Algorithm predicts ICU survival rates more precisely than previous models

A machine learning algorithm that takes into account an intensive care patient's long-term disease history can predict their chances of survival more accurately than previous methods, according to a study published in the June 2019 issue of The Lancet Digital Health.

Researchers at the University of Copenhagen trained the algorithm on the medical data of more than 230,000 Danish ICU patients. When only long-term disease history was analyzed, the algorithm performed similarly to three widely used mortality risk prediction tools — the Multimorbidity Index, SAPS II and APACHE II — and became more precise with more extensive medical history data.

However, when a neural network of the patients' acute physiology measures was also factored in, the algorithm consistently outperformed all three previous methods, all of which were based on static, linear representations rather than utilizing machine learning.

"Our study shows that more realistic interpretations of the risk of death can be achieved with neural network models than by studying the additive effects of disease history and ICU measures," the researchers wrote. "Explainable machine learning models are key in clinical settings, and our results emphasize how to progress toward the transformation of advanced models into actionable, transparent and trustworthy clinical tools."

More articles about AI:
IBM exec says data-related challenges are biggest reason AI projects fall through
Harvard Business Review: What boards need to know about AI
ZeOmega AI uses social determinants of health to predict risk of opioid abuse

© Copyright ASC COMMUNICATIONS 2019. Interested in LINKING to or REPRINTING this content? View our policies by clicking here.

 

Top 40 Articles from the Past 6 Months