Artificial intelligence trained on more than 440,000 electrocardiograms has identified a previously overlooked electrical pattern in routine heart scans that predicts sudden cardiac death more effectively than conventional medical tests, Science Daily reported.
More than 300,000 people in the United States die from sudden cardiac arrest each year, an emergency where the heart’s electrical system suddenly fails and stops pumping blood. Implanted defibrillators prevent some of these deaths, but doctors have long struggled to determine which patients need them most, according to Science Daily.
“Medical decisions are really hard, and I think that’s why AI is so exciting for me,” as stated to Science Daily by Ziad Obermeyer, lead author of the study and an associate professor at UC Berkeley’s School of Public Health. “We can not only make better decisions, but also start to understand what’s actually going on with these patients before their heart stops.”
AI Identifies Heart Risks Using Existing EKG Patterns
Researchers at UC Berkeley trained an AI system by comparing electrical patterns from EKGs in Sweden with information from death certificates. Science Daily reported that the algorithm was trained to pick up on subtle distinctions separating healthy subjects, individuals with known cardiac vulnerabilities, and those who ultimately succumbed to sudden cardiac arrest.
The team tested the technology using thousands of additional patient records from the United States and Taiwan. Traditional screening methods largely focus on how much blood the heart pumps with each contraction to assess if someone needs an implanted defibrillator, identifying a patient group with an annual sudden cardiac death rate of 4.6%, according to Science Daily.
By pinpointing a cohort experiencing a 7% annual mortality rate, the artificial intelligence architecture proved better equipped at singling out vulnerable individuals at severe risk. The algorithm also recognized a broader population at elevated risk, including thousands who would not normally raise concerns under existing medical guidelines.
In Plain English: The Clinical Takeaway
- Routine EKG Compatibility: The AI system analyzes ordinary EKG images already collected routinely in hospitals worldwide, meaning it does not require an entirely new type of medical test.
- Higher Predictive Accuracy: Traditional screening identifies high-risk groups with a 4.6% annual death rate, whereas the AI algorithm pinpointed a population with a 7% annual death rate.
- Catching Missed Patients: The technology flags thousands of individuals who appear low-risk under conventional pumping-capacity measurements but still face dangerous electrical instability.
“One thing that makes the problem very tragic, but also very well suited for AI, is that we have the cure for this problem,” Obermeyer told Science Daily, adding that patients would get a defibrillator if they knew they were at risk.
“Unfortunately, data like the ones we used for this study are incredibly hard to access,” Obermeyer told Science Daily, noting that assembling the datasets took approximately 10 years and involved contributions from Dandelion Health and Nightingale Open Science.
| Screening Method | Primary Diagnostic Metric | Identified Annual Death Rate |
|---|---|---|
| Conventional Testing | Ejection fraction / Blood volume pumped per contraction | 4.6% |
| AI EKG Algorithm | Complex electrical wave and spike patterns from standard EKGs | 7.0% |
Contraindications & When to Consult a Doctor
Sudden cardiac arrest results from a failure of electrical signals that coordinate the heartbeat, distinguishing it from a heart attack caused by blocked blood flow. Roughly 90% of people experiencing sudden cardiac arrest outside a hospital die within minutes.
Patients who receive a high-risk EKG assessment under the proposed AI approach would be offered a patch that continuously records electrical activity.
As noted by Science Daily, Obermeyer has additionally established an online portal where individuals curious about their personal risk can input basic demographic details and an email address so that investigators can reach out regarding EKG evaluations once the methodology sees wider rollout.
References
- Nature (Research findings published regarding artificial intelligence detection of sudden cardiac death risks)
- UC Berkeley School of Public Health (Clinical study data and international dataset evaluations)
- UCSF-UC Berkeley Computational Precision Health program (Data infrastructure and methodology)
