medRxiv

Artificial intelligence detected severe structural heart disease

The analysis increased the share of severe cases detected by 34.6 percentage points compared with the portable device’s standard rhythm reading.

Paper by Arya Aminorroaya, Sumukh Vasisht Shankar, Mariam Khan, +7 more authorsRohan Khera · Yale University · Yale New Haven Hospital
Figure 1. Study Design of the ACCESS-SHD Study
Open the figure at full size
Figure 1. Study Design of the ACCESS-SHD StudyStudy Design of the ACCESS-SHD StudyAminorroaya et al.

Our read

Biomarkers
Studied:
Preprint: cross-sectional study of 597 adults undergoing outpatient echocardiography (Independent replication)
What changed:
AI-ECG detected severe structural heart disease with 84.6% sensitivity against 50.0% with the device's native interpretation
The catch:
Testing patients already referred for echocardiography likely increased disease prevalence and overestimated positive predictive value
Stake:
An author founded Ensight-AI. Authors founded Evidence2Health.

Editor of record: Boris Djordjevic · AgingResearch.News

Follow @agingbrief

In a prospective observational study, 597 adults undergoing outpatient heart ultrasound at Yale New Haven Hospital recorded 30 seconds of their heart’s electrical activity with a portable device. Their median age was 61.7 years, and 51.4% were women. Ultrasound identified severe structural heart disease—serious problems involving heart pumping, valves or muscle—in 30 participants.

Researchers tested artificial intelligence analysis of the recordings against ultrasound findings. It detected 86.7% of severe cases and correctly classified 72.5% of people without severe disease. Only 14.4% of positive results corresponded to severe disease. Compared with the device’s standard rhythm reading, the analysis detected a larger share of severe cases but also flagged more people without severe disease.

Why it matters

Finding structural heart disease before symptoms develop is a screening question relevant to healthy aging. The authors suggest this approach could help guide screening on a large scale.

Caveats

Participants were already referred for heart ultrasound, so the results do not establish performance in general-population screening; people with implanted pacemakers or defibrillators were excluded. This preprint has not been peer-reviewed.

From the paper

Result

… AI-ECG detected severe SHD with substantially higher sensitivity than the device’s native interpretation (84.6% vs 50.0%; difference, 34.6 percentage points [95% CI, 13.0 to 56.0]), at the cost of lower specificity (73.0% vs 83.3%; difference, -10.3 percentage points [95% CI, -14.3 to -6.4]) …

Results, Incremental Value Over KardiaMobile Interpretation

Limitation

First, it was conducted in an echocardiography laboratory among patients already referred for TTE, which likely increased disease prevalence and may have overestimated PPV relative to community screening.

DISCUSSION

Quoted word for word from the full text on medRxiv.

Comments

Be the first to comment.