AI Tool Predicts Post-Heart Attack Trajectories to Personalize Follow-Up

Predicting which patients will deteriorate and where problems will arise remains difficult for cardiology and critical care teams. This uncertainty hinders timely, tailored follow-up after hospital discharge. Researchers have now developed an AI approach that identifies post–heart attack trajectories to guide more personalized care.

Researchers at the University of Surrey developed an AI tool that analyzes longitudinal health records to detect shared recovery patterns after acute myocardial infarction. The study, published on August 6, 2026 in the Journal of the American Medical Informatics Association, evaluated 12,701 UK Biobank participants who had sustained a heart attack. Investigators tracked the sequence and timing of new diagnoses after the index event and applied data‑driven machine learning to cluster patients following similar trajectories.

The analysis revealed three distinct five‑year courses. 63% percent developed cardiometabolic conditions, including hypertension, type 2 diabetes and dyslipidemia, alongside episodic cardiac and respiratory complications. 23% percent, thought to smoke, experienced progressive declines affecting the lungs, the musculoskeletal system and other organs, while 14% developed structural heart disease, arrhythmias and kidney disorders.

Outcomes differed substantially across groups. The smoking‑related trajectory had a 44% mortality rate, more than three times the rate observed in the largest cardiometabolic group. The team reported that trajectories could be predicted at the time of the heart attack using pre‑existing diagnoses and demographic information, enabling earlier risk‑aligned follow‑up.

To assess biological meaning, the researchers conducted genetic analyses. Each trajectory corresponded to distinct molecular pathways: immune activation and tissue remodeling in the largest cardiometabolic group; insulin signaling and lipid transport in the arrhythmia‑predominant group; and chronic inflammation and degeneration in the smoking‑related group. This confirmed biological distinctiveness across the patient clusters.

The trajectory information was also considered alongside commonly used secondary prevention risk assessments, such as the SMART score. Traditional risk scores remained the strongest single predictors of mortality in this study, while the trajectory patterns offered added detail beyond a standalone score. Together, these findings suggest a complementary role for trajectory modeling, highlighting where and why intervention may be needed.

"Clinicians typically use risk assessments, such as the SMART score, to help them understand how likely a patient is to have another heart event. We found that these tools are still the strongest single predictor of mortality in our study, but the trajectories added detail that a standalone score cannot provide. The patterns we have identified show that we can capture not just a patient's risk but, crucially, why and where intervention could be needed," said Professor Nophar Geifman, senior author of the study from the University of Surrey.

Related Links
University of Surrey