A smartphone application utilizing artificial intelligence (AI) successfully anticipated heart failure (HF) three weeks prior to hospitalization in a recent study by analyzing the patient’s voice patterns.
The groundbreaking research was unveiled on Monday, November 13, 2023, at the American Heart Association’s (AHA) Scientific Sessions in Philadelphia, Pennsylvania.
The integration of AI technology is increasingly prevalent in the healthcare sector, aiming to improve diagnostic accuracy. Notably, approximately 5% of medical facilities in the United States received incorrect treatments in 2021, as reported by the Southern Medical Association. To mitigate significant diagnostic errors, Babylon Health researchers introduced advanced AI symptom checkers three years ago. This recent innovation in AI applications promises to enhance heart failure therapy significantly, despite the established practice of using AI for medical diagnoses.
The AI system, known as the Cardio HearO program, analyzes various speech metrics such as pitch, volume, dynamics, and other characteristics to detect subtle changes in the patient’s voice patterns over time. These alterations in voice characteristics could signify an escalation in heart issues, a key indicator of HF progression. Impressively, the software accurately predicted over 75% of illnesses three weeks prior to their manifestation.
Lead researcher William T. Abraham, MD, FAHA, a distinguished professor of internal medicine at Ohio State University Wexner Medical Center, highlighted in the AHA press release that speech analysis represents a cutting-edge technology with the potential to revolutionize remote monitoring of HF patients. By offering early alerts of deteriorating cardiac conditions, this innovative approach enables proactive and personalized patient care, ultimately leading to improved health outcomes and reduced hospitalizations.
The study, conducted in Israel with 416 HF patients between March 2018 and April 2023, revealed promising results. The participants, predominantly male (75%) with an average age of 68, submitted five daily sentences in their native languages—Hebrew, Russian, Arabic, or English—via the application. While the efficacy testing phase involved only 153 participants, the AI algorithm was trained using data from 263 individuals.
During the training phase, the AI app accurately predicted a worsening of HF in 76% of cases approximately 24 weeks before hospitalization, issuing only three false alerts per person annually. Moreover, the app demonstrated a 71% accuracy rate in detecting HF progression within three weeks during the validation phase, with an average of three false alerts per person per year.
Despite the study’s limitations due to a relatively small participant pool, the results underscore the AI app’s potential in early HF prediction. Ongoing research in the United States is currently underway to further refine and validate the Cardio HearO system.
