基于超级学习的童年期状况与老年痴呆发病风险的因果推断研究

童年期状况对老年痴呆发病具有长期而深远的影响，然而相关研究面临着时间跨度长、混杂因素依时变化和影响机制错综复杂等严峻挑战，导致目前研究较少且难以阐明其因果关系。生命历程流行病学理论提出关键期、路径和累积三种假说，试图解释其因果关系。本项目在前期老年健康和老年痴呆调查基础上开展纵向调查，了解童年期状况、老年痴呆等的流行现状和趋势；设定逆概率权重，选取多种预测性能好的机器学习模型，通过交叉验证法训练得到个体学习器，并集成为最优的超级学习组合，估计逆概率权重，构建基于超级学习的边际结构模型；将该模型应用于分析童年期社会经济状况、不良经历和营养状况对老年痴呆发病风险的因果效应，进而探索老年期不良社会关系、心理障碍和健康风险行为因素的中介和调节效应，以验证生命历程的三种假说。本研究将阐明童年期状况与老年痴呆发病风险的因果效应，从生命历程视角为揭示其影响机制奠定基础，为早期防控老年痴呆提供新的思路。

Childhood conditions have a long-term and far-reaching impact on the onset of dementia. However, related studies are faced with severe challenges such as long time spans, time-varying confounding factors, and complicated influencing mechanisms. As a result, there are limited studies and it is difficult to clarify the causality. The theory of life course proposes three hypotheses of critical period, path and accumulation, trying to explain the causality. This project carried out a longitudinal survey based on the prior elderly health survey and dementia survey to understand the prevalence and trends of childhood conditions and the dementia; using super-learning methods, select a variety of machines Learning models with good predictive performance, which are trained by cross-validation to obtain individual learners and integrated into the optimal super-learning combination to estimate the inverse probability weights, construct a super-learning-based marginal structure model; We apply this model to analyze the causal effect of socio-economic status in childhood, adverse experiences, and nutritional status on the risk of dementia, and then explore the mediating and moderating effects of socioeconomic status, adverse social relationships, psychological disorders, and health risk behavioral factors in old age, trying to verify three hypotheses of the theory of life course. This study will effectively infer the causal relationship between childhood status and the risk of dementia, lay the foundation for revealing its influence mechanism, and provide new ideas for early prevention and control of dementia.