基于分子标志物联合构建阿尔茨海默病风险预测模型的巢式病例对照研究

Alzheimer's disease (AD) is often hidden from onset. The existing AD early diagnosis and prediction methods are evasive, expensive and have low accuracy. If we could build a prediction method that is cheap, convenient and accurate, it would be of great significance. On the basis of previous works of epidemiology and bio-informatics, we study AD prediction model, depending on the cohort which including more than 3000 elderly cases having had 3 years’ fellow-ups. Our main works: Through evidence-based medicine method, we make further confirmation of the molecular markers (SNPS, proteins, metabolites), and determine the molecular markers used in model through experiments. Using nested case-control design, through the SNP, ELISA and targeted metabolism technology, we measure expression of markers in baseline level. Using Logistic regression and generalized linear mixed effects model in horizontal and vertical comparison, combined with the information of epidemiology and follow-ups, we find differences among variables, and we build and verify AD risk prediction index system. Based on the system, we set up AD risk prediction model by random forests method. Using cross validation method, we evaluate prediction performance of AD risk prediction model, comparing the importance scores of different indexes in model, exploring the interaction among different indexes. This study can provide scientific basis for predicting the risk of AD.

阿尔茨海默病（AD）起病隐匿，现有早期诊断及预测方法多数侵入性大、价格昂贵且准确性不高，如能构建一种经济、便捷、准确的预测方法，其意义重大。本项目拟在前期流行病学及生物信息学研究的基础上，依托已随访3年的3000余例老年人自然队列，开展AD预测模型研究，主要工作有：采用循证医学方法，对前期筛选的分子标志物（SNP、蛋白、代谢物）做进一步确认，并通过预实验做最终确定；基于巢式病例对照设计，利用SNP分析、ELISA、靶向代谢组技术检测选定标志物在基线血样中水平；将其与流行病学及随访信息融合，使用Logistic回归及广义线性混合效应模型进行横、纵向比较，寻找重要差异变量，建立并验证AD风险预测指标体系；基于该体系，通过随机森林构建AD风险预测模型，同时借用交叉验证的方法评估模型的预测效能，比较不同指标在模型中的重要性评分，探索不同指标间的交互作用。本项目可为AD发病风险的预测提供科学依据。

阿尔茨海默病（AD）起病隐匿，现有早期诊断及预测方法多数侵入性大、价格昂贵且准确性不高，如能构建一种经济、便捷、准确的预测方法，其意义重大。本项目拟在前期流行病学及生物信息学研究的基础上，依托已随访3年的3000余例老年人自然队列，开展AD预测模型研究，主要工作有：采用循证医学方法，对前期筛选的分子标志物做进一步确认，并通过预实验做最终确定；基于巢式病例对照设计，利用靶向组学技术检测选定标志物在基线血样中水平；将其与流行病学及随访信息融合，使用Logistic回归及广义线性混合效应模型进行横、纵向比较，寻找重要差异变量，建立并验证AD风险预测指标体系；基于该体系，通过随机森林构建AD风险预测模型。本项目可为AD发病风险的预测提供科学依据。