Machine learning models of inflammatory markers used in the prediction of early and late atrial fibrillation recurrence in overweight and obese patients after catheter ablation

BACKGROUND: Inflammation is associated with atrial fibrillation (AF) recurrence after catheter ablation. This study aimed to create machine learning models of inflammatory biomarkers and clinical parameters to predict early recurrence of atrial fibrillation (ERAF) and late recurrence of atrial fibrillation (LRAF) after catheter ablation. METHODS: The study included 114 patients with abnormal body weight, defined as body mass index (BMI) > 25 kg/m², and paroxysmal, persistent, or long-standing persistent AF. Blood samples were collected at baseline and 24 hours after ablation. Almost 120 clinical and laboratory parameters were selected to develop machine learning models of ERAF and LRAF. Shapley additive explanations (SHAP) were derived to explain the obtained predictions. The models were trained and tuned for hyperparameters by five-fold cross-validation and subsequently evaluated on independent test sets using stratified sampling. RESULTS: ERAF was observed in 20.3% of patients. LRAF was diagnosed in 26.3% of patients. The ERAF model with 5 variables - monocyte count, platelet-to-lymphocyte ratio (PLR), fibrinogen, and parameters measured before ablation, such as glomerular filtration rate (GFR) and left atrial volume - performed well in the studied cohort (mean AUC, 0.70 ± 0.06). The LRAF model with parameters such as neutrophil-to-lymphocyte ratio (NLR), soluble vascular cell adhesion molecule-1 (sVCAM-1), troponin, monocyte-to-HDL-C ratio (MHR), adiponectin, and ERAF also showed valuable performance (mean AUC, 0.77 ± 0.11). CONCLUSIONS: The machine learning model based on inflammatory biomarkers is a valuable tool for predicting ERAF. The combination of ERAF and inflammatory biomarkers significantly improves LRAF prediction.