Development and validation of an interpretable machine learning model for early prediction of postoperative atrial fibrillation following on-pump cardiac surgery

Miaomiao Qian, Jie Yang, Dandan Geng

Abstract:

Postoperative atrial fibrillation (POAF) complicates 20%–40% of cardiac surgeries, increasing morbidity, length of stay, and healthcare costs. Early risk stratification could enable targeted prophylactic interventions. This study aimed to develop and validate an interpretable machine learning model for predicting incident POAF following on-pump cardiac surgery using routinely available perioperative parameters.

Introduction

Postoperative atrial fibrillation (POAF) represents the most frequently encountered arrhythmia following cardiac surgery, with reported incidence rates ranging from 20% to 40% among patients undergoing coronary artery bypass grafting (CABG) and valvular procedures (1). This complication substantially increases the risk of thromboembolic events, prolongs intensive care unit (ICU) and hospital length of stay, elevates healthcare costs, and is associated with increased short- and long-term mortality (2, 3).

Methods

This retrospective cohort study was conducted at the Department of Cardiovascular Surgery, First Affiliated Hospital of Nanjing Medical University. Consecutive adult patients (≥18 years) who underwent on-pump cardiac surgery between January and December 2025 were screened.

Results

Between January and December 2025, 2,260 patients underwent cardiac surgery, of whom 1,054 met eligibility criteria. The development cohort and temporal validation cohort were partitioned as previously described (Figure 1). Within the development cohort, 236 patients (27.1%) developed incident AF during the index ICU stay.

Discussion

The core finding of this study is that a ML model combining immediate postoperative laboratory indicators predicts POAF with moderate accuracy, with ChE and age serving as key drivers.

Conclusion

In summary, this study developed and temporally validated an interpretable RF model for predicting incident POAF following on-pump cardiac surgery. By utilizing LASSO regression to identify 11 optimal predictors and SHAP analysis to reveal nonlinear risk patterns, we identified ChE and age as the predominant model-derived drivers of POAF.

Citation: Qian M, Yang J and Geng D (2026) Development and validation of an interpretable machine learning model for early prediction of postoperative atrial fibrillation following on-pump cardiac surgery. Front. Surg. 13:1933263. doi: 10.3389/fsurg.2026.1933263

Received: 09 July 2026, Revised: 17 August 2026, Accepted: 25 August 2026, Published: 10 September 2026

Copyright: © 2026 Qian, Yang and Geng. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

Correspondence: 

Dandan Geng gengdandan1988@126.com
These authors have contributed equally to this work

Disclaimer

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.

Abbreviations

ABG, Arterial blood gas; AF, Atrial fibrillation; ALP, Alkaline phosphatase; ALT, Alanine aminotransferase; APTT, Activated partial thromboplastin time; AST, Aspartate aminotransferase; AUC, Area under the curve; BE, Base excess; BMI, Body mass index; BUN, Blood urea nitrogen; CABG, Coronary artery bypass grafting; ChE, Cholinesterase; DT, Decision tree; EHR, Electronic health records; FIB, Fibrinogen; GGT, Gamma-glutamyl transferase; ICU, Intensive care unit; INR, International normalized ratio; Lac, Lactate; LASSO, Least absolute shrinkage and selection operator; LR, Logistic regression; MAP, Mean arterial blood pressure; MCH, Mean corpuscular hemoglobin; MCHC, Mean corpuscular hemoglobin concentration; MCV, Mean corpuscular volume; ML, Machine learning; MPV, Mean platelet volume; NB, Naive Bayes; PaCO₂, Partial pressure of carbon dioxide; PaO₂, Partial pressure of oxygen; PCT, Plateletcrit; PDW, Platelet distribution width; PLT, Platelet count; POAF, Postoperative atrial fibrillation; PRAUC, Precision-recall area under the curve; PT, Prothrombin time; RBC, Red blood cell; RDW, Red cell distribution width; RF, Random forest; ROC, Receiver operating characteristic; SaO₂, Arterial oxygen saturation; SHAP, SHapley Additive exPlanations; SpO₂, Peripheral oxygen saturation; SVM, Support vector machine; t-SNE, t-distributed stochastic neighbor embedding; TRIPOD, Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis; TT, Thrombin time; XGBoost, Extreme gradient boosting.

Funding

The author(s) declared that financial support was received for this work and/or its publication. This study was supported by the Nursing Innovation Support Project of the Huai Nursing Fund, China Social Welfare Foundation (Grant No. HLCXKT-20230180).

Data availability statement

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.