Machine learning identifies three health trajectories after heart attack

At the moment of a heart attack, a patient’s future may already be splitting into different paths. Researchers at the University of Surrey used machine learning—an artificial intelligence method that finds patterns in data—to identify three distinct five-year health trajectories among 12,701 UK Biobank participants who had survived a heart attack.

The team tracked when new diagnoses appeared and grouped patients whose health records followed similar sequences. 63% developed cardiometabolic conditions such as hypertension, type 2 diabetes and dyslipidemia, together with episodic heart and respiratory complications. A second group, 23% of participants thought to smoke, experienced declines in the lungs, musculoskeletal system and other organs. The final 14% developed structural heart diseases, arrhythmias and kidney problems.

The difference between those paths was not merely statistical. The smoking-related group recorded a 44% mortality rate, more than three times the rate of the largest group. Genetic analysis also connected each trajectory to distinct molecular pathways: immune activation and tissue remodeling in the largest group, insulin signaling and lipid transport in the arrhythmia group, and chronic inflammation and degeneration in the smoking-related group.

The researchers say the model could predict a patient’s likely trajectory at the time of the heart attack, using pre-existing diagnoses and demographic data. Anthony Onoja, the study’s lead author, said respiratory conditions, older age and higher deprivation scores were key predictors of the highest-risk group. Existing risk assessments, including the SMART score, remained the strongest single predictor of mortality in the study; the trajectories added information about why and where intervention might be needed.

So what changes in practice? If the approach is validated and adopted, hospitals could identify higher-risk survivors earlier and tailor follow-up care to the problems most likely to emerge, rather than relying on a single estimate of another heart event. For now, the evidence comes from health-record analysis, and the researchers describe the work as an early step.