AI model predicts patient survival after cardiac surgery
An AI algorithm that can already spot patients with reduced left ventricular ejection fraction can now predict long-term survival following cardiac surgery, according to a new study published in Mayo Clinic Proceedings.
This advantage, according to the study's authors, provides clinicians with a valuable tool for evaluating risk as patients and their healthcare providers contemplate surgery.
The team behind the analysis believe it to be the first large-scale research initiative that focuses on whether or not AI algorithms built around a single ECG can predict cardiac surgery outcomes.
Researchers examined data from 20,627 patients treated at the Mayo Clinic from 1993 to 2019. All patients underwent coronary artery bypass grafting, valve surgery or both. They also all presented with a left ventricular ejection fraction of more than 35%.
Overall, 17,125 presented with a normal AI EKG screen and 3,502 had an abnormal screen. Patients with an abnormal screen were more likely to be older and present with more comorbidities.
The novel algorithm was applied to each patient's most recent ECG within 30 days of their procedure.
The probability of survival after five years was 86.2% for patients with a normal screen versus 71.4% for those with an abnormal screen. In addition, the 10-year probability of survival was 68.2% for a patient with a normal screen and 45.1% for those with an abnormal screen.
"Our study finds there is a clear correlation between long-term mortality and a positive AI ECG screen for reduced ejection fraction among patients without apparent severe cardiomyopathy," lead author Mohamad Alkhouli, MD, a cardiologist at the Mayo Clinic, said in a statement. "This correlation was consistent among patients undergoing valve, coronary bypass, or valve and coronary bypass surgery."
In addition, Alkhouli added, these findings "may aid in risk stratification of patients referred for surgery and facilitate shared decision-making."
The group also added that more studies were being conducted to determine whether the information provided by their algorithm can potentially improve diagnoses, decision-making and clinical outcomes.
The full study is available here.