AI model predicts risk of post-operative AFib
Researchers have built a new artificial intelligence (AI) model that uses electrocardiogram (ECG) data to identify patients who are more likely to develop atrial fibrillation (AFib) after surgery, sharing their findings in Circulation: Arrhythmia and Electrophysiology.[1]
Post-operative atrial fibrillation (POAF) was once viewed as “benign and transient,” the group explained, but more recent research suggests it can increase a patient’s risk of longer hospitalizations, spontaneous AFib and even stroke.
“It is imperative to establish practical screening for high-risk patients with POAF,” wrote first author Takeshi Tohyama, MD, PhD, a specialist with Kyushu University Hospital in Japan, and colleagues.
Tohyama et al. focused on data from nearly 43,980 pre-operative ECGs taken from 27,564 adult patients undergoing surgery at a high-volume hospital from 2015 to 2020. All patients included in the study were free of AFib and showed no signs of AFib in their pre-operative ECGs. After their surgery, each patient underwent an additional ECG within seven days.
To develop its deep learning-based AI model, the researchers started with more than 137,000 ECGs. They then turned to the previously mentioned 43,980 pre-operative ECGs, using them to train, tune and internally validate the algorithm. The POAF rate of the internal validation dataset was 3.6%, and the AI model achieved a sensitivity of 79.9%, specificity of 73.5%, positive predictive value of 10.2% and negative predictive value of 99%.
“Our results showed that the deep learning model could effectively predict POAF with high accuracy,” the authors wrote. “The low incidence of POAF might be responsible for the high negative predictive value, but low positive predictive value. However, the high negative predictive value is the most important requirement for clinical applications because prediction with this model is used as an initial screening to identify high-risk patients.”
One key limitation with this study was that it was performed with retrospective data; a follow-up analysis using newer, prospectively collected data is still necessary. Overall, though, the authors viewed their work as a clear success.
“Our model enables early detection and timely treatment of POAF and can prevent related adverse events and spontaneous AFib in the future,” they wrote.
Read the full study here.
Michael has more than 18 years of experience as a professional writer and editor. He has written at length about cardiology, radiology, artificial intelligence and other key healthcare topics.