老年多病共存现象普遍。传统流行病学方法在探索共病发生发展规律时存在局限。伴随医疗信息化的飞速发展，如何利用大数据资源对多病共存开展研究日益得到关注。本项目结合医学信息学、流行病学和大数据分析方法，利用医疗保险数据和北京大学附属医院电子病历数据的分析挖掘我国老年人多病共存特点，挖掘共病衍变规律，构建基于疾病发展轨迹的风险预测模型,具有重要理论和应用价值。研究内容包括：关联规则学习共病组合模式；隐马尔可夫模型探索共病发展规律，隐马尔科夫回归聚类模型建立精确细分的共病模式转变模型；半监督学习方法构建可预测临床结局的多层贝叶斯网络模型，利用老年多病共存临床队列数据验证并优化该网络模型，并通过与现有预测模型进行比较，评价预测效果。本研究数据源于真实世界管理型大数据，突破了传统研究方法对疾病间复杂关系的理解，对在当前老龄化的严峻形势下实现以患者为中心的个性化诊疗模式并提高医疗服务连续性意义重大。

Multimorbidity among older adults has become a common phenomenon. Traditional epidemiological methods have limitations in exploring how multiple chronic conditions emerge. With the rapid development of informatics, application of big data resources to carry out research on multimorbidity has attracted increasing attention. Using data from medical insurance data and electronic medical record data of Peking University affiliated hospitals, we propose to explore the characteristics of multimorbidity, the emergence of multiple chronic conditions over time, and construct a risk prediction model based on multimorbidity trajectories. Our research methods will include: association rules learning patterns of disease combinations; Hidden Markov Model exploring multimorbidity trajectories and using Latent Regression Markov Clustering to establish the accurate transition models of multimorbidity pattern; using semi-supervised learning method to establish and optimize Bayesian network model for predicting clinical events such as hospital readmission and emergency admission, validating and optimizing the network model with clinical cohort from electronic medical records, and comparing it with existing risk prediction models. Research based upon real-world big administrative data breaks through the understanding for the complex relationships between diseases of the conventional research methods, and is of great significance for improving patient-centered treatment and the continuity of medical services.