https://orcid.org/0000-0001-8684-2361
Summary
Phenotypic classification is a method of grouping patients with similar phenotypes.
We aimed to use phenotype classification based on a clustering process for risk stratification of patients with non-valvular atrial fibrillation (AF) and second, to assess the benefit of the Atrial Fibrillation Better Care (ABC) pathway.
Patients with AF were prospectively enrolled from 27 hospitals in Thailand from 2014 to 2017, and followed up every 6 months for 3 years. Cluster analysis was performed from 46 variables using the hierarchical clustering using the Ward minimum variance method. Outcomes were a composite of all-cause death, ischemic stroke/systemic embolism, acute myocardial infarction and heart failure.
A total of 3405 patients were enrolled (mean age 67.8 ± 11.3 years, 58.2% male). During the mean follow-up of 31.8 ± 8.7 months. Three clusters were identified: Cluster 1 had the highest risk followed by Cluster 3 and Cluster 2 with a hazard ratio (HR) and 95% confidence interval (CI) of composite outcomes of 2.78 (2.25, 3.43), P < 0.001 for Cluster 1 and 1.99 (1.63, 2.42), P < 0.001 for Cluster 3 compared with Cluster 2. Management according to the ABC pathway was associated with reductions in adverse clinical outcomes especially those who belonged to Clusters 1 and 3 with HR and 95%CI of the composite outcome of 0.54 (0.40, 073), P < 0.001 for Cluster 1 and 0.49 (0.38, 0.63), P < 0.001 for Cluster 3.
Phenotypic classification helps in risk stratification and prognostication. Compliance with the ABC pathway was associated with improved clinical outcomes.
Background
Despite the increasing use of oral anticoagulation (OAC) for stroke prevention in patients with atrial fibrillation (AF) over the past decade, the residual risks of mortality and morbidity remain high.1,2 Recent guidelines suggested that the management of AF patients should not focus only on OAC but adopt a more holistic or integrated care approach to better characterization3 and management, based on the Atrial Fibrillation Better Care (ABC) pathway approach which also includes symptom management and appropriate treatment for cardiovascular risk factors and comorbidities.4,5 Compliance with such an approach has been associated with improved clinical outcomes, including lower mortality, stroke, bleeding and hospitalizations.6
Importantly, single risk factors do not occur in isolation and occur in clusters of comorbidities. Indeed, clustering based on clinical phenotypes can help define ‘clinically complex’ patient groups7 that are associated with high risks of adverse outcomes. These high risk groups are often undertreated or discontinued treatments after started8 Clustering of clinical phenotypes may help in risk stratification and guiding therapy.9,10
Phenotype classification studies of AF patients have shown that Asian patients may be different from non-Asians in AF clinical phenotypes and clinical risk prediction.9,11,12 Such a clustering process helps to improve patients with increased risk, but there are limited data on how integrated care would impact outcomes in these complexity cluster phenotypes.
The primary objective of our study was to use phenotype classification based on a clustering process for risk stratification of patients with non-valvular AF and second, to assess the benefit from compliance to holistic or integrated care management based on the ABC pathway.
Materials and methods
Study subjects
This is a prospective non-valvular AF registry from 27 hospitals in Thailand conducted between 2014 and 2017 for the COhort of antithrombotic use and Optimal INR Level in patients with non-valvular Atrial Fibrillation in Thailand (COOL-AF Thailand) registry. Fourteen of those centers are university hospitals, and 13 are regional or general hospitals. Protocols were established and followed by the data management team and statisticians to ensure the integrity and quality of the data before the final analysis. Patients with age at least 18 years with documented AF by 12-lead ECG or ambulatory ECG monitoring were included. Exclusion criteria were as follows: rheumatic mitral disease, mechanical prosthetic heart valve, ischemic stroke within 3 months, AF from transient cause(s), hematologic conditions that increased risk of bleeding, refusal to participate, life expectancy <3 years, participating in the pharmaceutical clinical trial or unable to have follow-up visit(s). This study was approved by the Central Research Ethics Committee of Thailand. Written informed consent was given prior to the participation of each patient. The study was conducted in accordance with the International Conference on Harmonization for Good Clinical Practice Guidelines and the principles in the Declaration of Helsinki.
Study conduct
The investigators of each study site were informed to enroll patients consecutively. After the patient signed their informed consent, site investigators interviewed the patients and reviewed medical record and to acquire the data. Those data were then transferred from the CRF to a web-based system by a study nurse or research assistant. The web-based system was designed to centralize data and to eliminate the possibility of missing data or errors. Recorded forms were sent to the central site. Once received at the central data management site, the data from the CRF were checked and reentered into the system for purposes of data validation. A monitoring process was conducted for every study site to confirm data quality and ensure that the study was conducted in accordance with good clinical practice. The site investigators were required to record follow-up data every 6 months until 3 years.
Data collection
At enrolment, the following data were needed: demographic data, AF types and symptoms, vital signs, body weight, height, cardiovascular risk factors such as diabetes, hypertension, smoking, comorbidities, laboratory findings, ECG and echocardiogram results and medications. Components of the CHA2DS2-VASc score (C = congestive heart failure −1 point; H = hypertension −1 point; A = age ≥75 years −2 points; D = diabetes −1 point; S = stroke/transient ischemic attack (TIA) −2 points; V = vascular disease −1 point; A = age 65–74 years −1 point; and Sc = female sex −1 point); HAS-BLED score (uncontrolled hypertension, abnormal renal, or liver function; history of stroke; history of bleeding; labile INR; elderly (age ≥65 years)); and drugs or alcohol (1 point each) were recorded.
The definition of ABC pathway adherence was based on the original definition.13 Adherence to component A (avoid stroke) was achieved if the patient received an appropriate OAC strategy according to their stroke risk at baseline. The A adherence was met when male patients with CHA2DS2-VASc score ≥ 1 or female patients with CHA2DS2-VASc score ≥ 2 received an OAC, and male patients with CHA2DS2-VASc score = 0 or female patients with CHA2DS2-VASc score ≤ 1 did not receive an OAC. Component B (better symptom management) adherence was fulfilled if patients had a European Heart Rhythm Association score ≤ 2. For component C (cardiovascular risk factors and comorbidity management), adherence was considered if patients had appropriate management. This included (i) hypertension management with angiotensin-converting enzyme inhibitors (ACEi)/angiotensin (II) receptor blockers (ARB), calcium channel blockers (CCB), diuretics or beta-blockers (BB), (ii) coronary artery disease management with ACEi/ARB, BB, statins and antiplatelet, (iii) ischemic stroke/TIA management with statins and antiplatelet, (iv) heart failure management with ACEi/ARB and BB and (v) diabetes management with oral antidiabetics or insulin. Adherence to ABC pathway was achieved when the patient met all three components. Definitions of ABC adherence are shown in Supplementary Table S1.
Data during the follow-up visits were similar to the baseline visit. The clinical outcomes were also recorded. Supporting documents were required to be uploaded to the web system for verification.
Outcomes
The main outcome of this study was a primary composite outcome which included ‘all-cause death, ischemic stroke/systemic embolism (SSE), acute myocardial infarction (MI) and heart failure (HF)’. Ischemic stroke was defined as a sudden focal neurological deficit lasting at least 24 h. TIA was similarly defined with complete recovery in 24 h. Imaging data such as computerized tomography (CT) brain scan or magnetic resonance imaging (MRI) were required to be uploaded into the web. Systemic embolism requires both clinical and objective evidence of sudden loss of organ perfusion. Acute MI was defined as the detection of a rise and/or fall of cardiac troponin with at least one value above the 99th percentile upper reference limit and with at least one of the following: symptoms, new significant ST–T changes or new left bundle branch block (LBBB), development of pathological Q waves or imaging evidence of new loss of viable myocardium or new regional wall motion abnormality. HF was defined as a hospital admission, or an urgent, unscheduled clinic/office/emergency department visit with a primary diagnosis of HF, with new or worsening HF symptoms, objective evidence of HF and received initiation or intensification of HF treatment.
Site investigators were required to upload the supporting documents for clinical events into the web-based system for event verification and adjudication process. All outcomes were adjudicated by an adjudication committee. Queries were made for the site investigators during the verification process.
Statistical analysis
The clustering process was performed by hierarchical clustering by the use of the Ward minimum variance method. The Euclidean distance was used as a measure of distance or dissimilarity. A dendrogram was used to display the clusters according to the Euclidean distance. We use the following 46 variables for the clustering process: age, body mass index (BMI), estimated glomerular filtration rate (eGFR), hematocrit, diastolic blood pressure, systolic blood pressure, heart rate, dyslipidemia, coronary artery disease (CAD), history of myocardial infarction (MI) or unstable angina, history of coronary artery bypass graft (CABG), history of percutaneous coronary intervention (PCI), history of drug eluting stent, type of AF, pacemaker, New York Heart Association (NYHA) class III/IV, internal cardioverter defibrillator (ICD), biventricular pacing, biventricular pacing with ICD (BiV/ICD), intraventricular conduction defect (IVCD), female gender, hypothyroidism, hypertension, diabetes, history of gastrointestinal (GI) bleeding, renal replacement therapy (RRT)/kidney transplantation, anemia, dementia, peripheral artery disease (PAD), history of ischemic stroke or TIA, cardioversion history, rhythm control medications, catheter ablation of AF, AV node ablation pacemaker, LBBB, electrocardiographic evidence of left ventricular hypertrophy (LVH), current smoker, hyperthyroidism, liver function, alcohol abuse, mild left atrial (LA) enlargement, LA enlargement, severe LA enlargement, mild left ventricular systolic dysfunction, moderate left ventricular systolic dysfunction and severe LV dysfunction. The missing values were imputed. After the display of the dendrogram, an appropriate number of clusters was determined by the decision whether the final number of clusters represented a homogenous group with relatively similar characters.
Continuous data are shown as mean and standard deviation (SD). Categorical data are displayed as counts and percentages. Comparisons of the three clusters were performed by the ANOVA test with Bonferroni post-hoc analysis for continuous data and the chi-square test for categorical data. The incidence rate of the composite and individual outcome are shown as the incidence rate per 100 person-years. The comparisons of the incidence rate of outcome were made by the Poisson regression. Kaplan–Meier analysis was performed to assess the survival free of clinical events. Log-rank test was used to analyze the significant difference between groups of the survival analysis. Multivariable Cox proportional hazard model analysis was performed to assess clusters as predictors for clinical outcomes using the time data. The results are shown as hazard ratio (HR) and 95% confidence interval (CI). Multivariable Cox proportional hazard model analysis was also used to determine the effect of ABC compliance on the clinical outcomes of all patients as well as among each cluster.
Sensitivity analysis was performed to assess the predictive value of ABC adherence on clinical outcomes by considering the achievement of the target for certain variables; a time in the therapeutic range (TTR) of ≥65% for those who received warfarin and a blood pressure <140/90 mmHg for those with hypertension. The achievement of treatment targets for other conditions is difficult to assess due to lots of missing data for low-density lipoprotein (LDL) cholesterol and hemoglobin A1C. TTR was calculated by the Rosendaal method.14 Among 2233 patients with warfarin, 107 cannot calculate TTR (4.8%). There was no missing value for blood pressure.
A P value of <0.05 was considered statistical significance. Statistical analysis was performed by the R version 3.6.3 (www.r-project.org) and the SPSS statistical software version 18.0 (SPSS, Inc., Chicago, IL, USA).
Results
Patient characteristics
A total of 3405 patients were studied (mean age 67.8 ± 11.3 years, 58.2% male). A flow diagram of the study population is shown in Supplementary Figure S1. The location of enrolment was the outpatient department (OPD) of cardiology in 2226 (65.4%), the OPD of internal medicine in 961 (28.2%). The rest were enrolled from other departments. The dendrogram of the clustering process using Ward linkage is shown in Figure 1. When the Euclidean distance was <50, the clusters became more heterogeneous, hence we concluded that a cluster number of three groups is appropriate. Patients in Cluster 1 were older adults with a high proportion of cardiovascular risk factors and comorbidities (i.e. older adults and sick). Cluster 2 subjects were relatively young adults with a small proportion of risk factors or comorbidities (i.e. young and healthy) whereas Cluster 3 patients were mainly older adults but the prevalence of risk factors and comorbid conditions was not as high as Cluster 1 (i.e. older adults and less sick). Baseline clinical information of each cluster is shown in Table 1.
Dendrogram generated by hierarchical clustering process using Ward linkage. Each branch represents each patient. Vertical lines are clusters that are joined together depends on the relative degree of similarity between individual patient and the scale indicates the rescaled Euclidean distance at which clusters were joined; the more dissimilarity between clusters, the greater the height. The red line indicates the stopping location of the clustering process which results in three clusters for AF patients.
Baseline characteristics of study population
| Variables | All (n = 3405) | Cluster 1 (n = 722) | Cluster 2 (n = 1302) | Cluster 3 (n = 1381) | P value |
|---|---|---|---|---|---|
| Age (years) | 67.8 ± 11.3 | 75.6 ± 6.6 | 56.5 ± 7.2 | 74.3 ± 6.8 | <0.001 |
| Female gender | 1424 (41.8%) | 325 (45.0%) | 449 (34.5%) | 650 (47.1%) | <0.001 |
| Hypertension | 2330 (68.4%) | 475 (65.8%) | 742 (57.0%) | 1113 (80.6%) | <0.001 |
| Diabetes | 839 (24.6%) | 172 (23.8%) | 308 (23.7%) | 359 (26.0%) | 0.316 |
| Dyslipidemia | 1917 (56.3%) | 415 (57.5%) | 648 (49.8%) | 854 (61.8%) | <0.001 |
| Current smoker | 104 (3.1%) | 12 (1.7%) | 68 (5.2%) | 24 (1.7%) | <0.001 |
| Type of AF | 0.138 | ||||
| New or paroxysmal | 1148 (33.7%) | 224 (31.0%) | 464 (35.6%) | 460 (33.3%) | |
| Persistent | 645 (18.9%) | 131 (18.1%) | 253 (19.4%) | 261 (18.9%) | |
| Permanent | 1612 (47.3%) | 367 (50.8%) | 585 (44.9%) | 660 (47.8%) | |
| History of CAD | 547 (16.1%) | 238 (33.0%) | 155 (11.9%) | 154 (11.2%) | <0.001 |
| History of Ml or unstable angina | 207 (6.1%) | 116 (16.1%) | 52 (4.0%) | 39 (2.8%) | <0.001 |
| History of PCI | 253 (7.4%) | 147 (20.4%) | 69 (5.3%) | 37 (2.7%) | <0.001 |
| Drug eluting stent | 167 (4.9%) | 104 (14.4%) | 44 (3.4%) | 19 (1.4%) | <0.001 |
| History of CABG | 65 (1.9%) | 29 (4.0%) | 15 (1.2%) | 21 (1.5%) | <0.001 |
| History of peripheral artery disease | 44 (1.3%) | 14 (1.9%) | 10 (0.8%) | 20 (1.4%) | 0.066 |
| History of ischemic stroke or TIA | 592 (17.4%) | 136 (18.8%) | 182 (14.0%) | 274 (19.8%) | <0.001 |
| History of hyperthyroidism | 160 (4.7%) | 25 (3.5%) | 91 (7.0%) | 44 (3.2%) | <0.001 |
| History of hypothyroidism | 90 (2.6%) | 19 (2.6%) | 35 (2.7%) | 36 (2.6%) | 0.991 |
| Dementia | 29 (0.9%) | 8 (1.1%) | 6 (0.5%) | 15 (1.1%) | 0.148 |
| Cardioversion history | 87 (2.6%) | 8 (1.1%) | 57 (4.4%) | 22 (1.6%) | <0.001 |
| Catheter ablation of AF | 91 (2.7%) | 10 (1.4%) | 63 (4.8%) | 18 (1.3%) | <0.001 |
| AV node ablation pacemaker | 7 (0.2%) | 0 (0.0%) | 6 (0.5%) | 1 (0.1%) | 0.033 |
| GI bleeding history | 92 (2.7%) | 29 (4.0%) | 22 (1.7%) | 41 (3.0%) | 0.006 |
| Renal replacement therapy/kidney transplantation | 40 (1.2%) | 11 (1.5%) | 14 (1.1%) | 15 (1.1%) | 0.618 |
| Alcohol abuse | 140 (4.1%) | 15 (2.1%) | 99 (7.6%) | 26 (1.9%) | <0.001 |
| Rhythm control medications | 277 (8.1%) | 42 (5.8%) | 144 (11.1%) | 91 (6.6%) | <0.001 |
| NYHA class III/IV | 89 (2.6%) | 18 (2.5%) | 36 (2.8%) | 35 (2.5%) | 0.908 |
| Pacemaker | 269 (7.9%) | 107 (14.8%) | 53 (4.1%) | 109 (7.9%) | <0.001 |
| BiV | 1 (0.0%) | 0 (0.0%) | 0 (0.0%) | 1 (0.1%) | 0.480 |
| ICD | 55 (1.6%) | 16 (2.2%) | 15 (1.2%) | 24 (1.7%) | 0.171 |
| BiV/ICD | 16 (0.5%) | 8 (1.1%) | 7 (0.5%) | 1 (0.1%) | 0.004 |
| BMI (kg/m2) | 25.2 ± 4.7 | 23.5 ± 4.4 | 26.4 ± 4.9 | 24.9 ± 4.4 | <0.001 |
| Diastolic blood pressure (mmHg) | 128.5 ± 18.4 | 117 ± 15.5 | 126.1 ± 17.7 | 136.7 ± 16.6 | <0.001 |
| Systolic blood pressure (mmHg) | 75.2 ± 12.7 | 65.0 ± 9.8 | 76.9 ± 12.3 | 79.0 ± 11.5 | <0.001 |
| Heart rate (bpm) | 77.4 ± 16.2 | 70.9 ± 11.6 | 77.6 ± 15.4 | 80.5 ± 17.9 | <0.001 |
| Electrocardiographic evidence of LVH | 696 (20.4%) | 164 (22.7%) | 291 (22.4%) | 241 (17.5%) | 0.002 |
| Intraventricular conduction: non-specific IVCD | 145 (4.3%) | 49 (6.8%) | 32 (2.5%) | 64 (4.6%) | <0.001 |
| Intraventricular conduction: LBBB | 110 (3.2%) | 46 (6.4%) | 30 (2.3%) | 34 (2.5%) | <0.001 |
| eGFR | 62.1 ± 15.7 | 48.7 ± 4 | 81.2 ± 3.4 | 51 ± 5.6 | <0.001 |
| Hematocrit (%) | 39.6 ± 3.9 | 37.8 ± 4.1 | 41.3 ± 3.6 | 39 ± 3.5 | <0.001 |
| Anemia | 1293 (38.0%) | 448 (62.0%) | 205 (15.7%) | 640 (46.3%) | <0.001 |
| Liver function | 78 (2.3%) | 19 (2.6%) | 33 (2.5%) | 26 (1.9%) | 0.417 |
| LVEF (%) | 60.2 ± 13.7 | 59.6 ± 14.4 | 57.7 ± 14.6 | 62.8 ± 11.9 | <0.001 |
| LA dimension (mm) | 43.9 ± 8.3 | 43.2 ± 8.9 | 44.0 ± 7.5 | 44.0 ± 8.6 | 0.075 |
| Variables | All (n = 3405) | Cluster 1 (n = 722) | Cluster 2 (n = 1302) | Cluster 3 (n = 1381) | P value |
|---|---|---|---|---|---|
| Age (years) | 67.8 ± 11.3 | 75.6 ± 6.6 | 56.5 ± 7.2 | 74.3 ± 6.8 | <0.001 |
| Female gender | 1424 (41.8%) | 325 (45.0%) | 449 (34.5%) | 650 (47.1%) | <0.001 |
| Hypertension | 2330 (68.4%) | 475 (65.8%) | 742 (57.0%) | 1113 (80.6%) | <0.001 |
| Diabetes | 839 (24.6%) | 172 (23.8%) | 308 (23.7%) | 359 (26.0%) | 0.316 |
| Dyslipidemia | 1917 (56.3%) | 415 (57.5%) | 648 (49.8%) | 854 (61.8%) | <0.001 |
| Current smoker | 104 (3.1%) | 12 (1.7%) | 68 (5.2%) | 24 (1.7%) | <0.001 |
| Type of AF | 0.138 | ||||
| New or paroxysmal | 1148 (33.7%) | 224 (31.0%) | 464 (35.6%) | 460 (33.3%) | |
| Persistent | 645 (18.9%) | 131 (18.1%) | 253 (19.4%) | 261 (18.9%) | |
| Permanent | 1612 (47.3%) | 367 (50.8%) | 585 (44.9%) | 660 (47.8%) | |
| History of CAD | 547 (16.1%) | 238 (33.0%) | 155 (11.9%) | 154 (11.2%) | <0.001 |
| History of Ml or unstable angina | 207 (6.1%) | 116 (16.1%) | 52 (4.0%) | 39 (2.8%) | <0.001 |
| History of PCI | 253 (7.4%) | 147 (20.4%) | 69 (5.3%) | 37 (2.7%) | <0.001 |
| Drug eluting stent | 167 (4.9%) | 104 (14.4%) | 44 (3.4%) | 19 (1.4%) | <0.001 |
| History of CABG | 65 (1.9%) | 29 (4.0%) | 15 (1.2%) | 21 (1.5%) | <0.001 |
| History of peripheral artery disease | 44 (1.3%) | 14 (1.9%) | 10 (0.8%) | 20 (1.4%) | 0.066 |
| History of ischemic stroke or TIA | 592 (17.4%) | 136 (18.8%) | 182 (14.0%) | 274 (19.8%) | <0.001 |
| History of hyperthyroidism | 160 (4.7%) | 25 (3.5%) | 91 (7.0%) | 44 (3.2%) | <0.001 |
| History of hypothyroidism | 90 (2.6%) | 19 (2.6%) | 35 (2.7%) | 36 (2.6%) | 0.991 |
| Dementia | 29 (0.9%) | 8 (1.1%) | 6 (0.5%) | 15 (1.1%) | 0.148 |
| Cardioversion history | 87 (2.6%) | 8 (1.1%) | 57 (4.4%) | 22 (1.6%) | <0.001 |
| Catheter ablation of AF | 91 (2.7%) | 10 (1.4%) | 63 (4.8%) | 18 (1.3%) | <0.001 |
| AV node ablation pacemaker | 7 (0.2%) | 0 (0.0%) | 6 (0.5%) | 1 (0.1%) | 0.033 |
| GI bleeding history | 92 (2.7%) | 29 (4.0%) | 22 (1.7%) | 41 (3.0%) | 0.006 |
| Renal replacement therapy/kidney transplantation | 40 (1.2%) | 11 (1.5%) | 14 (1.1%) | 15 (1.1%) | 0.618 |
| Alcohol abuse | 140 (4.1%) | 15 (2.1%) | 99 (7.6%) | 26 (1.9%) | <0.001 |
| Rhythm control medications | 277 (8.1%) | 42 (5.8%) | 144 (11.1%) | 91 (6.6%) | <0.001 |
| NYHA class III/IV | 89 (2.6%) | 18 (2.5%) | 36 (2.8%) | 35 (2.5%) | 0.908 |
| Pacemaker | 269 (7.9%) | 107 (14.8%) | 53 (4.1%) | 109 (7.9%) | <0.001 |
| BiV | 1 (0.0%) | 0 (0.0%) | 0 (0.0%) | 1 (0.1%) | 0.480 |
| ICD | 55 (1.6%) | 16 (2.2%) | 15 (1.2%) | 24 (1.7%) | 0.171 |
| BiV/ICD | 16 (0.5%) | 8 (1.1%) | 7 (0.5%) | 1 (0.1%) | 0.004 |
| BMI (kg/m2) | 25.2 ± 4.7 | 23.5 ± 4.4 | 26.4 ± 4.9 | 24.9 ± 4.4 | <0.001 |
| Diastolic blood pressure (mmHg) | 128.5 ± 18.4 | 117 ± 15.5 | 126.1 ± 17.7 | 136.7 ± 16.6 | <0.001 |
| Systolic blood pressure (mmHg) | 75.2 ± 12.7 | 65.0 ± 9.8 | 76.9 ± 12.3 | 79.0 ± 11.5 | <0.001 |
| Heart rate (bpm) | 77.4 ± 16.2 | 70.9 ± 11.6 | 77.6 ± 15.4 | 80.5 ± 17.9 | <0.001 |
| Electrocardiographic evidence of LVH | 696 (20.4%) | 164 (22.7%) | 291 (22.4%) | 241 (17.5%) | 0.002 |
| Intraventricular conduction: non-specific IVCD | 145 (4.3%) | 49 (6.8%) | 32 (2.5%) | 64 (4.6%) | <0.001 |
| Intraventricular conduction: LBBB | 110 (3.2%) | 46 (6.4%) | 30 (2.3%) | 34 (2.5%) | <0.001 |
| eGFR | 62.1 ± 15.7 | 48.7 ± 4 | 81.2 ± 3.4 | 51 ± 5.6 | <0.001 |
| Hematocrit (%) | 39.6 ± 3.9 | 37.8 ± 4.1 | 41.3 ± 3.6 | 39 ± 3.5 | <0.001 |
| Anemia | 1293 (38.0%) | 448 (62.0%) | 205 (15.7%) | 640 (46.3%) | <0.001 |
| Liver function | 78 (2.3%) | 19 (2.6%) | 33 (2.5%) | 26 (1.9%) | 0.417 |
| LVEF (%) | 60.2 ± 13.7 | 59.6 ± 14.4 | 57.7 ± 14.6 | 62.8 ± 11.9 | <0.001 |
| LA dimension (mm) | 43.9 ± 8.3 | 43.2 ± 8.9 | 44.0 ± 7.5 | 44.0 ± 8.6 | 0.075 |
AF, atrial fibrillation; CAD, coronary artery disease; MI, myocardial infarction; PCI, percutaneous coronary intervention; TIA, transient ischemic attack; AV, atrioventricular; GI, gastrointestinal; NYHA, New York Heart Association; BiV, biventricular pacemaker; ICD, internal cardioverter defibrillator; BMI, body mass index; LVH, left ventricular hypertrophy; IVCD, intraventricular conduction delay; LBBB, left bundle branch block; eGFR, estimated glomerular filtration rate; LVEF, left ventricular ejection fraction; LA, left atrial.
Baseline characteristics of study population
| Variables | All (n = 3405) | Cluster 1 (n = 722) | Cluster 2 (n = 1302) | Cluster 3 (n = 1381) | P value |
|---|---|---|---|---|---|
| Age (years) | 67.8 ± 11.3 | 75.6 ± 6.6 | 56.5 ± 7.2 | 74.3 ± 6.8 | <0.001 |
| Female gender | 1424 (41.8%) | 325 (45.0%) | 449 (34.5%) | 650 (47.1%) | <0.001 |
| Hypertension | 2330 (68.4%) | 475 (65.8%) | 742 (57.0%) | 1113 (80.6%) | <0.001 |
| Diabetes | 839 (24.6%) | 172 (23.8%) | 308 (23.7%) | 359 (26.0%) | 0.316 |
| Dyslipidemia | 1917 (56.3%) | 415 (57.5%) | 648 (49.8%) | 854 (61.8%) | <0.001 |
| Current smoker | 104 (3.1%) | 12 (1.7%) | 68 (5.2%) | 24 (1.7%) | <0.001 |
| Type of AF | 0.138 | ||||
| New or paroxysmal | 1148 (33.7%) | 224 (31.0%) | 464 (35.6%) | 460 (33.3%) | |
| Persistent | 645 (18.9%) | 131 (18.1%) | 253 (19.4%) | 261 (18.9%) | |
| Permanent | 1612 (47.3%) | 367 (50.8%) | 585 (44.9%) | 660 (47.8%) | |
| History of CAD | 547 (16.1%) | 238 (33.0%) | 155 (11.9%) | 154 (11.2%) | <0.001 |
| History of Ml or unstable angina | 207 (6.1%) | 116 (16.1%) | 52 (4.0%) | 39 (2.8%) | <0.001 |
| History of PCI | 253 (7.4%) | 147 (20.4%) | 69 (5.3%) | 37 (2.7%) | <0.001 |
| Drug eluting stent | 167 (4.9%) | 104 (14.4%) | 44 (3.4%) | 19 (1.4%) | <0.001 |
| History of CABG | 65 (1.9%) | 29 (4.0%) | 15 (1.2%) | 21 (1.5%) | <0.001 |
| History of peripheral artery disease | 44 (1.3%) | 14 (1.9%) | 10 (0.8%) | 20 (1.4%) | 0.066 |
| History of ischemic stroke or TIA | 592 (17.4%) | 136 (18.8%) | 182 (14.0%) | 274 (19.8%) | <0.001 |
| History of hyperthyroidism | 160 (4.7%) | 25 (3.5%) | 91 (7.0%) | 44 (3.2%) | <0.001 |
| History of hypothyroidism | 90 (2.6%) | 19 (2.6%) | 35 (2.7%) | 36 (2.6%) | 0.991 |
| Dementia | 29 (0.9%) | 8 (1.1%) | 6 (0.5%) | 15 (1.1%) | 0.148 |
| Cardioversion history | 87 (2.6%) | 8 (1.1%) | 57 (4.4%) | 22 (1.6%) | <0.001 |
| Catheter ablation of AF | 91 (2.7%) | 10 (1.4%) | 63 (4.8%) | 18 (1.3%) | <0.001 |
| AV node ablation pacemaker | 7 (0.2%) | 0 (0.0%) | 6 (0.5%) | 1 (0.1%) | 0.033 |
| GI bleeding history | 92 (2.7%) | 29 (4.0%) | 22 (1.7%) | 41 (3.0%) | 0.006 |
| Renal replacement therapy/kidney transplantation | 40 (1.2%) | 11 (1.5%) | 14 (1.1%) | 15 (1.1%) | 0.618 |
| Alcohol abuse | 140 (4.1%) | 15 (2.1%) | 99 (7.6%) | 26 (1.9%) | <0.001 |
| Rhythm control medications | 277 (8.1%) | 42 (5.8%) | 144 (11.1%) | 91 (6.6%) | <0.001 |
| NYHA class III/IV | 89 (2.6%) | 18 (2.5%) | 36 (2.8%) | 35 (2.5%) | 0.908 |
| Pacemaker | 269 (7.9%) | 107 (14.8%) | 53 (4.1%) | 109 (7.9%) | <0.001 |
| BiV | 1 (0.0%) | 0 (0.0%) | 0 (0.0%) | 1 (0.1%) | 0.480 |
| ICD | 55 (1.6%) | 16 (2.2%) | 15 (1.2%) | 24 (1.7%) | 0.171 |
| BiV/ICD | 16 (0.5%) | 8 (1.1%) | 7 (0.5%) | 1 (0.1%) | 0.004 |
| BMI (kg/m2) | 25.2 ± 4.7 | 23.5 ± 4.4 | 26.4 ± 4.9 | 24.9 ± 4.4 | <0.001 |
| Diastolic blood pressure (mmHg) | 128.5 ± 18.4 | 117 ± 15.5 | 126.1 ± 17.7 | 136.7 ± 16.6 | <0.001 |
| Systolic blood pressure (mmHg) | 75.2 ± 12.7 | 65.0 ± 9.8 | 76.9 ± 12.3 | 79.0 ± 11.5 | <0.001 |
| Heart rate (bpm) | 77.4 ± 16.2 | 70.9 ± 11.6 | 77.6 ± 15.4 | 80.5 ± 17.9 | <0.001 |
| Electrocardiographic evidence of LVH | 696 (20.4%) | 164 (22.7%) | 291 (22.4%) | 241 (17.5%) | 0.002 |
| Intraventricular conduction: non-specific IVCD | 145 (4.3%) | 49 (6.8%) | 32 (2.5%) | 64 (4.6%) | <0.001 |
| Intraventricular conduction: LBBB | 110 (3.2%) | 46 (6.4%) | 30 (2.3%) | 34 (2.5%) | <0.001 |
| eGFR | 62.1 ± 15.7 | 48.7 ± 4 | 81.2 ± 3.4 | 51 ± 5.6 | <0.001 |
| Hematocrit (%) | 39.6 ± 3.9 | 37.8 ± 4.1 | 41.3 ± 3.6 | 39 ± 3.5 | <0.001 |
| Anemia | 1293 (38.0%) | 448 (62.0%) | 205 (15.7%) | 640 (46.3%) | <0.001 |
| Liver function | 78 (2.3%) | 19 (2.6%) | 33 (2.5%) | 26 (1.9%) | 0.417 |
| LVEF (%) | 60.2 ± 13.7 | 59.6 ± 14.4 | 57.7 ± 14.6 | 62.8 ± 11.9 | <0.001 |
| LA dimension (mm) | 43.9 ± 8.3 | 43.2 ± 8.9 | 44.0 ± 7.5 | 44.0 ± 8.6 | 0.075 |
| Variables | All (n = 3405) | Cluster 1 (n = 722) | Cluster 2 (n = 1302) | Cluster 3 (n = 1381) | P value |
|---|---|---|---|---|---|
| Age (years) | 67.8 ± 11.3 | 75.6 ± 6.6 | 56.5 ± 7.2 | 74.3 ± 6.8 | <0.001 |
| Female gender | 1424 (41.8%) | 325 (45.0%) | 449 (34.5%) | 650 (47.1%) | <0.001 |
| Hypertension | 2330 (68.4%) | 475 (65.8%) | 742 (57.0%) | 1113 (80.6%) | <0.001 |
| Diabetes | 839 (24.6%) | 172 (23.8%) | 308 (23.7%) | 359 (26.0%) | 0.316 |
| Dyslipidemia | 1917 (56.3%) | 415 (57.5%) | 648 (49.8%) | 854 (61.8%) | <0.001 |
| Current smoker | 104 (3.1%) | 12 (1.7%) | 68 (5.2%) | 24 (1.7%) | <0.001 |
| Type of AF | 0.138 | ||||
| New or paroxysmal | 1148 (33.7%) | 224 (31.0%) | 464 (35.6%) | 460 (33.3%) | |
| Persistent | 645 (18.9%) | 131 (18.1%) | 253 (19.4%) | 261 (18.9%) | |
| Permanent | 1612 (47.3%) | 367 (50.8%) | 585 (44.9%) | 660 (47.8%) | |
| History of CAD | 547 (16.1%) | 238 (33.0%) | 155 (11.9%) | 154 (11.2%) | <0.001 |
| History of Ml or unstable angina | 207 (6.1%) | 116 (16.1%) | 52 (4.0%) | 39 (2.8%) | <0.001 |
| History of PCI | 253 (7.4%) | 147 (20.4%) | 69 (5.3%) | 37 (2.7%) | <0.001 |
| Drug eluting stent | 167 (4.9%) | 104 (14.4%) | 44 (3.4%) | 19 (1.4%) | <0.001 |
| History of CABG | 65 (1.9%) | 29 (4.0%) | 15 (1.2%) | 21 (1.5%) | <0.001 |
| History of peripheral artery disease | 44 (1.3%) | 14 (1.9%) | 10 (0.8%) | 20 (1.4%) | 0.066 |
| History of ischemic stroke or TIA | 592 (17.4%) | 136 (18.8%) | 182 (14.0%) | 274 (19.8%) | <0.001 |
| History of hyperthyroidism | 160 (4.7%) | 25 (3.5%) | 91 (7.0%) | 44 (3.2%) | <0.001 |
| History of hypothyroidism | 90 (2.6%) | 19 (2.6%) | 35 (2.7%) | 36 (2.6%) | 0.991 |
| Dementia | 29 (0.9%) | 8 (1.1%) | 6 (0.5%) | 15 (1.1%) | 0.148 |
| Cardioversion history | 87 (2.6%) | 8 (1.1%) | 57 (4.4%) | 22 (1.6%) | <0.001 |
| Catheter ablation of AF | 91 (2.7%) | 10 (1.4%) | 63 (4.8%) | 18 (1.3%) | <0.001 |
| AV node ablation pacemaker | 7 (0.2%) | 0 (0.0%) | 6 (0.5%) | 1 (0.1%) | 0.033 |
| GI bleeding history | 92 (2.7%) | 29 (4.0%) | 22 (1.7%) | 41 (3.0%) | 0.006 |
| Renal replacement therapy/kidney transplantation | 40 (1.2%) | 11 (1.5%) | 14 (1.1%) | 15 (1.1%) | 0.618 |
| Alcohol abuse | 140 (4.1%) | 15 (2.1%) | 99 (7.6%) | 26 (1.9%) | <0.001 |
| Rhythm control medications | 277 (8.1%) | 42 (5.8%) | 144 (11.1%) | 91 (6.6%) | <0.001 |
| NYHA class III/IV | 89 (2.6%) | 18 (2.5%) | 36 (2.8%) | 35 (2.5%) | 0.908 |
| Pacemaker | 269 (7.9%) | 107 (14.8%) | 53 (4.1%) | 109 (7.9%) | <0.001 |
| BiV | 1 (0.0%) | 0 (0.0%) | 0 (0.0%) | 1 (0.1%) | 0.480 |
| ICD | 55 (1.6%) | 16 (2.2%) | 15 (1.2%) | 24 (1.7%) | 0.171 |
| BiV/ICD | 16 (0.5%) | 8 (1.1%) | 7 (0.5%) | 1 (0.1%) | 0.004 |
| BMI (kg/m2) | 25.2 ± 4.7 | 23.5 ± 4.4 | 26.4 ± 4.9 | 24.9 ± 4.4 | <0.001 |
| Diastolic blood pressure (mmHg) | 128.5 ± 18.4 | 117 ± 15.5 | 126.1 ± 17.7 | 136.7 ± 16.6 | <0.001 |
| Systolic blood pressure (mmHg) | 75.2 ± 12.7 | 65.0 ± 9.8 | 76.9 ± 12.3 | 79.0 ± 11.5 | <0.001 |
| Heart rate (bpm) | 77.4 ± 16.2 | 70.9 ± 11.6 | 77.6 ± 15.4 | 80.5 ± 17.9 | <0.001 |
| Electrocardiographic evidence of LVH | 696 (20.4%) | 164 (22.7%) | 291 (22.4%) | 241 (17.5%) | 0.002 |
| Intraventricular conduction: non-specific IVCD | 145 (4.3%) | 49 (6.8%) | 32 (2.5%) | 64 (4.6%) | <0.001 |
| Intraventricular conduction: LBBB | 110 (3.2%) | 46 (6.4%) | 30 (2.3%) | 34 (2.5%) | <0.001 |
| eGFR | 62.1 ± 15.7 | 48.7 ± 4 | 81.2 ± 3.4 | 51 ± 5.6 | <0.001 |
| Hematocrit (%) | 39.6 ± 3.9 | 37.8 ± 4.1 | 41.3 ± 3.6 | 39 ± 3.5 | <0.001 |
| Anemia | 1293 (38.0%) | 448 (62.0%) | 205 (15.7%) | 640 (46.3%) | <0.001 |
| Liver function | 78 (2.3%) | 19 (2.6%) | 33 (2.5%) | 26 (1.9%) | 0.417 |
| LVEF (%) | 60.2 ± 13.7 | 59.6 ± 14.4 | 57.7 ± 14.6 | 62.8 ± 11.9 | <0.001 |
| LA dimension (mm) | 43.9 ± 8.3 | 43.2 ± 8.9 | 44.0 ± 7.5 | 44.0 ± 8.6 | 0.075 |
AF, atrial fibrillation; CAD, coronary artery disease; MI, myocardial infarction; PCI, percutaneous coronary intervention; TIA, transient ischemic attack; AV, atrioventricular; GI, gastrointestinal; NYHA, New York Heart Association; BiV, biventricular pacemaker; ICD, internal cardioverter defibrillator; BMI, body mass index; LVH, left ventricular hypertrophy; IVCD, intraventricular conduction delay; LBBB, left bundle branch block; eGFR, estimated glomerular filtration rate; LVEF, left ventricular ejection fraction; LA, left atrial.
Outcomes
The mean follow-up duration was 31.8 ± 8.7 months. During follow-up, the primary composite outcomes, all-cause death, SSE, MI, and HF occurred in 645 (18.9%), 380 (11.2%), 134 (3.9%), 41 (1.2%) and 247 (7.3%) patients, respectively. The incidence rates of these outcomes were 7.53 (6.96–8.13), 4.21 (3.81–4.65), 1.51 (1.26–1.78), 0.46 (0.33–0.62) and 2.84 (2.49–3.21), respectively. SSE rate and incidence rate (95% CI) according to CHA2DS2-VASc score for patients with OAC (n = 2568) and no OAC (n = 837) are shown in Supplementary Table S2.
The incidence rates of clinical outcomes among each cluster are shown in Table 2, and differences between cluster groups are shown as histogram bar graphs in Figure 2. Cluster 1 had the highest risks and Cluster 2 was the lowest risk group.
Bar graphs of incidence rate of composite and individual clinical outcomes according to clusters (SSE, ischemic stroke/systemic embolism; MI, myocardial infarction; HF, heart failure).
Incidence rate of clinical outcomes according to clusters
| Number of patients | Number of events | 100 person-years | Rate per 100 person-years (95% CI) | |
|---|---|---|---|---|
| Composite outcomesa | ||||
| All patients | 3405 | 645 | 85.67 | 7.53 (6.96–8.13) |
| Cluster 1 | 722 | 202 | 16.85 | 11.99 (10.39–13.76) |
| Cluster 2 | 1302 | 148 | 34.35 | 4.31 (3.64–5.06) |
| Cluster 3 | 1381 | 295 | 34.47 | 8.56 (7.61–9.59) |
| All-cause death | ||||
| All patients | 3405 | 380 | 90.27 | 4.21 (3.81–4.65) |
| Cluster 1 | 722 | 118 | 18.33 | 6.44 (5.33–7.71) |
| Cluster 2 | 1302 | 80 | 35.47 | 2.26 (1.79–2.81) |
| Cluster 3 | 1381 | 182 | 36.47 | 4.99 (4.29–5.77) |
| SSE | ||||
| All patients | 3405 | 134 | 88.99 | 1.51 (1.26–1.78) |
| Cluster 1 | 722 | 33 | 18.07 | 1.83 (1.26–2.56) |
| Cluster 2 | 1302 | 27 | 35.18 | 0.77 (0.51–1.12) |
| Cluster 3 | 1381 | 74 | 35.74 | 2.07 (1.63–2.59) |
| Myocardial infarction | ||||
| All patients | 3405 | 41 | 89.94 | 0.46 (0.33–0.62) |
| Cluster 1 | 722 | 13 | 18.20 | 0.71 (0.38–1.22) |
| Cluster 2 | 1302 | 8 | 35.43 | 0.23 (0.09–0.44) |
| Cluster 3 | 1381 | 20 | 36.31 | 0.55 (0.34–0.85) |
| Heart failure | ||||
| All patients | 3405 | 247 | 87.08 | 2.84 (2.49–3.21) |
| Cluster 1 | 722 | 88 | 17.13 | 5.14 (4.12–6.33) |
| Cluster 2 | 1302 | 68 | 34.66 | 1.96 (1.52–2.49) |
| Cluster 3 | 1381 | 91 | 35.29 | 2.58 (2.08–3.17) |
| Number of patients | Number of events | 100 person-years | Rate per 100 person-years (95% CI) | |
|---|---|---|---|---|
| Composite outcomesa | ||||
| All patients | 3405 | 645 | 85.67 | 7.53 (6.96–8.13) |
| Cluster 1 | 722 | 202 | 16.85 | 11.99 (10.39–13.76) |
| Cluster 2 | 1302 | 148 | 34.35 | 4.31 (3.64–5.06) |
| Cluster 3 | 1381 | 295 | 34.47 | 8.56 (7.61–9.59) |
| All-cause death | ||||
| All patients | 3405 | 380 | 90.27 | 4.21 (3.81–4.65) |
| Cluster 1 | 722 | 118 | 18.33 | 6.44 (5.33–7.71) |
| Cluster 2 | 1302 | 80 | 35.47 | 2.26 (1.79–2.81) |
| Cluster 3 | 1381 | 182 | 36.47 | 4.99 (4.29–5.77) |
| SSE | ||||
| All patients | 3405 | 134 | 88.99 | 1.51 (1.26–1.78) |
| Cluster 1 | 722 | 33 | 18.07 | 1.83 (1.26–2.56) |
| Cluster 2 | 1302 | 27 | 35.18 | 0.77 (0.51–1.12) |
| Cluster 3 | 1381 | 74 | 35.74 | 2.07 (1.63–2.59) |
| Myocardial infarction | ||||
| All patients | 3405 | 41 | 89.94 | 0.46 (0.33–0.62) |
| Cluster 1 | 722 | 13 | 18.20 | 0.71 (0.38–1.22) |
| Cluster 2 | 1302 | 8 | 35.43 | 0.23 (0.09–0.44) |
| Cluster 3 | 1381 | 20 | 36.31 | 0.55 (0.34–0.85) |
| Heart failure | ||||
| All patients | 3405 | 247 | 87.08 | 2.84 (2.49–3.21) |
| Cluster 1 | 722 | 88 | 17.13 | 5.14 (4.12–6.33) |
| Cluster 2 | 1302 | 68 | 34.66 | 1.96 (1.52–2.49) |
| Cluster 3 | 1381 | 91 | 35.29 | 2.58 (2.08–3.17) |
SSE, ischemic stroke/systemic embolism.
Composite outcomes = all-cause death or SSE or myocardial infarction or heart failure.
Incidence rate of clinical outcomes according to clusters
| Number of patients | Number of events | 100 person-years | Rate per 100 person-years (95% CI) | |
|---|---|---|---|---|
| Composite outcomesa | ||||
| All patients | 3405 | 645 | 85.67 | 7.53 (6.96–8.13) |
| Cluster 1 | 722 | 202 | 16.85 | 11.99 (10.39–13.76) |
| Cluster 2 | 1302 | 148 | 34.35 | 4.31 (3.64–5.06) |
| Cluster 3 | 1381 | 295 | 34.47 | 8.56 (7.61–9.59) |
| All-cause death | ||||
| All patients | 3405 | 380 | 90.27 | 4.21 (3.81–4.65) |
| Cluster 1 | 722 | 118 | 18.33 | 6.44 (5.33–7.71) |
| Cluster 2 | 1302 | 80 | 35.47 | 2.26 (1.79–2.81) |
| Cluster 3 | 1381 | 182 | 36.47 | 4.99 (4.29–5.77) |
| SSE | ||||
| All patients | 3405 | 134 | 88.99 | 1.51 (1.26–1.78) |
| Cluster 1 | 722 | 33 | 18.07 | 1.83 (1.26–2.56) |
| Cluster 2 | 1302 | 27 | 35.18 | 0.77 (0.51–1.12) |
| Cluster 3 | 1381 | 74 | 35.74 | 2.07 (1.63–2.59) |
| Myocardial infarction | ||||
| All patients | 3405 | 41 | 89.94 | 0.46 (0.33–0.62) |
| Cluster 1 | 722 | 13 | 18.20 | 0.71 (0.38–1.22) |
| Cluster 2 | 1302 | 8 | 35.43 | 0.23 (0.09–0.44) |
| Cluster 3 | 1381 | 20 | 36.31 | 0.55 (0.34–0.85) |
| Heart failure | ||||
| All patients | 3405 | 247 | 87.08 | 2.84 (2.49–3.21) |
| Cluster 1 | 722 | 88 | 17.13 | 5.14 (4.12–6.33) |
| Cluster 2 | 1302 | 68 | 34.66 | 1.96 (1.52–2.49) |
| Cluster 3 | 1381 | 91 | 35.29 | 2.58 (2.08–3.17) |
| Number of patients | Number of events | 100 person-years | Rate per 100 person-years (95% CI) | |
|---|---|---|---|---|
| Composite outcomesa | ||||
| All patients | 3405 | 645 | 85.67 | 7.53 (6.96–8.13) |
| Cluster 1 | 722 | 202 | 16.85 | 11.99 (10.39–13.76) |
| Cluster 2 | 1302 | 148 | 34.35 | 4.31 (3.64–5.06) |
| Cluster 3 | 1381 | 295 | 34.47 | 8.56 (7.61–9.59) |
| All-cause death | ||||
| All patients | 3405 | 380 | 90.27 | 4.21 (3.81–4.65) |
| Cluster 1 | 722 | 118 | 18.33 | 6.44 (5.33–7.71) |
| Cluster 2 | 1302 | 80 | 35.47 | 2.26 (1.79–2.81) |
| Cluster 3 | 1381 | 182 | 36.47 | 4.99 (4.29–5.77) |
| SSE | ||||
| All patients | 3405 | 134 | 88.99 | 1.51 (1.26–1.78) |
| Cluster 1 | 722 | 33 | 18.07 | 1.83 (1.26–2.56) |
| Cluster 2 | 1302 | 27 | 35.18 | 0.77 (0.51–1.12) |
| Cluster 3 | 1381 | 74 | 35.74 | 2.07 (1.63–2.59) |
| Myocardial infarction | ||||
| All patients | 3405 | 41 | 89.94 | 0.46 (0.33–0.62) |
| Cluster 1 | 722 | 13 | 18.20 | 0.71 (0.38–1.22) |
| Cluster 2 | 1302 | 8 | 35.43 | 0.23 (0.09–0.44) |
| Cluster 3 | 1381 | 20 | 36.31 | 0.55 (0.34–0.85) |
| Heart failure | ||||
| All patients | 3405 | 247 | 87.08 | 2.84 (2.49–3.21) |
| Cluster 1 | 722 | 88 | 17.13 | 5.14 (4.12–6.33) |
| Cluster 2 | 1302 | 68 | 34.66 | 1.96 (1.52–2.49) |
| Cluster 3 | 1381 | 91 | 35.29 | 2.58 (2.08–3.17) |
SSE, ischemic stroke/systemic embolism.
Composite outcomes = all-cause death or SSE or myocardial infarction or heart failure.
Cox proportional hazard model and survival analysis
Survival analysis using time-to-event data confirmed that Cluster 1 was the highest risk group and Cluster 2 was the lowest risk group using log-rank tests, P < 0001 (Figure 3). Cox proportional hazard model analysis demonstrated that for the primary composite outcome, Cluster 1 had an HR 2.78 (95% CI 2.25–3.43) and Cluster 3 HR of 1.99 (1.63-2.42) when compared to Cluster 2. Similar results were shown for the individual outcomes (Figure 4). The adjusted analysis was not performed since every baseline factor was used to generate the clusters.
Survival graph of composite outcomes according to clusters.
Forest plot from results of the Cox proportional hazard model of composite and individual outcomes according to clusters (SSE, ischemic stroke/systemic embolism; MI, myocardial infarction; HF, heart failure).
Clinical outcomes according to compliance to ABC pathway
Cox proportional hazard model analysis demonstrated that compliance to the ABC pathway was associated with a reduction in composite outcomes as well as individual outcomes. ABC pathway compliance for each cluster group showed that the greatest benefit was seen for AF patients in Cluster 1 and Cluster 3, while a non-statistical trend for benefit was seen for Cluster 2 (the lowest risk group). The forest plots are shown in Figure 5.
Forest plot of composite and individual outcomes according to ABC pathway compliance of each cluster (SSE, ischemic stroke/systemic embolism; MI, myocardial infarction; HF, heart failure).
Sensitivity analysis for the relation of ABC adherence on clinical outcomes using ABC adherence with the achievement of TTR ≥65% for those with warfarin and blood pressure <140/90 mmHg for those with hypertension showed that ABC adherence is associated with better outcomes for all clusters (Supplementary Figure S2).
Discussion
The phenotypic clustering process on this prospective multicentre nationwide AF registry conducted in Thailand demonstrates that the phenotypic classification was useful for risk stratification and guiding therapy. Second, management according to the ABC pathway was associated with a beneficial effect in reducing clinical outcomes in AF patients, especially amongst who belonged to higher-risk cluster groups.
This is the first analysis of phenotypic clusters in Asian patients, which also shows the impact of compliance with the ABC pathway on outcomes. The initial focus for risk reduction during the management of AF patients was focused on stroke reduction. However, the high residual risks of adverse outcomes including stroke and death remained even amongst OAC-treated patients.15 Hence, recent guidelines have recommended a more holistic or integrated care management approach that includes the management of symptoms, cardiovascular risk factors and comorbidities, including lifestyle changes.4,5,13
Our study shows that phenotypic classification using the clustering process may provide an additional value for the risk stratification and prognosis since it integrates all available information to classify patients into different clinical risk groups.7 The clustering analysis of our studies used 46 variables as described earlier. This is in contrast to approaching risk factors in a binary (yes/no) manner, given the close associations of many risk factors with each other, thus defining a phenotypic ‘cluster’ of comorbidities. Indeed, human pathophysiology may be too complex for hypothetical reasoning even with the aid of linear or non-linear statistical methods. Clustering algorithms use data input and distribution patterns to form cluster groups of certain diseases or populations that share similar and distinct properties. When the clustering process is properly applied, specific disease phenotype clustering can help demonstrate obscured associations that help clinicians understand disease complexity, and guide appropriate management.7,16 For example, phenotypic classification of HF patients not only helps in risk prediction but also guiding management such as that seen in the TOPCAT study.10
Even in phenotypic groups with similar risk, the classification helped to identify the HF patient groups that responded well to spironolactone.10 Similar to our study in AF patients, phenotype clustering not only helped in risk prediction but also identified those who would benefit most from ABC pathway management. Nonetheless, phenotypic clusters may partly vary according to the characteristics of the patient population being studied, necessitating more evidence for different cohorts and patient groups. For example, the ORBIT-AF registry revealed that among the four cluster subgroups, patients with atherosclerotic comorbidities had the worst clinical outcomes during the two-year follow-up.11 The FUSHIMI-AF investigators identified six clusters in Japanese AF patients and demonstrated that patients who were very elderly, with prior stroke and atherosclerotic comorbidities were the highest risk groups.12 Identification of the high-risk subjects with AF should help clinicians to intensify management and ABC pathway-based integrated care. The results of our sensitivity analysis showed that patients with ABC adherence with the achievement of good TTR for warfarin and good blood pressure control for hypertension had better clinical outcomes in all clusters even in the low-risk cluster. These results emphasize the benefit of the adherence to ABC pathway in patients with AF.
Limitations
There were some limitations in this study. First, our study enrolled patients mainly from large-sized hospitals, and may not be generalized to the general population. Second, the variables used in the clustering process of different AF cohorts may provide different results, hence comparing results between different studies needs consideration of the variables and methods used for clustering.
Conclusion
Phenotypic classification by cluster analysis can demonstrate clusters of AF patients with similar phenotypes which help in risk stratification and prognostication. Compliance with the ABC pathway was associated with improved clinical outcomes.
Supplementary material
Supplementary material is available at QJMED online.
Acknowledgements
The authors gratefully acknowledge Ahthit Yindeengam and Poom Sairat for data management.
Funding
This study was funded by the Health Systems Research Institute (HSRI) (59-053), and the Heart Association of Thailand under the Royal Patronage of H.M. The King. Neither of the two funding sources influenced any aspect of this study or the authors’ decision to submit this manuscript for publication.
Conflict of interest
G.Y.H.L.: consultant and speaker for BMS/Pfizer, Boehringer Ingelheim, Anthos and Daiichi-Sankyo. No fees are directly received personally. G.Y.H.L. is co-principal investigator of the AFFIRMO project on multimorbidity in AF, which has received funding from the European Union’s Horizon 2020 research and innovation program under grant agreement no. 899871. Other authors hereby declare no personal or professional conflicts of interest relating to any aspect of this particular study.
Author contributions
Rungroj Krittayaphong (Conceptualization [equal], Data curation [equal], Methodology [equal], Project administration [equal], Resources [equal], Supervision [equal], Writing—original draft [equal], Writing—review & editing [equal]), Sukrit Treewaree (Formal analysis [equal], Investigation [equal], Writing—review & editing [supporting]), Wattana Wongtheptien (Investigation [equal], Writing—review & editing [equal]) and Pontawee Kaewkumdee (Investigation [equal], Methodology [equal], Writing—review & editing [equal])
Data availability
The dataset that was used to support the conclusion of this study is included within the manuscript. Any other additional data will be made available upon request to Rungroj Krittayaphong ([email protected]).
