Statistical methods in mHealth to signal interventional needs for mental health patients

项目摘要 随着智能手机普及率的增长，它们作为可扩展健康监测的潜力也在增长 治疗精神疾病的工具。有自杀意念的个体的行为警告信号， 到目前为止，躁郁症、进食障碍、抑郁症、精神分裂症和其他精神障碍 在不良事件发生之前，如自杀未遂或复发，很难识别。 数字表型，个体水平的人类表型的时刻量化 在……里面 使我们能够量化这些警告信号，并提示 适时的干预。目前发表的变化点和异常检测在数字上的应用 到目前为止，表型数据一直是证明数字表型潜力的主要研究 用于行为和健康监测。这项提案旨在推进的更广泛的目标可以描述为 分三步走，按照以下具体目标排序。目标1：开发新的统计方法 能够考虑纵向特征的变化点和异常检测方法 数据缺失的普遍和普遍模式。目标2：开发降维技术以改进 统计功率和减少数字表型中的噪音。这将大大提高 目标1中提出的方法。对这两个目标都至关重要的是发展计算效率 软件。目标3：通过我们正在进行的和新的协作，在患者群体中实施此软件 以便在上传新数字表型数据时对其进行分析，并在以下情况下提供临床医生通知 检测到行为警告信号。这最后一步是拟议工作的最终目标，也是成功的 完成后将对患者的健康产生立竿见影的影响，使干预措施能够防止 各种各样的上瘾和障碍。利用我们在统计方法、数字表型和 软件开发，结合我们广泛的数字表型合作网络，我们做得很好 定位于开发识别行为警告信号所需的统计方法和软件 从数字表型数据中获取数据，并通过合作研究实施这些方法。成功 该项目的完成将对全球个性化医疗和移动医疗产生立竿见影的影响 治疗精神疾病。 使用来自个人数字设备的数据

Project Summary As smartphones have grown in prevalence, so too has their potential grown as a scalable health monitoring tool for the treatment of psychiatric disorders. Behavioral warnings signs in individuals with suicidal ideation, bipolar disorder, eating disorders, depression, schizophrenia, and other psychiatric disorders have, until this point, been difficult to identify prior to the occurrence of an adverse event, such as a suicide attempt or relapse. Digital phenotyping, the moment-by-moment quantification of the individual-level human phenotype in situ , has enabled us to quantify these warnings signs and prompt an appropriately-timed intervention. Current published uses of change point and anomaly detection on digital phenotyping data so far have been proof-of-principal studies demonstrating the potential of digital phenotyping for behavioral and health monitoring. The wider goal that this proposal aims to advance can be characterized in three steps, which are ordered according to the following specific aims. Aim 1: Develop novel statistical methods for change point and anomaly detection capable of accounting for longitudinal features with widespread and general patterns of missing data. Aim 2: Develop dimensional reduction techniques to improve statistical power and reduce noise in digital phenotypes. This will greatly improve the performance of the methods proposed in aim 1. Crucial to both of these aims is the development of computationally efficient software. Aim 3: Implement this software on patient populations through our ongoing and new collaborations so as to analyze new digital phenotyping data as it is uploaded and provide clinicians notifications when behavioral warning signs are detected. This final step is the ultimate goal of the proposed work, as successful completion will lead to an immediate impact on patient health, enabling interventions to prevent relapse in a wide variety of addictions and disorders. Using our expertise in statistical methods, digital phenotyping and software development, combined with our wide network of digital phenotyping collaboration, we are well positioned to both develop the statistical methods and software necessary to identify behavioral warnings signs from digital phenotyping data, as well as implement these methods through collaborative studies. Successful completion of this project will have an immediate impact on personalized medicine and mobile health in the treatment of psychiatric disorders. using data from personal digital devices