复杂数据下半参数转换模型及其在老年慢性病发展中的应用研究

本项目的目标是，针对复杂数据（包括区间删失数据，半竞争风险数据，高维特征数据）给统计建模带来的挑战，探索新的半参数转换模型，并应用于老年慢性病发展的研究。本项目将开展三方面工作。一是，构建复杂删失半参数转换模型（STMCC），建立新的sieve极大似然估计方法并证明估计量的统计性质。二是，在高维特征与疾病发展呈线性的情形下，通过构建惩罚似然函数，建立基于STMCC的线性变量选择方法。三是，在高维特征与疾病发展呈非线性的情形下，通过构建深度学习惩罚似然损失函数，建立基于STMCC的深度学习变量选择方法。本项目方法能够适用于不同老年慢性病发展中的复杂删失情况，灵活涵盖众多模型假设，还可用于线性或非线性情形下高维特征的非结构和结构化变量选择。本项目聚焦老年慢性病发展研究中的复杂数据挑战，为揭示影响老年慢性病发展的危险因素提供科学支撑，为老年慢性病的预防和管理提供有力依据，有助于健康老龄化的实现。

The main goals of this project are to establish novel semiparametric transformation methods in response to the challenges brought by complex data (including interval-censored data, semi-competing risk data, and high-dimensional covariates) to statistical modeling, and apply them to the analysis of the progression of chronic diseases in the elderly. The project includes three parts. The first part involves establishing a semiparametric transformation model under complex censoring (STMCC), developing a new sieve maximum likelihood estimation method, and proving the statistical properties of the estimators. The second part is to establish variable selection methods based on STMCC by constructing penalized likelihood functions when the high-dimensional covariates are linear. The third part is to establish deep learning variable selection methods based on STMCC by constructing deep learning penalized likelihood loss functions when the high-dimensional covariates are nonlinear. These novel methods can readily handle complex censoring in the progression of various chronic diseases in the elderly, while flexibly incorporating various model assumptions and can be used for non-structured and structured variable selection when the high-dimensional covariates are linear and nonlinear, respectively. Overall, this project aims to tackle the complex data challenges that arise from chronic disease progression and deliver scientific support for revealing novel risk factors associated with chronic disease progression in the elderly, which contributes to the prevention and management of diseases, and the realization of healthy aging.