A new machine learning system is better at predicting the likelihood of patients with cardiovascular problems dying within ten years than healthcare professionals’ methods, according to a study presented at the EuroEcho 2021, a scientific meeting of the European Society Cardiology.
Unlike traditional methods based solely on clinical data, the new machine learning system also includes results from imaging scans on the heart, measured by stress cardiovascular magnetic resonance (CMR). During this exam, patients receive a drug that mimics the effect of exercise on the heart and then undergo imaging using a magnetic resonance imaging scanner.
Assessing the risk of death is commonly done in these patients. Usually, doctors use a limited amount of clinical information, including age, sex, smoking, blood pressure, and cholesterol levels. Patients with at least two risk factors, such as hypertension, dyslipidaemia, diabetes, and smoking, are considered high risk. This allows healthcare professionals to tailor care to prevent heart attacks or strokes. The problem is that this process is not always accurate, and many patients don’t receive the care they need.
This study aimed to find a better way to detect high-risk patients. A team of cardiologists from the Hospital Lariboisiere in Paris, France, wanted to see if combining machine learning using stress CMR data with clinical data would improve predictions regarding 10-year all-cause mortality in patients with coronary artery disease.
The study included 31,752 patients who underwent stress CMR between 2008 and 2018 in Paris. The patients were referred for this exam due to chest pain, shortness of breath on exertion, or a high risk of cardiovascular disease. Participants were, on average, 64 years old, and about 2/3 were men. Doctors collected information on 23 clinical and 11 CMR parameters and followed patients for six years, on average. During that period, 2,679 (8.4%) patients died.
Machine learning was done in two steps. First, the system selected which of the clinical and CMR parameters could actually be used to predict death, which were then used in the second step to build an algorithm allocating different importance to each parameter to create the best prediction. Participants received a score between 0 (low risk) and 10 (high risk), indicating how likely they could die within ten years.
The machine learning score regarding which patients would be alive or dead after ten years with 76% accurate. This is significantly higher than any other established methods. “This means that in approximately three out of four patients, the score made the correct prediction,” said study author Dr. Theio Pezel. “Some information we collect from patients may not seem relevant for risk stratification. But machine learning can analyse a large number of variables simultaneously and may find associations we did not know existed, thereby improving risk prediction.”
“This is the first study to show that machine learning with clinical parameters plus stress CMR can very accurately predict the risk of death,” continued Dr. Pezel. “The findings indicate that patients with chest pain, dyspnoea, or risk factors for cardiovascular disease should undergo a stress CMR exam and have their score calculated. This would enable us to provide more intense follow-up and advice on exercise, diet, and so on to those in greatest need.”