Phenotypes of South Asian patients with atrial fibrillation and holistic integrated care management: cluster analysis of data from KERALA-AF Registry

Abstract

Background: Patients with atrial fibrillation (AF) frequently experience multimorbidity. Cluster analysis, a machine learning method for classifying patients with similar phenotypes, has not yet been used in South Asian AF patients.

Methods: The Kerala Atrial Fibrillation Registry is a prospective multicentre cohort study in Kerala, India, and the largest prospective AF registry in South Asia. Hierarchical clustering was used to identify different phenotypic clusters. Outcomes were all-cause mortality, major adverse cardiovascular events (MACE), and composite bleeding events within one-year follow-up.

Findings: 3348 patients were included (median age 65.0 [56.0-74.0] years; 48.8% male; median CHA₂DS₂-VASc 3.0 [2.0-4.0]). Five clusters were identified. Cluster 1: patients aged ≤65 years with rheumatic conditions; Cluster 2: patients aged >65 years with multi-comorbidities, suggestive of cardiovascular-kidney-metabolic syndrome; Cluster 3: patients aged ≤65 years with fewer comorbidities; Cluster 4: heart failure patients with multiple comorbidities; Cluster 5: male patients with lifestyle-related risk factors. Cluster 1, 2 and 4 had significantly higher MACE risk compared to Cluster 3 (Cluster 1: OR 1.36, 95% CI 1.08-1.71; Cluster 2: OR 1.79, 95% CI 1.42-2.25; Cluster 4: OR 1.76, 95% CI 1.31-2.36). The results for other outcomes were similar. Atrial fibrillation Better Care (ABC) pathway in the whole cohort was low (10.1%), especially in Cluster 4 (1.9%). Overall adherence to the ABC pathway was associated with reduced all-cause mortality (OR 0.26, 95% CI 0.15-0.46) and MACE (OR 0.45, 95% CI 0.31-0.46), similar trends were evident in different clusters.

Interpretation: Cluster analysis identified distinct phenotypes with implications for outcomes. There was poor ABC pathway adherence overall, but adherence to such integrated care was associated with improved outcomes.

Funding: Kerala Chapter of Cardiological Society of India.

Keywords: ABC pathway; Atrial fibrillation; Clustering analysis; Kerala; Phenotype classification; South Asia.

Fig. 1

Clustering process of this study. The cluster analysis strategy was to minimise the variance of attribute differences within clusters using Ward's approach and the algorithm of squared Euclidean distance. Initially each sample is considered as a separate cluster, the distance metric between clusters is repeatedly updated during the aggregation of clusters, and finally formed different clusters with similar characteristics. The complete hierarchical clustering process was visualised as a dendrogram, where different vertical lines indicated different clusters and the Y-axis represented the distance measure of different clusters with the further away from the end of the tree and the greater the differences between clusters. By observing the dendrogram generated by the clustering process, the median Y-axis value was around 40, and the distance between the last four mergers in the clustering process was significantly greater than the previous mergers, therefore 5 clusters were selected.

Fig. 2

Patient characteristics of different clusters by hierarchical clustering. AF, atrial fibrillation; AR, aortic regurgitation; BMI, body mass index; CVA, cerebrovascular accident; LBBB, left bundle branch block; LVH, left ventricular hypertrophy; MR, mitral regurgitation; NYHA, New York Heart Association; PAH, pulmonary hypertension; RBBB, right bundle branch block; RWMA, regional wall motion abnormalities; SE, systemic embolism; TIA, transient ischaemic attack.

Fig. 3

AF treatment managment of different clusters by hierarchical clustering. ACEI, angiotensin-converting enzyme inhibitors; AF, atrial fibrillation; ARB, angiotensin II receptor blockers; CCB, calcium channel blockers; ICD, Implantable cardioverter defibrillator; LAAO, left atrial appendage occlusion.

Fig. 4

Risk for outcomes on multivariate logistic regression analysis among different clusters by hierarchical clustering. Adjust for beta-blockers, rate-limiting calcium channel blockers, digoxin, angiotensin-converting enzyme inhibitors, angiotensin II receptor blockers, dihydropyridine calcium channel blockers, diuretics, statins, Class I AAD, Class III AAD, antiplatelet agents, anticoagulants, pacemaker implantation, surgery for atrial fibrillation, cardiac defibrillator implantation, left atrial appendage occlusion, and catheter ablation. AAD, antiarrhythmic drug; ACS, acute coronary syndrome; CI, confidence interval; CVA, cerebrovascular accident; HF, heart failure; MACE, major adverse cardiovascular events; OR, odds ratio; SE, systemic embolism; TIA, transient ischemic attack.

Fig. 5

The A criterion, B criterion, C criterion, and ABC criteria in the whole cohort and each cluster by hierarchical clustering.

Fig. 6

Impact of adherence to the ABC pathway on all-cause mortality and mace in the whole cohort and each cluster by hierarchical clustering. Adjusted for age, sex, type of atrial fibrillation, diabetes mellitus, heart failure, hypertension, dyslipidaemia, coronary artery disease, and prior cerebrovascular disease or transient ischaemic attack. CI, confidence interval; MACE, major adverse cardiovascular events; OR, odds ratio.