准确的痴呆患病率估计是客观评估痴呆疾病负担的重要基础，对于合理布局痴呆等疾病的防治资源具有重要意义。但是目前采用的两阶段横断面调查实施难度大，第二阶段无应答率往往超过35%，影响了患病率的准确估计。于是，第二阶段痴呆诊断缺失的估计和填补成为了关注的焦点。本研究前期基于中国老年退伍军人慢性非传染性疾病临床研究平台（CVCR）的两阶段痴呆患病率调查，比较了应答对象与一、二阶段无应答对象，发现在人口学特征、认知功能上均存在差异。二阶段调查对象认知功能更差、无应答率更高，因此不考虑无应答，或将一、二阶段无应答合并考虑均不能很好的校正患病率。本研究计划依托CVCR平台数据，基于一、二阶段数据完整的数据集，模拟出二阶段无应答对象，建立模拟数据集。以此验证常用的统计学模型、机器学习模型在估计二阶段缺失诊断、校正患病率中的准确性；利用该数据集探索基于LRS理论在缺失诊断估计和患病率校正中的可行性。

It is important to estimate the prevalence of dementia accurately, in terms of objective assessment of the burden of dementia and appropriate allocation of resources for prevention and treatment. However, it is difficult to execute the two-stage cross sectional survey and the non-response rate in the second stage is more than 35%, which significantly impact the accurate estimation of dementia prevalence. As a result, it becomes the hot pot to find out the approach to fill in the missing date and estimate the missing diagnosis of dementia in the second phase. In this study, based on the platform of China Veterans Clinical Research（CVCR）, which aim to investigate the prevalence of dementia through two-stage method, we compared the subjects who response in both 2 stages and non-response in first or second stage and found that there are differences on both demography and cognitive function among these different subgroup subjects. Basically, the cognitive function is worse and non-response rate is higher in second stage than in first stage, hence the prevalence will not be estimated well if investigators do not consider the non-response rate or only consider the general non-response rate combined the two stages. In this study, based on CVCR platform, we will use database without any missing data in both two stages to simulate non-response subjects in second stage and establish simulation data set. Then we will verify the accuracy of estimated prevalence by the common statistic model and statistical machine learning model; and explore the feasibility to estimate the missing diagnosis and adjust the prevalence by LRS theory.