TOPLINE:
An artificial intelligence (AI) deep-learning algorithm interpreting preoperative ECGs can identify risk for postoperative death in those undergoing cardiac surgery, noncardiac surgery, and interventional procedures, a large new study showed. The algorithm was more effective in identifying high-risk patients who went on to experience postoperative mortality than a widely used risk tool.
METHODOLOGY:
- Researchers evaluated the performance of an AI algorithm (PreOpNet) trained on preoperative ECGs in 36,839 patients, mean age 65 years, undergoing procedures at Cedars-Sinai Medical Center (CSMC) from 2015 to 2019 who had at least one 12-lead ECG performed within 30 days before the procedure.
- The main outcome was mortality after cardiac surgery, noncardiac surgery, and procedures performed in the catheterization laboratory or endoscopy suite, up to 30 days post-procedure.
- Researchers compared the performance of PreOpNet with the Revised Cardiac Risk Index (RCRI), an established risk calculator that uses preoperative clinical characteristics from electronic medical records.
- To assess the accuracy of PreOpNet in hospital settings with diverse patient populations, researchers applied the algorithm to cohorts from two separate external healthcare systems: Stanford Healthcare (SHC) and Columbia University Medical Center (CUMC).
TAKEAWAY:
- The algorithm discriminated mortality with an area under the curve (AUC) of 0.83 (95% CI, 0.79-0.87) compared to conventional RCRI (AUC, 0.67; 95% CI, 0.61-0.72).
- Patients determined to be high risk by the deep-learning model had an unadjusted odds ratio (OR) for postoperative mortality of 9.17 (95% CI, 5.85-13.82) compared with an unadjusted OR of 2.08 (0.77-3.50) for RCRI scores of more than 2, an indicator of high risk.
- PreOpNet performed similarly in discriminating mortality in patients undergoing cardiovascular surgery (AUC, 0.85; 95% CI, 0.77-0.92) and in those undergoing noncardiac surgery (AUC, 0.83; 95% CI, 0.79-0.88); however, for the RCRI score, the AUC was 0.62 (95% CI, 0.52-0.72) in patients undergoing cardiac surgery and 0.70 (95% CI, 0.63-0.77) in those undergoing noncardiac surgery.
- The external validation analysis showed the algorithm discriminated postoperative mortality with AUCs of 0.75 (95% CI, 0.74-0.76) in the SHC and 0.79 (95% CI, 0.75-0.83) in the CUMC cohort, with similar specificity, sensitivity, and positive and negative predictive value as with the CSMC cohort.
IN PRACTICE:
"Current clinical risk prediction tools are insufficient," study lead author David Ouyang, MD, Department of Cardiology, Smidt Heart Institute and Division of Artificial Intelligence in Medicine, Department of Medicine, CSMC, Los Angeles, said in a press release, adding this AI model "could potentially be used to determine exactly which patients should undergo an intervention and which patients might be too sick."
SOURCE:
The study was carried out by Ouyang and colleagues. It was published online on December 7, 2023, in The Lancet Digital Health.
LIMITATIONS:
The algorithm might not be applicable to low-risk patients who don't require preoperative ECG. As RCRI is designed to be evaluated in patients undergoing noncardiac surgery, the most direct comparison is in this setting (AUC, 0.83 vs 0.70 for PreOpNet and RCRI, respectively). All analyses were performed on retrospective cohorts.
DISCLOSURES:
The study received funding from the National Heart, Lung, and Blood Institute. Ouyang reports support from the National Institutes of Health and Alexion and consulting or honoraria for lectures from EchoIQ, Ultromics, Pfizer, InVision, the Korean Society of Echo, and the Japanese Society of Echo; see paper for disclosures of other authors.
Comments