Self-taught computer algorithms could save thousands of lives and improve our understanding of how our bodies work.
Heart disease, including heart attacks and strokes, accounts for more than half a million deaths per year in the U.S. It’s the leading cause of death for both men and women, and most instances of heart disease are preventable. Doctors are pretty good at identifying risk factors and warning people who are susceptible to heart attacks, but it turns out a computer might be even better.
Researchers at the University of Nottingham have developed a set of computer programs that could predict heart attacks better than doctors. The algorithms were trained on real patient records and developed criteria that outperformed the current guidelines set by the American Heart Association.
The AHA criteria identifies several risk factors like age, high blood pressure, and obesity that increase a person’s chance of suffering a heart attack. Doctors use these criteria when predicting who is most at risk of heart disease, and this method has worked pretty well. But there’s always room for improvement.
The Nottingham researchers developed four different computer programs, each using a different algorithm. These programs studied a large data set of patient records and attempted to develop their own set of criteria to predict heart attacks. Each of these programs was able to outperform doctors, with the best of the four making an additional 7.6 percent correct diagnoses with 1.6 percent fewer false positives.
In the United States alone, that extra accuracy could account for thousands of lives saved every year. But just as importantly, these computer programs could help doctors better understand the complex causes of heart disease.
This article was originally written for and posted by Popular Mechanics USA.