TL;DR
Get health and wellness essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
A Korean study presented at the American Heart Association meeting found that an AI model’s smartwatch ECG results for reduced left ventricular function could differ by device and vary across repeated readings. Researchers said the findings do not support relying on a single negative home recording, and called for more research before real-world use.
A Korean study presented at the American Heart Association annual meeting found that an AI model screening smartwatch ECGs for reduced heart-pumping function produced inconsistent results between Apple and Samsung watches and across repeated home readings. The findings temper expectations that smartwatch recordings are ready to serve as a dependable way to identify left ventricular systolic dysfunction (LVSD) outside clinical settings.
The device comparison involved 2,211 people at Incheon Sejong Hospital who had a clinical reason to undergo echocardiography. Researchers paired participants with Apple Watches and Samsung Galaxy Watches and applied a shared AI-model cutoff. LVSD was defined as an ejection fraction of 40% or less, measured on an ECG within 14 days of echocardiography. The cohort included people with known disease: 6.6% had known LVSD and 31% had known heart failure.
The model’s area under the curve was 0.907 for Apple Watch and 0.895 for Galaxy Watch. Reported sensitivities were 69.2% and 53.4%, respectively, while specificities were 88.9% and 93.4%. Despite those discrimination measures, 9% of participants had discordant results at the shared cutoff: 158 tested positive only on Apple Watch and 42 only on Galaxy Watch. Presenter Hak Seung Lee said Apple Watch scores tended to be higher.
In a separate group of 146 people taking ECGs at home with one smartwatch, 49.3% had both positive and negative recordings within the same week at the same cutoff. Galaxy Watches were used by most of this group. Attempts to reduce false positives, including averaging readings over a week, had trade-offs: a stricter rule lowered false positives but also missed more cases of LVSD.
Why Inconsistent Readings Matter
The findings matter because a smartwatch screening result could affect whether someone seeks clinical assessment, while inconsistent readings may prompt unnecessary follow-up or provide misleading reassurance. Lee said a single negative home recording is not reassuring on its own, and that mixed results should lead to further evaluation. The study did not establish that the devices improve diagnosis or patient outcomes.
Jessica Golbus, a University of Michigan cardiologist who spoke at the AHA press conference, said health systems will need to weigh the need to find people with symptomatic or asymptomatic LVSD against the burden of false positives. She also emphasized that screening only helps if people with detected LVSD can be connected to care and effective treatment. Whether smartwatch findings lead to earlier treatment or better outcomes remains untested in this study.
Products Worth Considering
Fitbit Sense 2 Health and Fitness Smartwatch, Shadow Grey/Graphite
Garmin Forerunner 165 GPS Running Smartwatch, 43 mm, Whitestone
Garmin Forerunner 935 GPS Running and Multisport Watch, Black
As an affiliate, we earn on qualifying purchases.
How the Watch Model Was Tested
The AI-ECG model, called AiTiALVSD-1L, uses a single-lead recording. Lee said it was pre-trained on nearly 3 million ECGs without labels and then fine-tuned for LVSD using about 360,000 clinical lead I ECGs. Researchers selected a cutoff of 23.5 in a separate Galaxy Watch tuning set at 98% specificity, then applied that same cutoff to Apple Watch recordings.
Interest in the approach follows a 2022 study that reported good accuracy for an AI model using smartwatch single-lead ECGs to detect LVSD. The current research tested two prospective cohorts at one South Korean hospital. The home-recording group was different from the 2,211-person device-comparison cohort and had a higher prevalence of LVSD: 33.6% had the condition and 55.5% had known heart failure.
“A single negative home recording is not reassuring on its own, and mixed results should prompt further evaluation.”
— Hak Seung Lee, MD, study presenter
Who Was Not Studied
The study was conducted at a single hospital, and the researchers did not test screening performance in asymptomatic people without heart failure. Its results therefore do not establish how well the approach would work in the general population or in different health systems. The source report also does not provide enough detail to determine how the model performs across age, race, ethnicity, or ranges of ejection fraction.
It remains unclear which device-specific thresholds, if any, could produce reliable results across watches and repeated recordings. The study did not assess whether using the readings changes clinical decisions, leads to earlier treatment, or improves health outcomes. Golbus said performance should be examined in different patient groups, and that systems for follow-up would need to be in place.
Validation Before Routine Screening
Researchers will need to test the model in larger, more diverse populations and determine how its performance changes by device, patient characteristics, and recording conditions. Further work must also establish how to balance false positives against missed LVSD cases and whether a device-specific cutoff or repeated-reading strategy is useful.
Lee called for more evidence before real-world use. Golbus said an important next step is evaluating how smartwatch findings translate into clinical outcomes and whether people identified by screening can be linked to appropriate care. The source report does not specify a timeline for additional studies or any planned change to clinical practice.
Key Questions
Did the study show that smartwatches can reliably detect reduced heart function?
No. The model showed discrimination in the study, but results differed between watch brands and were inconsistent across repeated home readings. The researchers said more evidence is needed before real-world use.
What condition was the AI model screening for?
It screened for left ventricular systolic dysfunction, defined in this study as an ejection fraction of 40% or less. Echocardiography was used as the clinical reference, with the study’s ECG measurement taken within 14 days.
How often did home readings vary?
In a separate group of 146 people recording ECGs at home, 49.3% had both positive and negative readings within the same week at the same cutoff. Most in that group used a Galaxy Watch.
Does one negative smartwatch reading rule out LVSD?
No such conclusion can be drawn from this study. Presenter Hak Seung Lee said a single negative home recording is not reassuring on its own. The study does not establish a smartwatch reading as a substitute for clinical assessment.
What needs to happen before this is used routinely?
Researchers need to validate performance in more settings and patient groups, refine device-specific thresholds, and study whether screening improves clinical outcomes. Health systems would also need processes to follow up on positive or mixed results.
Source: medpagetoday.com
