Preoperative electrocardiography for predicting cardiovascular events after noncardiac surgery: a secondary analysis of two prospective cohorts

Abstract

Background

The predictive value of preoperative resting ECGs for cardiovascular events after noncardiac surgery is unclear. This study evaluated whether incorporating conventional ECG features or the output of a deep-learning algorithm for ECG waveform analysis (PreOpNet) improves risk prediction beyond established clinical risk scores.

Methods

We conducted a secondary analysis of two prospective cohorts of patients undergoing major noncardiac surgery in China between 2019 and 2023. Eight prespecified conventional ECG features were extracted from ECG reports, and raw ECG waveforms were analysed using PreOpNet. The primary outcome was a composite of any cardiovascular events within 30 days after surgery; the secondary outcome was major adverse cardiac events (MACEs). We built nested logistic regression models based on clinical risk scores, with or without ECG findings. Predictive performance was evaluated using area under the receiver-operating-characteristic curve (AUC), risk reclassification metrics, and decision curve analyses.

Results

Among 6080 patients, 733 (12.1%) experienced postoperative cardiovascular events and 174 (2.9%) had MACEs. Adding conventional ECG features (ΔAUC 0.035) or PreOpNet (ΔAUC 0.032) to the Revised Cardiac Risk Index improved model discrimination modestly. When added to the Gupta Perioperative Myocardial Infarction or Cardiac Arrest risk score, both approaches yielded minimal gains (ΔAUC 0.010 for conventional ECG features; 0.006 for PreOpNet). Risk reclassification and decision curve analyses had limited incremental predictive value.

Conclusions

Preoperative ECG findings, assessed by including conventional waveform features and the PreOpNet algorithm, provide limited incremental predictive value for cardiovascular risk assessment before noncardiac surgery.