Statistical methods for biosignals with varying domains

描述（由申请人提供）：临床护理和大型观察性研究的特点是在医院就诊期间进行密集的健康监测，然后在就诊之间进行长时间的低强度或无监测。住院期间获得的数据来自许多新技术，例如非常密集采样的生物信号记录（脑电图、心电图、健康评分）和高分辨率多模态成像（MRI、CT、PET）。此类数据的一个主要特征是它是在特定主题的一段时间内收集的。事实上，不同受试者的住院时间和监测量各不相同，并且对于研究医院内和出院后的健康结果都具有丰富的信息。许多例子之一是最近一项针对因急性呼吸窘迫综合征 (ARDS) 入住重症监护病房 (ICU) 的受试者的研究。对于每个受试者，在 ICU 住院期间每天收集序贯器官衰竭评估 (SOFA) 评分，这是一种常用的评分系统，用于测量 ICU 中的器官功能障碍。 ICU 长度为 停留时间因主题而异，并且可能对当前和未来的健康结果提供大量信息。在此应用程序中，一系列相关问题被概念化并提炼为统计目标，以解决与此类数据采样相关的特定复杂性。具体来说，该提案解决了收集高密度生物信号的研究中以下基本未解决的问题：1）引入统计模型来分析 高密度生物信号，支持和健康结果不均匀； 2）开发功能性预测配准模型，改变生物信号的支持，以提供健康结果的最佳预测； 3）开发模型来描述在罕见但强烈的健康监测研究中获得的生物信号的横截面和纵向变异性。虽然重点在于在特定主题的时间长度内收集准连续超高分辨率生物信号的研究，但方法将可推广到具有类似数据采样结构的许多其他研究。 2

DESCRIPTION (provided by applicant): Clinical care and large observational studies are characterized by periods of intense health monitoring during hospital visits followed by long periods of low-intensity or no-monitoring between visits. Data obtained during in-hospital visits come from a host of new technologies, such as very densely sampled biosignal recordings (EEG, ECG, health scores) and high resolution multi-modality imaging (MRI, CT, PET). A major characteristic of this type of data is that it is collected for a period of time that is subject-spcific. Indeed, the in-hospital length and amount of monitoring varies between subjects, and is highly informative both for studying health outcomes in the hospital and after discharge. One among many examples is a recent study of subjects admitted to the Intensive Care Unit (ICU) with Acute Respiratory Distress Syndrome (ARDS). For each subject the Sequential Organ Failure Assessment (SOFA) score, a commonly- used scoring system to measure organ dysfunction in the ICU, was collected daily for each subject for the duration of their ICU stay. The ICU length of stay is different by subject and likely to be highly informative of current and future health outcomes. In this application, a set of relevant problems are conceptualized and distilled to statistical aims to address specific complexities associated with this type of data sampling. Specifically, the proposal addresses the following fundamental unsolved problems in studies that collect high density biosignals: 1) introducing statistical models for the association between high density biosignals with uneven support and health outcomes; 2) developing functional registration-by-prediction models that transform the support of biosignals to provide best prediction of health outcomes; and 3) developing models for describing the cross-sectional and longitudinal variability of biosignals obtained in studies with rare -but intense- health monitorin. While focus lies on research studies that collect quasi- continuous ultra-high resolution biosignals for subject-specific lengths of time, methods will be generalizable to many other studies with similar data sampling structures. 2