Machine Learning Boasts 90% Accuracy Rate for Predicting Heart Attack, Death

A recent study from the European Cardiology Society has found that machine learning may be a more efficient method for predicting heart attacks and death in humans.

The study, which was presented at the 2019 International Conference on Nuclear Cardiology and Cardiac CT, found that the machine learning model is more than 90% accurate in analyzing 85 variables to determine a person's risk of suffering a heart attack or death in the future.

"Humans have a very hard time thinking further than three dimensions (a cube) or four dimensions (a cube through time). The moment we jump into the fifth dimension we're lost. Our study shows that very high dimensional patterns are more useful than single dimensional patterns to predict outcomes in individuals and for that we need machine learning," said Luis Eduardo Juarez-Orozco, MD, PhD, lead author and research fellow at the Turku PET Centre.

Investigators enrolled 950 patients from the Turku PET Center in Finland with chest pain who had undergone usual protocol to look for coronary artery disease. Investigators obtained 58 pieces of data from coronary computed tomography angiography scans. Investigators analyzed data from positron emission tomography (PET) scans, which produced 17 variables on blood flow. They also included 10 clinical variables, including sex, age, smoking and diabetes, from medical records.

The 85 variables were entered into the machine learning algorithm called LogitBoost. LogitBoost is designed to with a machine learning algorithm to adapt to its individual user. Study authors noted that LogitBoost has the ability to analyze all 85 variables repeatedly until it develops the most efficient way to predict who had a heart attack or died. LogitBoost was able to identify patterns correlating the variables to death and heart attack with more than 90% accuracy.

"The algorithm progressively learns from the data and after numerous rounds of analyses, it figures out the high dimensional patterns that should be used to efficiently identify patients who have the event. The result is a score of individual risk," said Juarez-Orozco.

During the follow-up period, which lasted an average of 6 years, investigators observed there were 24 heart attacks, 49 deaths from any cause, and 109 patients underwent early revascularization.

Authors noted the predicative performance using the 10 clinical variables by themselves was modest, with an area under the curve of .65. When investigators added PET data the AUC increased to 0.69. When combing CCTA data to clinical and PET data, there was an AUC of .82.

In their conclusion, the authors wrote that machine learning for the analysis of clinical and hybrid sequential cardiac PET/CCTA data can improve the identification of symptomatic with CAD who will develop MI or death.

This study, titled “Refining the long-term prognostic value of hybrid PET/CT through machine learning,” was presented at the 2019 meeting of the International Conference on Nuclear Cardiology and Cardiac CT in Portugal.