基于因果推断框架下的最优治疗方案选择的统计方法研究

Personalized treatment is one of hot research areas in modern medicine, which allows individual patients to select the optimal treatment according to their own characteristics. However, the existing statistical methods about the treatment selection mainly focus on overall average treatment effects or quantile treatment effects of various treatment methods, without considering how to elect the best treatment for a patient, based on the patient’s characteristics..In order to solve the above problems, based on the theory of causal inference, we plan to develop new semi-parametric statistical methods for optimal treatment selection. Those new methods are based on new concepts: CATE curve, CQTE curve, and the corresponding CATE curve’s confidence bands, CQTE curve’s confidence bands. .This project mainly focuses on the variable selection in the model, CATE, CQTE curve estimation and approximate large sample properties. The intersection points of CATE curve’s confidence bands divide the value range of a clinical marker will help a doctor to choose an optimal treatment for a given a patient.

个性化治疗就是现代医学最热门的研究方向之一。患者需要根据自己的病情选择最优的治疗方案，医生也需要根据各治疗方案的特点选择最适宜的患者进行个性化的治疗。个体化医疗研究设计和分析需要新的统计方法，然而现有的关于治疗方案的选择的统计方法主要是对各种治疗方法计算试验总体的平均因果效应或分位数因果效应，没有考虑如何利用患者的临床标记来选择最优的治疗方案。.为了解决上述问题，本课题提出基于因果推断框架下的半参数统计方法进行治疗方案选择。这套方法是基于CATE（Conditional Average Treatment Effect）曲线、CQTE（Conditional Quantile Treatment Effect）曲线以及对应的置信带。其中CATE曲线表示给定患者临床标记值的条件下治疗方案的平均因果效应；CQTE曲线表示给定临床标记值的条件下治疗方案的分位数因果效应. .本项目主要研究模型中变量的选择，CATE、CQTE曲线的估计以及近似大样本性质，通过CATE曲线、CQTE曲线的置信带的交点，对临床标记的取值做分段讨论，从而进行治疗方案的选择。

个性化治疗旨在根据个体的协变量特征，选择最优的治疗策略。本项目提出了协变量特异的因果作用（CSTE）曲线。通过单指标模型把协变量整合成单一指标，绘制CSTE曲线，可支持最优个性化治疗方案的选择。本项目研究了结局为二值变量或生存时间时的CSTE估计，推导了估计量的大样本性质，构建同时置信带，支持高维协变量以及存在未观测到的混杂变量。本项目用R语言编写了估计CSTE曲线和置信带的软件包。