Worsening heart failure is accompanied by a build-up of fluid in the lungs. An AI smartphone app that picks up changes in a heart failure patient’s voice quality caused by this fluid accumulation and then alerts the physician about them – nearly 3 weeks before that ongoing decompensation would necessitate hospitalization and/or lead the physician to urgently introduce intravenous diuretics – is getting experts to sit up and take notice.
“In this incredibly prevalent waxing and waning condition, finding ways to identify worsening heart failure to prevent hospitalization and progressive disease is incredibly important,” observed American Heart Association (AHA)-appointed discussant David Ouyang, MD, assistant professor, Smidt Heart Institute, Division of Artificial Intelligence in Medicine, Cedars Sinai, Los Angeles. “Heart failure remains among the most common causes of hospitalization for older adults in the United States.
“The other standout feature is that we all use our cell phones on a daily basis,” Dr. Ouyang said at a late-breaking trial press briefing at the AHA 2023 annual meeting where results of the HearO Community Study were presented. “The ability to capture data from routine speech (patients speak five sentences into their phones every morning) is remarkable ... The HearO® technology was able to detect a substantial proportion of worsening heart failure events, with an average per individual of only three false positives over the course of a year. And, adherence to the study protocol was 81%. That’s higher than in many other kinds of routine patient monitoring studies,” he added.
Accumulating fluid changes speech
(e.g., pharynx, velum, tongue, and vocal folds). In the Israeli study, investigators enrolled 416 adults (75% were male, average age was 68 years) whose New York Heart Association (NYHA) 2-3 heart failure with either reduced or preserved ejection fraction was stable but placed them at-risk for heart failure events. The study goal was to analyze their speech data using the HearO® system to refine and test its ability to detect impending heart failure deterioration. Patients recorded five sentences in their native language (Hebrew, Russian, Arabic, or English) into the smartphone app daily. In a training phase of the study, distinct speech measures from 263 participants were used to develop the AI algorithm. Then, the algorithm was used in the remaining 153 participants to validate the tool’s effectiveness. In its ultimate form, once a deviation from the patient’s predefined baseline is detected, the app will generate a notice and send it to the health care practitioners.
Lead study author William T. Abraham, MD, FAHA, professor of medicine, physiology, and cell biology; and a College of Medicine Distinguished Professor in the division of cardiovascular medicine at The Ohio State University in Columbus, reported that between Mar. 27, 2018, and Nov. 30, 2021, subjects in the training phase made recordings on 83% of days. They were followed for up to 44 months. The test group made recordings on 81% of days between Feb. 1, 2020, and Apr. 30, 2023, and were followed for up to 31 months. Heart failure events were defined as hospitalization or outpatient intravenous diuretic treatment for worsening heart failure.
In the training phase, the app accurately predicted 44 of 58 heart failure events (76%) and 81% of first events (n = 35) on average 24 days before hospitalization or need for intravenous fluids. In the validation phase, the app was 71% accurate in detecting 10 of 14 heart failure events and 77% of first events (n = 10) on average 26 days in advance of events. In both periods, the app generated about 3 unnecessary alerts per patient year.
Dr. Abraham concluded, “This technology has the potential to improve patient outcomes, keeping patients well and out of the hospital, through the implementation of proactive, outpatient care in response to voice changes.”
The HearO® technology is being evaluated in an ongoing pivotal trial in the United State4s, Dr. Abraham said. The study is limited, he added, by the small number of patients and heart failure events, particularly in the test group.
“We continue to struggle with the burden of heart failure morbidity,” observed AHA press briefing moderator (and past AHA president) Clyde Yancy, MD, Magerstadt Professor at Northwestern University, Chicago. “So any tool that we can utilize and further refine that helps us address the need for hospitalization becomes very important. The idea that speech evaluation might give us sufficient early warning to forestall any admissions – and consider the cost savings attributable to that – is a very credible goal that we should continue to follow.” He pointed out that the technology enables assessments in the home environment for older patients who are less mobile.
In response to a press briefing question about the potential for physicians to be trained to hear early subtle voice changes on their own, Dr. Abraham stated, “I guess that is unknown, but the important difference is the system’s ability to take data in every day from patients and then process it automatically with AI.”
Joining in, Dr. Yancy said, “You know, this is interesting because even if you saw a patient once a month, which is an incredible frequency for any practice, there’s still 353 days that you haven’t seen the patient.” He noted that the AHA had just announced a multi-million dollar program to more deeply understand telemanagement. “So I think this is here to stay,” Dr. Yancy said.
Dr. Ouyang posed a further question. “Like with most AI recognition tools, we can now identify individuals at risk. How do we get from that step of identifying those at risk to improving their outcomes? This has been a critical question about heart failure, remote management, and remote monitoring, and I think it is a critical question for many of our AI tools.”
Dr. Abraham disclosed that he has received personal fees from Cordio Medical. Dr. Ouyang said that he had no disclosures relevant to this presentation.