基于贝叶斯竞争风险理论的农村人群脑卒中发生和转归的适宜性风险预警策略与方法研究

The heavy burdens of stroke were seen shifting from urban to rural areas in China. Without an assessment tool and method for accuraterisk prediction, effective early detection and intervention becomes infeasible. Because Framingham modelwas designed many years ago in the United States, its limitation in study design, variable selection, and risk calculation makes it not suitable forpredicting risks of stroke and other cardiovasculardiseases among rural populations of China. In this project, following the principal of suitability and adaptability, we propose to develop a new strategy of risk prediction using Bayesian hierarchical competing-risk modeling approach. This innovative strategy is able to accommodate data from multiple sources with different epidemiologic study designs (i.e., a multi-center cohort, a case-control study, and a stroke registry). By dividing lifespan into pre-stroke and post-stroke phases, within the framework of Bayesian hierarchical modeling with random-effects indescribing the heterogeneity between sub-populations, we willdevelop risk-competing models for predicting the risks of different types of stroke and those of their prognostic outcomes. The method of Bayesian variable selection will be used in selecting significant predictors from a list of risk factors that are pre-categorized into sub-groups according to their cost and convenience in acquisition and clinical relevance. By developing a set of candidate models, each with distinctive combination of risk factors and baseline risk function estimates, the approach of Bayesian Model Averaging will be applied to calculate averaged absolute risks, lifespan risks, and relative absolute risksto jointly describe risks of a target individual or community. By adopting this strategy, we aim to study the feasibility of developing the next-generation flexible, adaptable and suitable risk assessment methods and computing tools (instruments and computer software packages) to specifically serve varying needs of different rural populations in China. Accomplishment of this project will offer evidence in developing new methods and tools for effectively and adaptively predicting risks of not only stroke but other chronic diseases in rural area of China.

脑卒中严重负担状态已由城市转向农村，因缺乏其适宜风险预警工具而难以尽早干预。Framingham模型为代表的预测工具对农村人群缺乏适宜性且其设计构思、指标筛选、风险计算、建模方法均有局限性。本项目提出风险预警适宜性原则和贝叶斯竞争风险建模策略。依此，针对不同设计（多中心队列、全人群病例对照、病例随访队列）进行创新研究：构思上，将脑卒中自然史划为发生和转归两阶段分别建模；指标上，用适宜性分级策略遴选风险预警因子；风险计算上，将绝对风险、终身风险和相对绝对风险相结合，全面评估个体和群体风险状态；建模上，基于贝叶斯多水平竞争风险模型，利用贝叶斯模型选择，通过调整模型参数先验和建立备选模型集以适宜于不同目标人群；实践上，将复杂模型开发为简易量表工具及软件包。预期目标是以脑卒中为例建立慢病风险预警新方法，构建农村人群脑卒中发生和转归的适宜预警模型、量表工具及软件包；为慢性病风险评估提供新策略新方法。

本项目主要完成如下5项工作：1）建立了3个研究队列：建立了山东省多中心慢性病大型人群观察队列。同时在这些地区构建了大气复合污染实时云采集数据库。建立大型城市人群健康管理纵向监测队列。2）开发完成了预警指标采集工具： “预期寿命危险因素累积标尺”暴露危险因素采集法。3）筛选出了若干脑卒中及其相关疾病的新的预测因子：①证明了纤维蛋白原是预测女性代谢综合征的预测因子。②证实了红细胞参数（红细胞计数、血红蛋白含量和红细胞压积等）是代谢综合征的预测因子。③首先阐明了非酒精性脂肪肝是导致代谢综合征的独立危险因子，进而发现代谢综合征也是导致非酒精性脂肪肝的独立危险因子，最后，基于因果图模型系统阐明了非酒精性脂肪肝与代谢综合征的双向因果作用机制。④证实了育龄期妇女鸡蛋摄入量增加对绝经期后冠心病的发生具有强烈的保护作用⑤证实了高密度脂蛋白是预测冠状动脉粥样硬化的良好预测因子⑥发现甲状腺功能减退不仅是血脂异常的独立危险因子，而且与慢性肾病相关。.4）构建了脑卒中及其心脑血管病风险预警模型体系：①构建了网络结构驱动的疾病风险预测模型，为提高预测精度提供了新策略。②通过探索性因子分析方法进行正交独立旋转，进而在竞争风险框架下，建立了心脑血管病的新型预测模型③建立了中国城市汉族成年人群的高尿酸血症预测模型。④采用贝叶斯模型选择和模型平均策略，构建了筛选基因标记的新型策略。⑤构建了中国汉族人群首个心房颤动预测模型。⑥构建了脑卒中风险预测模型。⑦采用多中心竞争风险模型理论，构建了多中心疾病竞争风险预测模型(MCCRM)。⑧将贝叶斯推断理论进一步应用于上述多中心竞争风险模型中，拟构建更加高效的适用于不同地区不同人群的脑卒中及心脑血管病预测模型。.5）成果推广应用：将上述研发的各种脑卒中及心脑血管病及其相关疾病预测模型，开发成为在线“山大-康平健康/疾病风险评估及个性化健康干预系统”，并已在全国多省市示范推广应用。