Study population
This study included 9,895 participants, aged 40–69 years, from the Ansung–Ansan cohort of the Korean Genome Epidemiology Study, conducted by the Korea Disease Control and Prevention Agency. This study investigated the genetic and environmental factors contributing to the prevalence of metabolic and cardiovascular diseases. Individuals residing in rural (Ansung) and urban (Ansan) communities were enrolled between June 2001 and January 2003. The detailed protocols have been described in previous publications23,24.
A total of 10,030 eligible individuals who had resided in Ansung (n = 5,018) or Ansan (n = 5,012) for at least 6 months were enrolled in the study. Participants diagnosed with AF at baseline (n = 74) and those without SRH records (n = 61) were excluded, resulting in a final population of 9,895 participants.
A myriad of comprehensive health examinations, detailed on-site interviews, and meticulous laboratory tests were conducted during the baseline visit to a tertiary hospital. Six serial reassessments, following the entire cohort protocol, were performed through scheduled revisits every other year until 2014. All participants voluntarily enrolled in the study and provided written informed consent at the baseline assessment and each follow-up visit. The study adhered to the principles of the Declaration of Helsinki and was approved by the Korean National Research Institute of Health and the Institutional Review Board (IRB) of Hanyang University Medical Center (IRB no. HYUH 2017-12-033).
Assessment of lifestyle and medical history, physical examination, and laboratory tests
Well-trained investigators conducted comprehensive on-site interviews, collected critical lifestyle and clinical data, and performed physical examinations at tertiary hospitals during each visit. A structured questionnaire was used to obtain data on smoking, alcohol intake, education level, and specific medical conditions, such as HTN, DM, dyslipidaemia, cerebrovascular disease, CAD, and heart failure. Higher education was defined as obtaining a college degree or higher. The type and duration of physical activity were assessed using detailed questionnaires and quantified using estimated daily metabolic equivalent task scores. Blood pressure was measured by trained examiners using a mercury sphygmomanometer positioned at the level of the heart. Measurements were taken at least twice with the participant in a sitting position, and the results were averaged. If a blood pressure difference of ≥ 5 mmHg was observed between the two measurements, a third measurement was taken, and the last two measurements were averaged. WC was measured at the midpoint between the lowest rib and iliac crest at the end of expiration in a standing position.
Blood samples were collected after overnight fasting and analysed to determine the lipid profiles, haemoglobin A1c levels, white blood cell counts, and CRP levels using an automated analyser. Laboratory evaluations were performed in a single core clinical laboratory accredited and participating annually in inspections and surveys by the Korean Association of Quality Assurance for Clinical Laboratories. Blood concentrations of glucose, total cholesterol, high density lipoprotein (HDL)-cholesterol, and triglyceride were measured using the enzyme method (ADVIA 1650 and ADVIA 1800; Siemens Healthineers). Low-density lipoprotein (LDL)-cholesterol levels were calculated using the Friedewald formula25. CRP was measured using a turbidimetric assay method (ADVIA 1650 and ADVIA 1800; Siemens Healthineers). HTN was defined as a diagnosis of HTN made by a physician or the regular use of antihypertensive medications. DM was defined as a diagnosis of DM by a physician, regular use of antidiabetic medications, or a haemoglobin A1c level of ≥ 6.5%. Dyslipidaemia was defined as a diagnosis of dyslipidaemia by a physician, the regular use of statin without a history of cardiovascular disease or DM, or the presence of at least one of the following abnormal laboratory test results: total cholesterol level ≥ 240 mg/dL, triglyceride level ≥ 150 mg/dL, or HDL cholesterol level < 45 mg/dL.
Assessment of self-rated health
Participants were asked to evaluate their overall health by responding to the question, ‘How do you generally perceive your health?’ The responses were initially divided into five categories: ‘very poor’, ‘poor’, ‘fair’, ‘good’, and ‘very good’. Due to the relatively small numbers of participants who responded ‘very poor’ (n = 411) and ‘very good’ (n = 154), we reclassified the responses into three categories: ‘poor (very poor/poor)’ SRH (n = 3,380), ‘fair’ SRH (n = 3,521), and ‘good (good/very good)’ SRH (n = 2,994).
Standard 12-lead ECG and identification of AF
Standard 12-lead ECG (GE Marquette MAC 5000®, GE Marquette Inc., Milwaukee, WI, USA) was performed in all participants at baseline and during every revisit. All ECG tracings were recorded at a paper speed of 25 mm/s and an amplitude of 0.1 mV/mm. The results were interpreted by a cardiologist and coded according to the Minnesota code classification system. AF was identified either through the presence of AF on a 12-lead ECG or a self-reported history of AF using a questionnaire administered by a physician before the baseline visit or during follow-up visits. The Minnesota codes 8-3-1, 8-3-2, 8-3-3, and 8-3-4 were used to classify AF. Newly developed AF was defined as the first identification of AF between visits. The date of new AF development was defined as the date when AF was first detected on ECG or when it was diagnosed by a physician.
Statistical analysis
The baseline characteristics of the participants were compared between the groups. One-way analysis of variance was used for analysing continuous variables, such as BMI, WC, and LDL-cholesterol levels, while Pearson’s chi-square test was used for analysing categorical variables, including sex, comorbidities, and smoking history. Post-hoc analyses using Bonferroni correction were used to perform multiple comparisons. For continuous variables with a skewed distribution, comparisons between groups were performed using the Kruskal–Wallis test. The Shapiro–Wilk test was used to assess the normality of the distribution of continuous variables.
All variables had approximately 1% missing values. The missing values of individual variables ranged from 0 to 4.9%. Among the covariates in the multivariate regression models, the missing values of individual variables ranged from 0 to 1.04% (Supplementary Fig. S1). Rather than excluding cases with missing data, multiple imputations were performed using a bootstrap expectation-maximization algorithm26. Five possible imputed datasets were generated. The average value was used for continuous variables, while the most frequent value was adopted for categorical variables.
The associations between SRH and incident AF were evaluated using a CPH model adjusted for age, sex, residence, education, BMI, physical activity, comorbidities (including HTN, DM, heart failure, dyslipidaemia, MI, non-MI CAD, asthma, and chronic lung disease), smoking status, alcohol consumption, and laboratory data (including LDL cholesterol levels and CRP levels. The factors associated with the ‘poor’ SRH group were analysed using logistic regression analysis. To determine whether SRH provided an additional predictive value for incident AF when combined with conventional risk factors, we developed prediction models for new-onset AF with and without the SRH variable using multivariate CPH models. The goodness of fit of these prediction models was estimated using Harrell’s C-index and Akaike information criterion (AIC). Harrell’s C-indices were compared using the method proposed by Haibe–Kains et al.27. The linear predictors identified using multivariate CPH models were employed as risk predictors for the C-index estimation. A difference of > 10 between the two AIC values was considered significant.
To evaluate the impact of reclassifying SRH responses from five to three categories based on the survey results, we performed a sensitivity analysis using a Kaplan–Meier survival analysis and a multivariate CPH model with the original five-category SRH responses. Due to the small number of ‘very good’ responses, the ‘good’ response was used as a reference when calculating the coefficients in the multivariate CPH model.
All statistical analyses were performed using the statistical software R-4.3.2 (R Core Team, R Foundation for Statistical Computing, Vienna, Austria) and its packages ‘tableone’, ‘rms’, ‘Amelia’, ‘survival’, ‘BiocManager’, and ‘survcomp’ in RStudio-2023.12.1, Build 402 (RStudio Team, RStudio Inc., Boston, MA, USA). A P value of < 0.05 was considered significant.
link