Predictive value of systemic inflammatory indices for new-onset atrial fibrillation after acute myocardial infarction: an interpretable machine learning study

BACKGROUND: New-onset atrial fibrillation (NOAF) after acute myocardial infarction (AMI) is associated with adverse outcomes. Inflammation contributes to atrial fibrillation (AF) onset and progression. We developed an interpretable machine learning (ML) framework integrating complete blood count (CBC)-derived inflammatory indices with routine clinical parameters for NOAF risk assessment. METHODS: We retrospectively analyzed 3,077 AMI patients without prior AF (2018-2023). Multivariable logistic regression, restricted cubic splines (RCS), and threshold analyses evaluated six inflammatory indices: neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), systemic immune-inflammation index (SII), monocyte-to-lymphocyte ratio (MLR), systemic inflammation response index (SIRI), and aggregate index of systemic inflammation (AISI). Boruta and LASSO selections were intersected. Eight ML models were evaluated, with Shapley Additive Explanations (SHAP) for interpretation. Estimated change points were approximately 5.5 (NLR), 1217 (SII), 2.9 (SIRI), and 430 (AISI). The support vector machine (SVM) achieved the highest mean training out-of-fold (OOF) area under the precision-recall curve (PR-AUC; 0.357) and a held-out test PR-AUC of 0.407. Mean absolute SHAP values were highest for left atrial diameter (LAD), age, and B-type natriuretic peptide (BNP); SIRI ranked first among inflammatory features. Fixed-feature ablation showed modest PR-AUC and Brier-score improvements, uncertain ROC-AUC gains, and no consistent net benefit on decision-curve analysis. CONCLUSIONS: CBC-derived inflammatory indices were associated with NOAF, but their incremental predictive value beyond routine clinical variables remains uncertain. Given moderate predictive performance and no external validation, multicenter validation is needed before clinical implementation.