Phenotype-Based Risk Stratification in Patients With New-Onset Atrial Fibrillation Following Myocardial Infarction Using a Data-Driven Approach.

BACKGROUND

Atrial fibrillation (AF) after myocardial infarction (MI) is associated with significant morbidity and mortality, yet existing risk stratification tools perform poorly in this population. Conventional predictors fail to capture multidimensional risk, while data-driven approaches may identify latent phenotypes with improved prognostic value.

AIMS: We sought to identify phenotypes of new-onset AF after MI using principal component analysis (PCA), evaluate their association with adverse outcomes, and develop a simple decision-tree model for clinical risk stratification.

METHODS

We analyzed 451 patients with new-onset AF after MI. PCA identified low-, intermediate-, and high-risk phenogroups. The primary endpoint was 3-year mortality, stroke, or MACE. Kaplan-Meier and multivariable Cox models assessed outcomes. Decision trees predicted phenogroup membership, with angiographic features compared across risk tiers.

RESULTS

PCA identified three phenogroups (low, intermediate, high risk) driven mainly by LVESV, LVEDV, LVEF, sex, and CHA2DS2-VASc. Clusters showed clear separation on PCA mapping. Event-free survival differed significantly across groups (log-rank p < 0.0001), with the lowest survival in the high-risk cluster. After multivariable adjustment for clinical risk factors, phenogroups remained independently associated with adverse outcomes. A simplified decision tree using LVEF, CHA2DS2-VASc, LVEDV, age, and TR Vmax predicted risk with 72.3% accuracy, supporting bedside risk stratification in post-MI AF patient populations.

CONCLUSION

Phenomapping identified three distinct AF risk groups with different clinical outcomes. LV size and function were the strongest contributors to phenotypic separation. The decision tree offered a clear, accurate, and interpretable framework that supports a new strategy for identifying appropriate patients with AF after MI.