口学特征是影响老人身体不舒适时的就诊决策的关键因素,应当将改善老人的医疗状况融入社会整体发展,通过综合干预改善贫困地区农村老人的医疗服务利用水平。

Objective： To explore the influencing factors of elderly outpatient visits and to provide evidence for poverty reduction in health in the poor rural areas. Methods： Through stratified sampling,a total of 1 271 aged people in four poverty Qi/County of Ulanqabcity were surveyed,including Qahar Right Wing Front Banner,Qahar Right Wing Middle Banner,Qahar Right Wing Rear Banner and Liangcheng County. Their socio-economic and demographic characteristics,daily consumption,Euro Qol five dimensions questionnaire（ EQ-5 D） and visual analogue scale（ VAS）,social support,health service needs and utilization were collected through cross-sectional household questionnaires. 1 039 aged people who had experienced physical discomfort in the past 30 days were selected as subjects for the study. The differences between the groups were analyzed by chi-square test. A Logistic regression equation and a decision tree of elderly visits were built to find factors influencing decision-making of the aged. Results： The average age of the research subjects was 71. 8 years,with 52. 2% being illiterate and 85. 8% with middle social support. 58. 5% of the subjects living with their spouses,mostly living in 15 min medical circle and without any financial support from their children. The 30-day visiting rate when having physical discomfort was 31. 0%. The chi-square test showed that the differences in visit rates among age,ethnic,residence patterns,daily consumption index,housing types,social support scores,grown children＇s economic assistance,travel time to medical institutions,and health self-assessment scores were statistically significant. Compared with Logistic analysis,the decision tree showed lower error rate of classification.Logistic regression model＇s error rate of classification was 31. 4%,showing that the differences in visit rates among age,ethnic,residence patterns,daily consumption index,social support scores,travel time to medical institutions,and health self-assessment scores were statistically significant. The decision tree model＇s error rate of classification was 28. 6%,showing six main influencing factors,including the travel time to medical institutions,cohabitants,education level,age,whether adult children provide economic support and social support score. The importance of these predictors were 0. 42,0. 21,0. 13,0. 11,0. 07 and 0. 06,respectively. Conclusion： In poor rural areas,medical resources,economic affordability,family and individual socio-demographic characteristics are the key factors affecting decision-making for the aged. It is necessary to integrate the improvement of the health care of the aged into the overall development of the society. And comprehensive interventions should be adopted to improve the outpatient utilization for aged in poor rural areas.