Researchers at the University of Surrey have identified three distinct health trajectories followed by people during the five years after a heart attack, using a form of machine learning designed to detect patterns in medical records over time. The findings suggest that recovery after acute myocardial infarction is not a single, predictable process. Instead, survivors may enter markedly different paths involving cardiovascular, metabolic, respiratory, musculoskeletal and kidney-related disease. By identifying these patterns at the moment of the heart attack, the researchers say clinicians may eventually be able to offer more personalised follow-up care before complications become established.
The study, published in the Journal of the American Medical Informatics Association, analysed the health records of 12,701 UK Biobank participants who had experienced an acute myocardial infarction. Rather than examining only whether patients suffered another heart event, the research team tracked the sequence and timing of diagnoses recorded after the initial heart attack. This allowed the investigators to study multimorbidity as a dynamic process: not simply how many conditions a person developed, but which conditions appeared, when they emerged and how they clustered together.
To find these patterns, the team applied data-driven temporal machine learning. Conventional clinical risk scores generally combine measurements such as age, blood pressure, cholesterol and medical history to estimate the likelihood of a future event. Temporal machine learning takes a different approach by analysing the order and timing of events. In this study, the method grouped patients whose post-heart-attack diagnoses followed similar trajectories, revealing three broad health pathways that might otherwise have remained hidden in standard statistical analyses.
The largest trajectory included approximately 63 per cent of the patients. People in this group tended to develop cardiometabolic conditions, including hypertension, type 2 diabetes and dyslipidaemia, alongside episodic heart and respiratory complications. These disorders are biologically interconnected: high blood pressure can place additional strain on the heart and blood vessels, insulin resistance can promote vascular damage, and abnormal blood lipid levels can accelerate atherosclerosis. The combination may create a cycle in which metabolic dysfunction increases cardiovascular stress while recurrent illness further reduces physical resilience.
A second trajectory involved around 23 per cent of the participants and was associated with deterioration affecting the lungs, musculoskeletal system and other organs. The researchers describe this as a group thought to be characterised by smoking-related risk. Smoking can damage the respiratory system directly, but its effects extend far beyond the lungs. Tobacco exposure contributes to chronic inflammation, impaired blood-vessel function, reduced oxygen delivery and accelerated tissue degeneration. These processes may help explain why patients in this trajectory experienced a wider decline involving respiratory and musculoskeletal health rather than a narrowly cardiac pattern.
The third trajectory, representing approximately 14 per cent of the cohort, was marked by structural heart disease, arrhythmias and kidney problems. Structural changes can interfere with the heart’s ability to pump efficiently, while arrhythmias disrupt its electrical rhythm. Kidney dysfunction is closely linked to cardiovascular disease through fluid regulation, blood pressure control and vascular injury. When heart and kidney problems develop together, each can intensify the other, producing a clinically complex condition that may require coordinated monitoring across several medical specialties.
Mortality differed sharply between the trajectories. The smoking-related group had a mortality rate of 44 per cent, more than three times the rate observed in the largest group dominated by cardiometabolic conditions. The contrast indicates that a patient’s future risk may depend not only on the presence of disease, but also on the type of biological pathway that follows the initial heart attack. Respiratory illness, older age and higher levels of socioeconomic deprivation were among the strongest predictors of membership in the highest-risk trajectory, according to the researchers.
Dr Anthony Onoja, the study’s lead author and a research fellow at the University of Surrey, said the analysis showed that a patient’s likely trajectory could be predicted at the time of the heart attack using pre-existing diagnoses and demographic information. The model was particularly effective at identifying people most likely to enter the high-risk pathway. Such a system could eventually be integrated into hospital records to flag patients who may need intensive respiratory assessment, smoking-cessation support, rehabilitation or closer monitoring for complications outside the heart. The researchers emphasise, however, that the approach remains at an early stage and would need further validation before routine clinical use.
The team also investigated whether the statistical groups reflected meaningful biological differences rather than merely being artefacts of medical-record patterns. Genetic analysis found that each trajectory was associated with distinct molecular pathways. The largest cardiometabolic group was linked to immune activation and tissue remodelling, processes involved in inflammation and the repair or restructuring of damaged organs. The trajectory involving structural heart disease, arrhythmias and kidney problems was associated with insulin signalling and lipid transport, biological systems that influence energy use and cardiovascular metabolism. The smoking-related trajectory showed links to chronic inflammation and degeneration, consistent with the long-term effects of tobacco exposure and systemic tissue injury.
The researchers compared the machine-learning trajectories with established clinical tools, including the SMART score, which is used to estimate the likelihood of future cardiovascular events. Professor Nophar Geifman, senior author of the study, said conventional risk assessments remained the strongest single predictor of mortality in the analysis. However, the trajectories added information that a single risk score cannot provide. A score may indicate that a patient is at high risk, while a trajectory can suggest whether that risk is more likely to involve metabolic disease, rhythm and kidney complications, or widespread respiratory and organ decline. That distinction could help clinicians move from general risk estimation toward earlier, targeted intervention.
The findings offer a new way to think about recovery after myocardial infarction, treating it as a long-term sequence of interconnected health events rather than an isolated cardiac emergency. The approach could support more personalised surveillance by helping healthcare teams identify which organs and disease processes are most likely to require attention. At the same time, the study does not establish that the machine-learning patterns cause the observed outcomes, and the trajectories will need to be tested in other populations and healthcare systems. If the results are confirmed, temporal models could become a powerful complement to existing risk scores, transforming routinely collected medical records into an early-warning system for the diverse paths heart attack survivors may follow.
Subject of Research: Health trajectories and multimorbidity after acute myocardial infarction
Article Title: Explainable temporal machine learning of multimorbidity trajectories after acute myocardial infarction: complementing clinical risk scores with mechanistic phenotypes
News Publication Date: 6-Aug-2026
Web References: https://doi.org/10.1093/jamia/ocag135
References: Journal of the American Medical Informatics Association, DOI: 10.1093/jamia/ocag135
Keywords
heart attack, myocardial infarction, cardiovascular disease, machine learning, artificial intelligence, multimorbidity, cardiometabolic disease, arrhythmia, kidney disease, respiratory disease, health trajectories, precision medicine, UK Biobank
Tags: cardiovascular disease clusteringdynamic disease progressionearly intervention in myocardial infarctionhealth trajectory analysisheart attack recoverymachine learning in cardiologymedical records pattern detectionmultimorbidity patterns after heart attackpersonalized post-heart attack carepredictive modeling for heart attack survivorstailored follow-up care in cardiac patientsUK Biobank heart health study
