Researchers from the University of Nottingham in the U.K. have developed four computer-learning algorithms that significantly outperform American College of Cardiology/American Heart Association guidelines for predicting heart attacks or strokes.
The neural network algorithms achieved 74.5% to 76.4% accuracy, beating existing guidelines by 7.6% while raising 1.6% fewer false alarms.
The researchers trained the algorithms on data from 378,256 patients in the U.K., using about 295,000 records to generate the internal predictive models. The researchers then used the remaining records to test and refine the algorithms.
Out of the 83,000-patient set of test records, the new system could have saved 355 extra lives, according to the researchers.
In addition, the artificial intelligence systems identified a number of risk factors and predictors not covered in existing guidelines, such as severe mental illness and the consumption of oral corticosteroids.
View Full Article
Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA
No entries found