CAJ | 학술논문

目的构建多分类倾向性评分匹配方法，并将其应用到分组变量为三分类的流行病学调查数据中。方法数据来源于前期的流行病学调查数据，从五个城市中随机抽取3600名受访者，收集他们的人口学信息。让受访者自己评价其健康情况，分析自评健康为好、一般和差的受访者其慢性病发病率是否相同，并探讨慢性病发病与否同时还受其他混杂因素的影响。结果在不同倾向性评分匹配方法中，只有马氏距离匹配可消除不同组间协变量的不均衡，设定匹配时的平均马氏距离为0.12。结论马氏距离匹配适用于本研究的数据类型，可有效控制三组间的混杂因素。
목적구건다분류경향성평분필배방법，병장기응용도분조변량위삼분류적류행병학조사수거중。방법수거래원우전기적류행병학조사수거，종오개성시중수궤추취3600명수방자，수집타문적인구학신식。양수방자자기평개기건강정황，분석자평건강위호、일반화차적수방자기만성병발병솔시부상동，병탐토만성병발병여부동시환수기타혼잡인소적영향。결과재불동경향성평분필배방법중，지유마씨거리필배가소제불동조간협변량적불균형，설정필배시적평균마씨거리위0.12。결론마씨거리필배괄용우본연구적수거류형，가유효공제삼조간적혼잡인소。
Objective To establish propensity score matching to multiple data and apply it in the epide-miological data with three categories. Methods Data were obtained from previous epidemiological surveys, in which 3600 subjects from 5 cities were randomly selected. Demographic information was collected and self-rated health was obtained. We analyzed whether the incidence of chronic disease was different among the subjects whose self-rated health were good, fair, or bad, respectively. Also, we explored whether the incidence of chronic disease was also affected by other confounding factors. Results In different propensity score matching, only Mahalanobis metric matching could eliminate the confounding factors among different groups, with caliper of 0.12. Conclusion Mahalanobis metric matching is suitable for our data, and can effectively control the confounding factors among three groups.