慢乙肝患者治疗队列医疗费用及群体异质性的混合模型研究

Hepatitis B virus (HBV) infection is a major public health problem in China, which brings a great financial burden to patients and their families and seriously hinder the economic and social development. The cohort of chronic hepatitis B (CHB) patients from hospital information system could reflect medical cost in a real clinical environment, which is the ideal data source for pharmacoeconomic evaluation. However, it has been well documented that the modeling and predicting of cost data is often problematic due to its strong positive skewness, excess zeros, multimodality, and heavy right tails. The feature of disease characteristics and treatment mode of CHB patients, such as high-dimensional data, complex types of variables, and serious collinearity also bring a great challenge in statistical analysis. An additional problem in predicting costs over time is that cost measured on the same individual at different time points is usually correlated. When analyzing the cost data, traditional analytic methods only focus on measurable variables, with ignoring the internal relationship between variables and overall effect of factors, and thus limiting the reference values of decisions. Herein, from the point of view of the population heterogeneity between patients, we will comprehensively evaluate the medical cost for CHB patients and thoroughly explore its influencing factors in the framework of models based on mixtures of parametric distribution：(1) THe finite mixture model will be applied to estimate cost; (2) The latent class model will be used to detect the homogeneity of clinical characteristics and treatments mode, and subgroup analysis and sensitivity analysis will be conducted to explore the cluster effect of influencing factors on medical cost for CHB; (3) The growth mixture model will be used to reveal the total and individual growth trajectories of medical costs, and thereby the prediction model for the medical cost in various subgroup will be precisely constructed. This study will provide valuable clues for reasonably allocating of medical resources, and will offer indicative reference for the specific medical decision for CHB patients.

乙型肝炎是我国重大公共卫生问题之一，给个人、家庭、社会造成沉重经济负担，严重影响我国社会经济的发展。基于医院的慢乙肝患者队列能反映真实环境下的医疗成本，是药物经济学评价中理想的数据来源。但实际分析时医疗费用呈现极度正偏峰、厚尾和多态分布，患者疾病特征和治疗方式表现出维度高、类型复杂和严重共线性，队列具有层次结构和纵向变化等特征。传统分析方法只关注可测变量，忽视变量间内在联系和因素的整体效应，极大地影响了研究的决策价值。本项目以患者的“群体异质性”为切入点，提出在混合模型框架下基于慢乙肝患者治疗队列研究医疗费用：采用有限混合模型估计费用；构建潜在类别模型探测疾病特征和治疗方式的同质性潜在亚组，深入解析慢乙肝治疗成本影响因素的组群效应；构建增长混合模型揭示不同群体慢乙肝患者医疗费用的总体和个体增长轨迹，并籍此建立不同亚群的医疗费用预测模型，为合理配置医疗资源及制定特异性的决策方案提供重要依据。

乙型肝炎是我国重大公共卫生问题之一，给个人、家庭、社会造成沉重经济负担，严重影响我国社会经济的发展。基于医院的慢乙肝患者队列能反映真实环境下的医疗成本，是药物经济学评价中理想的数据来源。但实际分析时医疗费用呈现极度正偏峰、厚尾和多态分布，患者疾病特征和治疗方式表现出维度高、类型复杂和严重共线性，队列具有层次结构和纵向变化等特征。本项目以患者的“群体异质性”为切入点，在混合模型框架下基于慢乙肝患者治疗队列研究医疗费用：通过类蒙特卡洛模拟实验显示有限混合模型在估计慢乙肝和肝硬化患者医疗费用时可将异质性群体分为同质性亚组，有效提高医疗费用估计精度；利用两部模型拟合慢乙肝患者抗病毒药物的利用率、抗病毒治疗成本及其影响因素，结果表明两部模型适用于部分受限数据，医保政策是影响抗病毒治疗的最主要因素；构建潜在类别模型和SOM神经网络聚类模型探测慢乙肝患者疾病特征和治疗方式的同质性潜在亚组，利用类别变量对直接医疗成本建立预测模型，预测性能优于传统广义线性模型；采用分位数回归研究慢乙肝患者直接医疗费用及影响因素，结果显示医保政策和治疗方案是直接医疗费用最主要的影响因素；在生存分析框架下结合倾向评分调整，评价慢乙肝患者各种抗病毒治疗的远期疗效，并进一步采用倾向性评分简单加权估计分析各种抗病毒治疗方案的成本效果，结果显示INF和ETV的远期疗效最优，但LAM+ADV初始联合治疗方案最具成本-效果，其次为ETV。构建潜变量增长混合模型揭示不同群体慢乙肝患者医疗费用的总体和个体增长轨迹，结果表明模型能够识别慢乙肝初治患者治疗队列中潜在亚类医疗费用变化的异质性。基线HBV DNA水平和早期使用抗病毒药物是直接医疗费用轨迹的最主要决定因素。本项目为横断面和纵向设计的医疗成本相关研究提供有效的统计方法学工具，同时为慢乙肝疾病的临床诊治、疾病费用控制提供良好数据支持，为合理配置医疗资源及制定特异性的决策方案提供重要依据。