Building predictive algorithms to identify resilience and resistance to Alzheimer's disease

项目摘要 有两种观察到的现象与传统的阿尔茨海默病（AD）轨迹背道而驰; 抵抗AD病理学（淀粉样蛋白和/或tau蛋白）的积累，尽管有风险因素的证据， 存在AD病理，但对认知能力下降保持弹性。将这些可能会 在他们的一生中对AD表现出的抵抗力或恢复力对于告知临床实践和转化至关重要。 临床试验招募。目前尚不清楚风险因素的组合，无论是人口统计学，血管或 神经成像，可能有助于提高预测个体表现出耐药性的可能性的准确性 或对AD的恢复力。此外，很少有人了解性别，种族及其相互作用如何影响这些 现象。迄今为止，相对有限的样本量和低种族多样性阻碍了研究。整体 该提案的目标是通过以下方式开发和验证对AD的抵抗力和恢复力的鲁棒预测算法 协调来自13个特征良好且种族多样的临床正常老年人队列的数据 （n=~ 15，000）。这一创新的建议可以改变临床决策和临床治疗的方法。 审判基于一组简单的易于获取的医疗信息，如人口统计学、血管风险， APOEe 4状态和脑体积数据（如果可用），我们验证的模型将提供可解释的患者- 抵抗力和恢复力的水平预测以及AD病理负担和认知能力的10年风险估计 根据病人的情况拒绝同样，我们的预测算法将提供一个预测框架， 应邀请进行初步筛选，并用于预测那些最有可能积累Ab/tau或表现出短 在临床试验过程中的长期下降。我们建议协调来自13个队列约15，000的数据 临床上正常的个体，以实现以下目标：（1）建立预测算法来分类那些 对淀粉样蛋白或tau蛋白具有抗性的人，并验证这些模型以证明它们在临床实践中的实用性 和AD预防试验，（2）建立和验证预测算法，以分类那些认知弹性 面对淀粉样蛋白或tau蛋白的异常水平，以及（3）检查性别和种族之间的交叉如何能够 产生更精确的个体化风险特征，反映这两个关键的人口阶层， 已知的AD风险因素。我们强大的跨学科团队跨越认知神经科学的广度，PET 以及MR神经成像、生物统计学、行为神经学和流行病学。我们的多PI团队反映了四个 对本提案至关重要的关键专业领域：（1）数据协调，（2）神经成像，（3） 机器学习;（4）认知弹性。我们发布了一系列数据协调方法 用于认知和PET神经成像数据，可以灵活地应用于不同的数据类型。使用这些 方法，我们将确定多个风险因素（人口统计学，血管风险， 神经成像），以建立对AD的抵抗力和恢复力的个体化风险概况。这一创新 该提案有可能改变我们对待临床实践和临床试验设计的方式。

PROJECT SUMMARY There are two observed phenomena that defy the traditional Alzheimer’s disease (AD) trajectory; those who resist the accumulation of AD pathology (amyloid and/or tau) despite evidence of risk factors, and those who present with AD pathology but remain resilient to cognitive decline. Classifying these individuals who will likely manifest resistance or resilience to AD over their lifetime is critical for informing clinical practice and transforming clinical trial recruitment. It remains unclear how combinations of risk factors, whether demographic, vascular or neuroimaging, may help to increase accuracy for predicting an individuals’ likelihood of manifesting resistance or resilience to AD. Further, very little is understood about how sex, race and their interaction influence these phenomena. Relatively limited sample sizes and low racial diversity have so far hampered studies. The overall goal of this proposal is to develop and validate robust predictive algorithms of resistance and resilience to AD by harmonizing data from 13 well characterized and racially diverse cohorts of clinically normal older adults (n=~15,000). This innovative proposal could transform approaches for both clinical decision making and clinical trials. Based on a simple set of easily accessible medical information, such as demographics, vascular risk, APOEe4 status, and brain volumetric data when available, our validated models will provide interpretable patient- level predictions of resistance and resilience with 10-year risk estimates of AD pathological burden and cognitive decline given a patient’s profile. Similarly, our predictive algorithms will provide a predictive framework of who should be invited for initial screening and serve to predict those most likely to accumulate Ab/tau or exhibit short term decline within the course of a clinical trial. We propose to harmonize data from 13 cohorts of ~15,000 clinically normal individuals, to accomplish the following aims: (1) build predictive algorithms to classify those who are resistant to either amyloid or tau and validate these models to demonstrate their utility in clinical practice and AD prevention trials, (2) build and validate predictive algorithms to classify those who are cognitively resilient in the face of abnormal levels of amyloid or tau, and (3) examine how intersections between sex and race can produce more refined individualized risk profiles that are reflective of these two critical population strata that are known risk factors for AD. Our strong interdisciplinary team spans the breadth of cognitive neuroscience, PET and MR neuroimaging, biostatistics, behavioral neurology, and epidemiology. Our multi-PI team reflect four critical areas of expertise that are essential to this proposal: (1) data harmonization, (2) neuroimaging, (3) machine learning, and (4) cognitive resilience. We have published a range of data harmonization approaches for both cognitive and PET neuroimaging data, which can be flexibly applied to different data types. Using these approaches, we will identify higher-order interactions between multiple risk factors (demographics, vascular risk, neuroimaging) to build individualized risk profiles of both resistance and resilience to AD. This innovative proposal has the potential to transform the way we approach clinical practice and clinical trial design.