Collaborative Research: Integrative Heterogeneous Learning for Intensive Complex Longitudinal Data

该项目涉及与生物医学信息有关的几个基本统计问题。这些问题出现在复杂的生物医学研究中，当数据呈现高度异质性和决策阶段的数量是大的。研究人员的目标是开发新的统计方法和有效的计算工具，以解决实际应用，如创伤后应激障碍（PTSD）的治疗和使用移动的健康数据管理糖尿病。这项研究预计将在医疗保健方面发挥作用，并促进神经科学、心理健康、护理、传染病、表观遗传学、生物学和计算机科学的跨学科研究。将要开发的软件将广泛分发，供工业伙伴使用。该项目通过研究生参与研究提供培训。调查人员将研究三个研究课题。第一个动机是识别PTSD患者的异质性表观遗传效应。这项研究将通过开发一种新的模型来识别高维DNA甲基化介质，从而推动当前的界限。第二个主题的动机是移动的健康技术的最新进展，它有效地监测个人的健康状况，并提供个性化的治疗。"价值提升"的概念被用来选择最佳的治疗方法。当决策阶段数可以发散到无穷大时，研究者将研究一种新的估计量。最佳方案可以是稀疏的和自适应的，如果有许多治疗方案，则更有利。它也将是强大的意外情况，如临时药物短缺或预算限制。在第三个主题中，研究者计划开发一种双编码器方法，通过合并联合治疗产生的协同或拮抗效应等相互作用效应来估计最佳的全渠道个体化治疗规则。该奖项反映了NSF的法定使命，通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

This project addresses several fundamental statistical questions related to biomedical information. These questions arise in complex biomedical studies when the data present high heterogeneity and the number of decision stages is large. The investigators aim to develop new statistical methods and efficient computational tools to address practical applications such as treatment of post-traumatic stress disorder (PTSD) and the use of mobile health data in management of diabetes. The research is expected to be useful in health care and stimulate interdisciplinary research in neuroscience, mental health, nursing, infectious diseases, epigenetics, biology, and computer science. The software to be developed will be widely disseminated for use by industry partners. The project provides training through research involvement for graduate students. The investigators will study three research topics. The first is motivated by identifying heterogeneous epigenetic effects of PTSD patients. This research will push the current boundaries by developing a new model to identify high-dimensional DNA methylation mediators. The second topic is motivated by recent advances in mobile health technology, which effectively monitors individuals' health statuses and delivers personalized treatment. The concept of "value enhancement" is used to select the optimal treatment. The investigators will study a novel estimator when the number of decision stages can diverge to infinity. The optimal regime could be sparse and adaptive, making it more advantageous if there are many treatment options. It will also be robust to unexpected situations such as temporary medication shortages or budget constraints. In the third topic, the investigators plan to develop a double encoders approach to estimate the optimal omni-channel individualized treatment rule by incorporating interaction effects such as synergistic or antagonistic effects due to combination treatments.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.