近年来，我国糖尿病患病率呈持续上升趋势，其慢性并发症严重危害人民健康。故探讨其危险因素，对降低并发症的危害意义重大！“糖尿病随访研究数据”以“多相关生存结局”与“多重复测量协变量”为特点，“Bayesian联合模型”因具有综合生存结局和纵向协变量的“全面性”及处理复杂层次结构的“灵活性”的特点，参数估计更可靠，具有统计推断的“高效性”，为处理此类数据的更好选择。. 本研究基于2型糖尿病大型社区健康管理队列，建立随访子队列，定期收集人口学、饮食、运动、治疗干预措施及并发症等信息，结合生理、生化指标，建立多结局Bayesian联合生存模型，对糖尿病多重慢性并发症的发生时间和强度进行预测，用模拟技术评价方法性能，并在平行子队列中进行外推验证。本研究将为糖尿病人群的并发症危险因素确认、风险动态预测和干预措施评价等提供方法学支持，所建方法可推广到其他具有类似特征数据的研究。

The prevalence of diabetes has increased significantly in recent decades and the chronic complications of diabetes can do serious harms to Chinese population. It is of great importance to identify the potential risk factors for prevention of chronic complication and the related hazards..The following-up data from community-based cohort contain information on multiple correlated time-to-event outcomes and repeated-measurements of covariate, in which the traditional statistical methods are not appropriate. Bayesian joint modeling can consider multiple survival and longitudinal non-survival measurements simultaneously, to treat complicated hierarchical structure with great flexibility, and therefore yields higher efficiency in statistical inference, which justifies itself as a better choice compared to its traditional alternatives(e.g. cox proportional model with time-varying covariates)..The study will begin from exploring the baseline information of T2DM community cohort. Firstly, the sub-cohort for follow-up will be created by cluster sampling and multiple measurements including demographic characteristics, dietary, exercise and physiological and biochemical measurements will be collected dynamically. Then Bayesian joint survival model will be built and validated for prediction of multiple chronic complications. The constructed models will be assessed and validated in practical data as well as simulated data. This study will provide methodological supports for identifying the risk factors of chronic complications, providing individual dynamic prediction, and assessing the effect of community intervention to T2DM complications. The analysis strategy constructed from this research could be generalized to other studies with similar data features.