Machine learning techniques for passive, remote monitoring of elderly heart failure patients from home

摘要 心力衰竭（HF）是医疗保险计划中住院和再入院的最常见原因。 HF的高死亡率和通过再入院的高住院利用率导致了巨大的经济损失。 目前估计的负担超过300亿美元，预计到2030年HF的患病率将继续上升46%。 有机会部署远程患者监测（RPM）工具，用于测量预后生物标志物， HF恶化和急性失代偿和再次入院。Bender Tech（BT）开发了一种尿液 一种能够容易地连接到家庭马桶上的测试平台，用于准确和容易地收集纵向健康 和行为数据，而不需要人工样本收集和/或测试。我们建议调整我们的 通过使数据收集和传输完全被动，用于老年HF患者群体的系统。 我们建议集成传感器和固件，能够识别用户何时使用他们的家庭厕所 （具体目标1），开发必要的机器学习（ML）分类算法，以确定何时 执行测试序列（特定目标2），并使用ML方法来证明用户生物特征识别的可行性 通过尿液检测数据进行识别（具体目标3）。这一建议的一个成功结果将是一套 使用ML技术构建的分类模型，旨在实现纵向尿液特征的被动收集 用于远程管理老年HF患者的来自家庭厕所的数据。这将使产品准备好 前瞻性临床试验，将成为未来II期申报的主题，用于远程监测 利尿有效性和防止再入院。

Abstract Heart failure (HF) is the most common cause for both hospitalizations and readmissions in the Medicare program. HF’s high mortality rate and high hospitalization utilization rate via readmissions results in a large economic burden currently estimated at over $30B, and prevalence of HF is expected to continue rising by 46% by 2030. There is an opportunity to deploy remote patient monitoring (RPM) tools for measuring prognostic biomarkers of worsening HF and acute decompensation and hospital readmission. Bender Tech (BT) has developed a urine testing platform capable of easily attaching to a home-toilet for accurate and easy collection of longitudinal health and behavior data without the requirement of manual sample collection and/or testing. We propose to adapt our system for use in elderly HF patient populations by rendering data collection and transmission completely passive. We propose to integrate sensors and firmware capable of identifying when a user has used their home toilet (Specific Aim 1), develop the machine learning (ML) classification algorithms necessary for determining when to perform a testing sequence (Specific Aim 2), and use ML methods to demonstrate feasibility of user biometric identification via urinary testing data (Specific Aim 3). A successful outcome of this proposal will be a set of classification models built using ML techniques designed to enable passive collection of longitudinal urine profile data from a home-toilet for use in remotely managing elderly HF patients. This will ready the product for prospective clinical trials that would be the subject of a future phase II submission for use in remotely monitoring diuretic effectiveness and preventing hospital readmissions.