A team of researchers at the University of Oxford has published results from CardioSense, an AI system that predicts heart attacks and major cardiac events up to 10 years before they occur. The study, published in The Lancet, analyzed 1.2 million patient records from the UK Biobank and achieved 92% accuracy in identifying individuals who would experience a cardiac event within the following decade.
How It Works
CardioSense does not require specialized cardiac imaging or expensive tests. It analyzes data that most patients already generate during routine checkups: standard blood panels, basic ECG readings, blood pressure measurements, and demographic information.
The model detects subtle patterns in blood work that human clinicians cannot reliably identify. Micro-variations in lipid ratios, C-reactive protein levels, and metabolic markers form complex signatures that correlate with long-term cardiovascular risk. Individually, these markers are within normal ranges — collectively, they tell a different story.
"A cardiologist looks at a cholesterol number and says it's fine," said Professor Sarah Chen, who led the research team. "Our model looks at the relationship between that number and 200 other values over time. The patterns it finds are invisible to the human eye but highly predictive."
The Study
The research team trained CardioSense on records from 800,000 patients in the UK Biobank, then validated it on a held-out set of 400,000 patients whose outcomes were already known. The model was asked to predict which patients would experience a heart attack, stroke, or cardiac arrest within 10 years.
Key results:
- 92% sensitivity — correctly identified 92% of patients who did experience a cardiac event
- 87% specificity — correctly ruled out 87% of patients who did not
- 10-year horizon — maintained accuracy across the full prediction window, with higher accuracy at shorter time frames (96% at 5 years)
The model outperformed the current clinical standard, the QRISK3 score, by 34 percentage points on the same patient population.
NHS Pilot Program
The UK's National Health Service has moved quickly. NHS England announced a pilot program that will deploy CardioSense in 30 hospitals across England starting in September 2026. The system will run alongside existing clinical workflows, flagging high-risk patients for early intervention.
The intervention pathway matters as much as the prediction. Patients identified as high-risk will be enrolled in preventive programs including medication adjustments, lifestyle coaching, and more frequent monitoring. The Oxford team estimates that early intervention in identified patients could prevent approximately 30,000 cardiac events per year in the UK alone.
Broader Implications
The paper is part of a growing body of evidence that AI can extract clinically significant predictions from routine medical data. Similar approaches are being explored for cancer detection, kidney disease progression, and neurological conditions.
Regulatory approval outside the UK is expected to take longer. The FDA has been cautious about long-horizon predictive AI in clinical settings, and the European Medicines Agency is still developing its framework for AI-based diagnostic tools. The Oxford team said it is in early conversations with both agencies.
For now, the results add to the case that the most impactful medical AI will not come from exotic new data sources but from better analysis of the data hospitals already collect.
