Abstract Funding Acknowledgements Type of funding sources: None. Introduction Atrial fibrillation (AF) is a disease with heterogeneous underlying conditions which usually encompass several cardiovascular comorbidities. There is a lack of large studies investigating the heterogeneity of patients with AF in Asian population. A refined classification may contribute on individual precise treatment and prognosis. Methods This cohort study was conducted utilizing a database from a tertiary medical center. Between 2014 and 2019, a total of 5002 adult patients with AF were enrolled for analysis. We performed an unsupervised hierarchical cluster analysis based on CHA2DS2-VASc score after model assessment. The risk of transient ischemia accident (TIA)/ischemic stroke, heart failure (HF) hospitalization, cardiovascular mortality, and all-cause mortality were assessed. Results We identified four replicable distinct clusters of patients with AF: cluster I included diabetic patients with HF with preserved ejection fraction and chronic kidney disease; cluster II included elder patients with a low BMI and pulmonary hypertension; cluster III included patients with metabolic syndrome and atherosclerotic disease; and cluster IV included patients with left heart dysfunction including reduced ejection fraction and enlarged left atrium. The incidence of stroke were 12.7%, 15.1%, 9.9%, and 6.3% in clusters I to IV respectively. Cox regression analyses showed that the differences in risk of TIA/ischemic stroke risk across clusters (cluster I, II, III vs. IV) were statistically significant (Hazard ratio (HR) 1.87 [95% CI 1.00–3.48], 2.06 [95% CI 1.06–4.01], 1.70 [95% CI 1.02–2.84]). Cluster II was independently associated with highest risk for stroke (HR 2.06 [1.06-4.01]), HF hospitalization (HR 1.19 [95% CI 0.79-1.80]), cardiac death (HR 2.51 [95% CI 1.21-5.22]), and all-cause death (HR 2.98 [95% CI 1.98-4.50]). Conclusion The study indicated the feasibility of clinical application of the cluster analysis in a highly heterogeneous cohort of patients with AF after accounting for CHA2DS2-VASc risk scores. Patients can be divided into four phenotypes with distinct patient characteristics and cardiovascular outcomes.