离散选择实验潜类别logit模型在卫生服务领域的应用与Stata软件的实现  

Discrete Choice Experiment Latent Class Logit Model in Health Services with Stata Software Implementation

作　　者：张春丽 邱佳玲 陈莎 刘仲琦 古羽舟 鲁永恒 邓瑜 郝春[1,6] 郝元涛 Zhang Chunli;Qiu Jialing;Chen Sha(Department of Medical Statistics,School of Public Health,Sun Yat-Sen University(510080),Guangzhou)

机构地区：[1]中山大学公共卫生学院医学统计学系,卫生信息研究中心,广东省卫生信息学重点实验室,510080 [2]广东省妇幼保健院妇女儿童健康研究所 [3]中山大学附属第一医院质量管理评价处服务管理科 [4]广州市疾病预防控制中心艾滋病预防控制部 [5]广州市越秀区康源社区支持中心 [6]中山大学国家治理研究院全球卫生研究中心

出　　处：《中国卫生统计》2023年第4期502-506,共5页Chinese Journal of Health Statistics

基　　金：国家科技重大专项(2018ZX10715004);国家自然科学基金(71974212;72204059);广东省自然科学基金(2020A1515010737);广州市科技计划项目(202201010078)。

摘　　要：目的简要介绍离散选择实验的设计步骤和潜类别分析模型的基本原理,通过实例演示介绍潜类别logit模型在Stata中的实现过程,为该模型在离散选择实验中的实际应用提供方法学的参考。方法基于广州市艾滋病高危人群选择HIV自检试剂偏好的离散选择实验数据,通过实例演示潜类别logit分析模型的构建过程,并提供相应的Stata命令。结果最终确立了4个类别模型为最优分类,模型估计结果显示4个类别模型中类别1(偏好尿液试剂),类别2(偏好更便宜的血液试剂配套说明书讲解和自行判读结果),类别3(偏好更便宜的尿液试剂),类别4(偏好血液试剂配套视频讲解和专业人员判读结果)的差异具有统计学意义。结论潜类别logit模型用于离散选择实验数据的分析具有简便性与灵活性,但也有其应用的局限性,因此需要进一步结合其他模型来优化分析。Objective This study aims to introduce the design steps of discrete choice experiments and the basic principles of the latent class analysis model briefly,focusing on the implementation process of the latent class logit model in Stata through an example to provide methodological reference for the practical application of the model in the discrete choice experiments.Methods Based on the discrete choice experiment data of HIV self-testing reagents choice preferences among people at high risk for HIV in Guangzhou,we demonstrated the construction process of latent class logit analysis model by an example and provided the corresponding Stata commands.Results Finally,we established 4 Classes models as the optimal class,and the model estimation results show that the difference of Class 1(prefer urine reagents),Class 2(prefer cheaper blood reagents with instructions and judging results by themselves),Class 3(prefer cheaper urine reagents),and Class 4(prefer blood reagents with video and judging results by professionals)are statistically significant.Conclusion The latent class logit model for the analysis of discrete choice experiment data has the simplicity and flexibility that is superior to other models,but also has its limitations in application.Therefore,it needs to be further combined with other models to optimize the analysis.

关 键 词：离散选择实验 偏好异质性 潜类别logit模型 

分 类 号：R195.1[医药卫生—卫生统计学]