Researchers have developed a new machine learning model capable of anticipating when infective endocarditis (IE) may occur after transcatheter aortic valve replacement (TAVR), sharing their findings in the American Journal of Cardiology.
IE is a rare TAVR complication, the authors noted, but when it does occur, it has been associated with “significantly increased complications and mortality.”
For their analysis, the researchers explored data from nearly 78,000 TAVR hospitalizations. Just 404 of those patients—less than 1%—developed IE as a complication. All patients were treated in the United States from 2014 to 2017. While 70% of that data was used to train the team’s AI model, the final 30% was used to validate the model’s performance.
The study’s primary endpoint was IE within 180 days of the TAVR procedure. Overall, the advanced AI model achieved an area under the curve (AUC) of 0.75. A standard logistic regression algorithm, meanwhile, achieve an AOC of just 0.60.
“As a subset of artificial intelligence, machine learning provides an innovative approach to data analysis beyond what is provided by conventional statistics,” wrote lead author Agam Bansal, MD, an internal medicine specialist with the Cleveland Clinic’s Heart and Vascular Institute, and colleagues. “In our study, machine learning algorithm outperformed conventional logistic regression for estimating infective endocarditis. To the best of our knowledge, this is the first risk model developed for predicting IE post-TAVI.”
History of previous IE, advanced age, a prolonged hospitalization after TAVR, heart failure and atrial fibrillation were all associated with a greater risk of post-TAVR IE. Also, IE was more common among men than women.
The team did caution that its work has certain limitations. The event rate was very low, for instance, and “machine learning algorithms are often criticized for overfitting.” However, even with these limitations in mind, the group sees its findings as a helpful tool that could “potentially help clinicians and patients in prognostication.”
Click here to read the full study.