Postoperative New-Onset Heart Block in Noncardiac Surgery: Model Development, Validation, and Long-Term Prognostic Analysis

Comprehensive Summary

This retrospective cohort study by Zhang et al. investigates new-onset heart block after noncardiac related surgeries, created a predictive model to examine risk factors, and examined potential complications that follow this condition. Electronic medical records of all patients who underwent noncardiac surgery at the Chinese PLA General Hospital between January 1st, 2008, and August 1st, 2019 were collected and exclusion criteria included any surgeries that lasted less than 30 minutes and patients that were younger than 18 years old. The primary outcome was new-onset postoperative heart block that developed within 30 days after surgery and included left and right bundle branch blocks along with first, second, and third-degree AV blocks. These heart blocks were confirmed using 12-lead ECGs and read by ECG room physicians. In total, 281,497 noncardiac surgery patients met the criteria and 1000 patients developed postoperative new-onset heart blocks. The authors created a predictive model using a traditional logistic regression model and 4 machine learning algorithms: neural network classifier (Nnet), light gradient boosting machine (LGBM), support vector machine (SVM), and extreme gradient boosting (XGB) with classification trees. Area under the curve (AUC) tests were used to determine the discrimination of both the traditional logistic regression and the machine learning models. The incidence of new-onset heart block within 30 days after noncardiac surgery was determined to be 0.36% and several risk factors were identified. Female sex, history of coronary heart disease, history of heart failure, American Society of Anesthesiologists (ASA) score, surgical specialty, emergency surgery, blood loss, preoperative anticoagulant use, preoperative hyperglycemia, and surgery duration longer than 3 hours were identified as potential risk factors. The XGB algorithm had the highest predictive ability with AUCs of 0.808 and 0.804 for the training and validation sets, respectively. These AUCs were comparable to the AUCs constructed using logistic regression, highlighting that the ML model was as effective as traditional logistic regression at determining potential risk factors for new-onset heart block after noncardiac surgery. It was also determined that new-onset AV block increases earlier postoperative mortality, while new-onset RBBB significantly worsens late outcomes.

Outcomes and Implications

Overall, this study examined multiple important topics. First and foremost, this study suggests that machine learning models are just as effective as traditional regression models at predicting risk factors for noncardiac surgery postoperative new-onset heart block. Machine learning has become much more relevant in the past decade, and this paper cements its efficacy in medical research. The discovery that different types of heart block impact mortality at different stages suggests that postoperative care must be tailored to the specific type of conduction issue found. This differentiation shifts the focus toward long-term cardiological surveillance for seemingly stable patients, ensuring that those who develop even minor blocks are not lost to follow-up. Furthermore, these findings justify a more strategic allocation of cardiac monitoring resources to noncardiac surgical wards, particularly for prolonged or emergency procedures where the physiological stress on the heart's conduction system is highest. Ultimately, this research provides a framework for reducing preventable cardiac complications by integrating AI-assisted screening into routine preoperative planning and extending the window of postoperative observation.