Machine learning can predict long-term outcomes in transthyretin amyloid cardiomyopathy more accurately than conventional risk scores, according to a study published in JAMA Cardiology.
Researchers led by the University of Bern analyzed records from 850 people with the rare heart condition treated at specialist centers in Switzerland and Austria. Most patients were men around 79 years old. The model learned from data available at diagnosis, including symptoms, medications, blood tests, kidney function and echocardiography measurements, to predict death from any cause or hospitalization for heart failure.
Over a three-year follow-up, the AI tool was 3% to 19% more accurate than standard scores. Instead of a single training-test split, the team used internal-external cross-validation, repeatedly training on two geographic cohorts and validating on others.
The condition occurs when the protein transthyretin misfolds and deposits amyloid in the heart, stiffening the muscle and leading to heart failure. Left untreated, patients typically live anywhere from about 20 months to just over five years, with disease stage driving the range. An estimated 120,000 Americans live with the condition and 5,000 to 7,000 more are diagnosed each year, though it frequently goes undetected or is caught late.
The authors say clinicians still lack a widely validated contemporary risk model for guiding treatment decisions at diagnosis. The study is a step toward filling that gap, though prospective validation in broader populations would strengthen the case for clinical use.
