OR WAIT null SECS
"Our work suggests that more lives could be saved with the application of this new machine-learning-based algorithm," Mihaela van der Schaar said
With a bit of competitive flair, a new announcement from UCLA notes that a predictive cardiology algorithm developed at the school “outperformed predictions from machine learning methods that have been developed by other research groups.”
The method is meant to project survival rates for patients with heart failure, allowing doctors to make personalized assessment of patients awaiting heart transplant. In a new Plos One study, it did so impressively.
Clinicians and engineers collaborated on the machine learning method, which they are calling Trees of Predictors. For this particular application, it took into account 53 different data points—33 regarding potential organ recipient, 14 about the donors, and 6 about the compatibility between the pair—ranging from demographic data to clinical metrics.
They tested Trees of Predictors across 30 years of data from the United Network for Organ Sharing. For various endpoints, the algorithm achieved a significantly higher area under the curve (AUC). For 3-month post-transplantation survival, it achieved an AUC of 0.660; the typical clinical risk scoring method achieves 0.587. It also outperformed many other models by 14% when predicting which patients would be alive 3 years later.
"Our work suggests that more lives could be saved with the application of this new machine-learning-based algorithm," one researcher, Mihaela van der Schaar, said. "It would be especially useful for determining which patients need heart transplants most urgently and which patients are good candidates for bridge therapies such as implanted mechanical-assist devices."
In addition to being a computing engineer at UCLA, she’s also a Turing Fellow at the Alan Turing Institute in London. One of her co-authors, cardiologist Martin Cadeiras, said that the algorithm also helped pinpoint those who would be good transplant candidates but might not be detected by other methods.
"This methodology better resembles the human thinking process by allowing multiple alternative solutions for the same problem but taking into consideration the variability of each individual,” Cadeiras said.
The methodology behind Trees of Predictors has applications beyond heart disease. It could be applied to other medical porblems, according to the study announcement, and the researchers have even used the framework on other types of data sets to show that it can detect handwriting and predict credit card fraud.