Abstract Background Atrial fibrillation (AF) is associated with cognitive impairment even in the absence of previous stroke. However, data on specific patterns of cognitive impairment and their association with outcomes are limited. Purpose We aimed to identify different phenotypic groups of AF patients stratified according to specific cognitive domains using cluster analysis and evaluate the association between identified clusters and adverse outcomes. Methods We used data from a prospective observational study on AF patients excluding those with previous clinically overt stroke. We performed a hierarchical cluster analysis based on Ward’s Method using data from the Mini-Mental State Examination (MMSE) administered during the enrollment. The MMSE includes 6 distinct cognitive domains. We normalized each domain’s score from 0 to 100 to preserve the ranking and the interpretability of the test. Patients’ clusters were identified by examining the distance between cluster coefficients and by visual inspection of the dendrogram. All-cause mortality was the primary endpoint. Results A total of 872 stroke-free AF patients were included. Three clusters based on MMSE domains were identified: Cluster 1 (n=494, 56.7%); Cluster 2 (n=243, 27.9%) and Cluster 3 (n=135, 15.5%). The clusters greatly differed according to the cognitive domains affected and baseline characteristics (Table). The use of oral anticoagulants did not differ among clusters and it was substantially high (88.6%). Cluster 2 had a higher prevalence of cognitive impairment (37.9% vs 35.6% for Cluster 3 and 1.0% for Cluster 1, p<0.001) with a lower mean MMSE score compared to other clusters. The analysis of MMSE domains showed substantial deficits in the orientation, attention/calculation, and recall domains with peculiar differences among the clusters. Cluster 2 showed the most significant impairment in the copying domain (100%) while Cluster 3 had a higher impairment in attention/calculation and recall domains. After a median follow-up of 629 days, all-cause mortality was 15.9% with a significantly higher prevalence in Cluster 2 (26.9% vs 10.0% and 15.5% for Cluster 1 and Cluster 3 respectively, p<0.001). At the adjusted Cox regression analysis, Cluster 2 was independently associated with a higher risk of all-cause mortality compared with Cluster 1 (HR 1.83, 95% CI 1.18-2.82) while Cluster 3 was not independently associated with the outcome (Figure). Conclusions In a prospective observational cohort of stroke-free AF patients, we identified three clusters showing different patterns of cognitive impairment, affecting specific MMSE domains and resulting in a different risk of mortality. Our findings highlight the clinical implications of cognitive impairment and stress the need for a more holistic approach to AF patients including a comprehensive cognitive assessment, as well as the need for studies on interventions able to positively affect cognitive impairment.Figure