# Smartwatch AI Struggles to Reliably Detect Heart Dysfunction Korean study reveals inconsistencies in smartwatch-based detection of left ventricular systolic dysfunction. - **Source:** [MedPage Today](https://www.medpagetoday.com) - **Published:** 2026-11-09 - **Topic:** [health](https://brifing.si/c/health) - **Original article:** - **On Brifing:** https://brifing.si/s/ca260368-f9ec-45a1-8b9f-ab172db758cd ## Diagnostic Accuracy Challenges Researchers evaluated an AI-ECG model designed to identify left ventricular systolic dysfunction using Apple and Samsung smartwatches. While the model demonstrated high area under the curve scores of 0.907 and 0.895, consistent real-world performance remained elusive. The study highlights significant hurdles before these devices can reliably screen for heart failure in clinical settings. ## Device Discordance Issues At a shared model cutoff, 9% of the 2,211 participants showed discordant results between the two smartwatch brands. Specifically, 158 individuals tested positive for dysfunction on the Apple Watch while testing negative on the Galaxy Watch. These systemic differences suggest that device-specific tuning is required to ensure uniform diagnostic accuracy across different hardware. ## Intra-Patient Variability In a smaller cohort performing at-home ECG monitoring, 49.3% of patients recorded both positive and negative results within the same week using the same device. This high rate of inconsistency indicates that a single negative recording is insufficient for reassurance. Mixed results necessitate further clinical evaluation to confirm or rule out potential heart conditions. ## Limitations of Averaging Researchers attempted to improve accuracy by averaging ECG readings over a one-week period. This strategy successfully reduced false positives but simultaneously resulted in more missed cases of left ventricular systolic dysfunction. Consequently, there is no simple mathematical fix for the current limitations of smartwatch-based screening, requiring more sophisticated algorithmic adjustments and fine-tuning. ## Clinical Implementation Hurdles Experts emphasize that while effective heart failure treatments exist, early identification remains a major challenge. The primary concern for health systems is managing the burden of false positives generated by wearable technology. Balancing the urgent need for early detection against the risk of overwhelming medical resources is a critical next step for clinicians. ## Study Methodology Details The AI model, AiTiALVSD-1L, was pre-trained on nearly 3 million unlabeled ECGs and fine-tuned on 360,000 clinical lead I ECGs. The study defined dysfunction as an ejection fraction of 40% or less. Participants included 2,211 patients undergoing echocardiography, with 6.6% having known dysfunction and 31% having a history of heart failure. ## Future Research Requirements Current findings are limited by a single-center design and a lack of data on asymptomatic populations without heart failure. Future studies must assess performance across diverse demographics, including variations in age, race, and ethnicity. Understanding how these results translate into actual clinical outcomes is essential before widespread adoption of smartwatch screening tools. ## Need for Further Validation Smartwatch AI shows potential but requires significant evidence before real-world application is viable. Clinicians advise that mixed results should always prompt further professional evaluation rather than relying solely on wearable data. Developing specific device cutoffs and robust validation strategies remains the priority for integrating these technologies into standard cardiovascular care pathways. ## Sources covering this story - https://www.medpagetoday.com/meetingcoverage/aha/123800 — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._