Research Highlights:
- An artificial intelligence (AI) model analyzed ECG recordings from two different brands of smartwatches for more than 2,200 adults in South Korea. The study found that the AI model was able to identify people with and without reduced heart pumping function, or left ventricular systolic dysfunction (LVSD). However, for 9% of participants, the results differed between the two smartwatch brands: positive (suggesting possible reduced heart pumping function) on one and negative on the other.
- Among another set of 146 adults who recorded ECGs at home on their own smartwatches, requiring positive results on at least 2 of 3 consecutive days, compared to acting on results on a single positive day, halved false-positive results (from 18 to 9 among 70 people without LVSD) in the 108 who recorded for at least 5 days, but missed two more people with LVSD.
- The findings suggest that smartwatch ECGs analyzed with AI may help identify people who need additional heart screening such as an echocardiogram. More research is needed before smartwatch ECG readings can replace traditional screening exams.
- Note: The study featured in this news release is a research abstract. Abstracts presented at the American Heart Association's scientific meetings are not peer-reviewed, and the findings are considered preliminary until published as a full manuscript in a peer-reviewed scientific journal.
CHICAGO, Nov. 9, 2026 — An artificial intelligence (AI) software tasked with evaluating heart signals (ECG data) measured by smartwatches may help identify a hidden and dangerous heart condition known as left ventricular systolic dysfunction (LVSD), although results varied between smartwatch brands and across ECG recordings, according to new, preliminary, independent late-breaking science presented today at the American Heart Association's Scientific Sessions 2026. The meeting on Nov. 6-9 in Chicago is a premier global exchange of the latest scientific advancements, research, and evidence-based clinical practice updates in cardiovascular science.
LVSD occurs when the heart's main pumping chamber, the left ventricle, does not contract effectively. In this study, LVSD was defined as a left ventricular ejection fraction of 40% or less, meaning that the chamber pumps out no more than 40% of the blood it contains with each heartbeat. LVSD can be present without symptoms, and without treatment it can lead to symptomatic heart failure. The researchers examined whether AI analysis of ECG signals measured by smartwatches could help identify people who may need traditional screening exams such as an echocardiogram, or ultrasound scan of the heart. This study evaluated a novel AI model that analyzed smartwatch ECG recordings for signs of reduced heart pumping function.
"Some people have reduced heart pumping function without noticeable symptoms," said study author Hak Seung Lee, M.D., FAHA, chief medical officer of Medical AI Co., Ltd., which developed the AI model, and a cardiologist at the Digital Healthcare Institute, Sejong Medical Research Institute in Incheon, South Korea. "Our study explored whether AI analysis of smartwatch ECGs could help identify people who may benefit from additional screening for LVSD or other heart conditions."
What are the study's key findings?
- Among 2,211 adults scheduled for an echocardiogram at an outpatient cardiology clinic, ECGs were recorded on both an Apple Watch and a Samsung Galaxy Watch during the same visit. The AI model distinguished people with and without LVSD about equally well on both watches. However, when the same AI-score cutoff (set in advance using separate Galaxy Watch data) was applied to both groups, 200 participants (9%) were classified as positive on one watch and negative on the other.
- A second group of study participants included 146 adults who recorded ECGs at home via smartwatches. Among 108 of the 146 participants who recorded ECGs on at least five days in the first week, when positive results in the AI analysis were required for at least 2 of 3 consecutive recording days, false-positive LVSD classifications were reduced from 18 to 9 participants among the 70 participants who did not have LVSD. This method of requiring at least 2 of 3 consecutive recording days of positive data allowed the AI model to identify 31 of 38 participants with LVSD, compared with 33 participants when ECG data from single-positive days were analyzed. Thus, false positives were reduced by 50%; however, two additional cases of LVSD were missed compared with the single-positive-day rule. This was one of several repeat-testing rules examined.
- Among all 146 home-recording participants, 81% of those with consistently positive results had LVSD confirmed by echocardiography, versus about 6% of those with consistently negative results. Nearly half (49%) had a mix of positive and negative results, and 40% of this group had confirmed LVSD.
- Because LVSD was more common in this study population than in the general population — participants were patients referred for an echocardiogram, many of whom already had known heart failure — the researchers adjusted the analysis to account for this difference and estimated results at lower prevalence.
"A single smartwatch ECG is a snapshot, and repeated recordings can give different results," Lee said. "Requiring repeated positive results reduced false positives in our study, yet it also missed some additional cases that should have been positive. Larger studies are needed to determine how best to use these results to guide further evaluation."
What are the study's details, background, design and limitations?
- Two groups of adults were enrolled in the study at Incheon Sejong Hospital in South Korea. The first group was enrolled between July 2023 and May 2024, and the second group was enrolled between July 2024 and December 2025.
- Group 1 included 2,211 adults who were scheduled for an echocardiogram at an outpatient cardiology clinic. Each participant had ECGs recorded using both an Apple Watch and a Samsung Galaxy Watch, back-to-back during the same clinic visit, under staff supervision.
- Group 2 included 146 outpatients, enrolled separately, who recorded ECGs at home on their own smartwatches with staff reminders to keep recording (94% of participants in this group used a Galaxy Watch). Each participant in group 2 recorded a median of 2.4 ECGs per day, and recordings from the first week were analyzed.
- In both groups, ECGs were recorded on the smartwatch as a 30-second, single-lead recording of the heart's electrical signals. Group 1 participants were recorded on both an Apple Watch and a Samsung Galaxy Watch; Group 2 participants used their own smartwatch at home (94% Galaxy Watch). The AI model analyzed these recordings for signs of weakened heart function.
- A clinical diagnosis of LVSD was then confirmed by transthoracic echocardiography (TTE), an ultrasound of the heart to show how blood is flowing through the heart and heart valves. In both groups, ECGs were recorded within 14 days before or after the echocardiogram.
- The study has several limitations. It was conducted at one hospital in South Korea and included patients referred for echocardiography, so the results may not apply to the general population. The comparison of repeat-testing rules in Group 2 included only 108 participants, and these findings are exploratory. Poor-quality ECG recordings (less than 3%) were excluded from the analysis, and Group 2 participants received reminders to keep recording at home.
- The study did not determine whether using these AI results leads to earlier treatment, fewer hospitalizations or improved survival. In addition, 12-lead ECGs were not available for comparison in the group that had smartwatch ECG measures taken in the clinic.
Co-authors, disclosures and funding sources are listed in the abstract.
Statements and conclusions of studies that are presented at the American Heart Association's scientific meetings are solely those of the study authors and do not necessarily reflect the Association's policy or position. The Association makes no representation or guarantee as to their accuracy or reliability. Abstracts presented at the Association's scientific meetings are not peer-reviewed, rather, they are curated by independent review panels and are considered based on the potential to add to the diversity of scientific issues and views discussed at the meeting. The findings are considered preliminary until published as a full manuscript in a peer-reviewed scientific journal.
