An AI-assisted strategy for monitoring pulmonary congestion in acute heart failure patients in emergency settings - acute heart failure (AHF) is the second most common cause of hospitalization in…

An AI-assisted strategy for monitoring pulmonary congestion in acute heart failure patients in emergency settings - acute heart failure (AHF) is the second most common cause of hospitalization in the united states. Pulmonary congestion, resulting in dyspnea, is the primary reason why patients present to the hospital seeking care. Relief of congestion is a key goal of therapy. Incomplete decongestion increases the risk of short-term death or RE- hospitalization three-fold. Despite this, nearly 50% of patients are reported to leave the hospital congested. There is a critical unmet need to accurately monitor congestion. While traditional measures of congestion - physical exam, body weight, urine output, natriuretic peptides - remain important, patients continue to be discharged with congestion. An accurate, reliable, efficient, and facile method to measure congestion is needed. Lung ultrasound (LUS) is a novel, low-cost, easy to use, bedside tool with superior accuracy for detecting congestion compared to conventional measures. Preliminary data from our clinical studies blushed-AHF and carvd-AHF suggest: 1) daily LUS detects changes in congestion severity more rapidly compared to conventional measures, and 2) guiding decongestive therapy based on LUS results is feasible. Our preliminary findings provide validation for the link between inadequate decongestion, as evaluated by LUS, and an increased incidence of clinically significant adverse events. Furthermore, our data highlights a notable challenge in the field of LUS-based congestion monitoring, which is the necessity for a human rater to ensure accurate and consistent quantitative scoring. Our long-term goal is to improve the measurement of decongestion for AHF patients. In this project we propose to automate LUS congestion scoring using innovations in AI/ML methods and a unique high-quality dataset of heart failure patients. We will use the results to investigate the additive benefit of AI-automated LUS congestion scoring to the current clinical practice of utilizing physical exam, history, blood tests, and chest radiography reports. Successful completion of this work would change clinical practice. The current limitations of congestion monitoring, including with LUS, are reliability and objectivity related to rater experience and information loss across transitions of care. If successful, this work would significantly impact AHF management, elevating the role of LUS. An open-source prototype of the AI software for congestion measurement, pulmox, would be publicly released for use in future clinical trials for novel decongestion therapies.