The most alarming number in this story isn't about the heart at all — it's 44%. That's the death rate among one group of heart attack survivors, a rate more than three times higher than any other. And until now, doctors had no easy way to spot those patients early enough to help them.
Researchers at the University of Surrey in Guildford have built an artificial intelligence tool that sorts heart attack survivors into three distinct recovery paths. The idea is simple and powerful: if doctors know which path a patient is on, they can tailor care sooner. The study, published in the Journal of the American Medical Informatics Association, used the health records of 12,701 participants who had survived a heart attack, tracking the sequence and timing of every new diagnosis that came afterward.
Using machine learning, a type of AI that finds patterns in large amounts of data, the team grouped patients who followed similar health patterns. The largest group, 63% of survivors, developed cardiometabolic conditions — high blood pressure, type 2 diabetes, and dyslipidemia, a condition where fats build up in the blood — along with occasional heart and breathing complications. A second group, about 23% of patients who were thought to smoke, suffered declines in their lungs, muscles, bones, and other organs. This group had that striking 44% mortality rate. The smallest group, roughly 14%, developed structural heart disease, irregular heartbeats, and kidney problems.
The AI tool proved especially good at finding the highest-risk patients. Lead author Dr. Anthony Onoja explained that the system could predict a patient's trajectory at the moment of the heart attack itself, using only their existing diagnoses and basic demographic information. The strongest warning signs were respiratory conditions, older age, and higher deprivation scores — a measure of how disadvantaged a community is.
What makes this research stand out is that it's not just pattern-spotting. Genetic analysis confirmed each group maps to its own biological machinery: immune activation and tissue repair in the largest group, insulin and fat transport signals in the arrhythmia group, and chronic inflammation and degeneration in the smoking-related group. In plain terms, these three paths are biologically real.
The researchers are careful not to overpromise. Senior author Professor Nophar Geifman noted that traditional risk scores, like the SMART score, remain the strongest single predictor of another heart event. But the trajectories add something a single number can't: they show not just how much risk a patient faces, but why, and exactly where care could be targeted.
The team is honest that this is early days. But they believe the tool could one day help hospitals identify high-risk survivors early and design care that matches each person's specific path. For the millions of people who leave a hospital after a heart attack, that could mean care that finally fits them — not just a vague promise of recovery, but a map of the road ahead.