Risk factor modication reduced incidence of atrial brillation in an 18-year prospective cohort study: a time-updated analysis

Although atrial brillation (AF) is an increasing health burden worldwide, strategies for AF prevention are lacking. This study aimed to identify modiable risk factors (MRF) for and estimate their impact on AF risk in the midlife general population. Methods We assessed 9,049 participants who were free of prevalent AF at baseline from the Korean Genome and Epidemiology Study. Cox models with time-varying assessment of risk factors were used to identify signicant MRF for incident AF. The MRF burden was dened as the proportion of times presented MRF during follow-up, based on the number of visits.


Introduction
Atrial brillation (AF), the most common cardiac arrhythmia, is a leading cause of mortality and ischemic stroke (1)(2). The prevalence of AF is expected to increase due to a growing burden of risk factors, such as an aging population, hypertension, obesity, diabetes mellitus and ischemic heart disease (3). AF also results in signi cant use of health care resources and causes a substantial economic burden, with AFrelated Medicare expenses being approximately $16 billion annually in the United States (4). Although catheter ablation has been shown to be effective in suppressing AF, it is invasive and has the potential for serious complications despite improved experience and advances in ablation technology (5, 6). Thus, a Page 3/16 focus on primary prevention is a critical component of strategies to control the growing burden of AF at a population level.
Many studies of incident AF have focused primarily on risk prediction, AF treatment and AF-related stroke prevention (6-11). However, there is limited information on preventive strategies to reduce the incidence of AF. Earlier studies on modi able and non-modi able risk factors focused mainly on baseline assessment, and the results may therefore have immortal time bias. For primordial prevention interventions, an appropriate level of control of modi able risk factors (MRF) is important.
In the present study, we aimed to identify the MRF for incident AF using a time-updated model, investigate the cumulative effects of the MRF burden on AF risk, and provide targets for prevention of AF in a longitudinal population-based cohort.

Baseline characteristics
In 9,049 participants without AF at baseline, 182 participants developed new-onset AF over a median follow-up of 13.1 years (range, 1.4 to 16.6 years). Participants with AF, when compared with people who did not develop AF, were more likely to be older, male, rural, to have CVD, higher WC, higher BP, and lower LTPA (Table 1).  (Table S1).
In time-updated multivariable models accounting for changes in risk factors after the baseline survey, three MRF (SBP ≥140 mmHg, obesity with central obesity, and inactivity) were each signi cantly associated with the development of incident AF ( or high levels of LTPA (> 0 min/weeks), was associated with a 42% higher risk of AF. At baseline, 81.2% of all participants had at least one MRF associated with AF risk, and the incidence of AF increased markedly with increasing numbers of MRF (Table S2). Additionally, assuming a causal relationship between MRF and AF, 28.7% (95% CI: 14.6-40.4%) of incident AF in the study population was attributable to these three MRF, with inactivity being the greatest contributing risk factor. (Table 3). Optimal levels of modi able risk factor control for risk reduction of AF To assess the joint impact and nd optimal levels of these MRF to impact on the incidence of AF, we used multivariable adjusted models with time-updated assessment of risk factors after baseline, as well as baseline models (Fig. 1 , with a graded reduction in risk across progressively more optimal risk factor levels. Patients with optimal levels of all three modi able risk factors (SBP ≥140 mmHg, obesity with central obesity, inactivity) had a 71% lower risk of incident AF compared with participants with the least favourable risk factor pro le (three MRF). These results were similar to those using the baseline model, where participants with ≤1 MRF had a lower risk of incident AF compared with participants with three MRF.
Impact of modi able risk factor burden on the risk of AF Figure 2 depicts the risk of AF associated with continuous measures of MRF burden during follow-up, using restricted cubic spline analysis (with a burden of 100% as reference). The proportion of participants with more than two MRF (burden of more than two MRF) was signi cantly associated with the risk of AF

Discussion
This study had three principal ndings. Firstly, in a time-updated model, three MRF (SBP ≥140 mmHg, obesity with central obesity, and inactivity) were signi cant in the midlife general population. Secondly, participants who maintained or achieved an optimal risk factor pro le had a signi cantly reduced risk of AF. These ndings highlight the potential population and individual level impact of risk factor modi cation on AF risk. Thirdly, decreased MRF burden was associated with a decreased risk of AF. Our study shows that reducing the MRF burden and maintenance or achievement of MRF ≤1 plays a crucial role in reducing the risk of AF.
Worldwide, a rapid upward trajectory of prevalence and incidence of AF is occurring (12). The global burden-of-disease study highlighted an alarming twofold increase in AF-related mortality between 1990 and 2010 (7), as well as estimated death rates of 20% and 50%, one and ve years, respectively, after initial diagnosis of AF in older adults (13). The burden of AF in Asia is rapidly increasing given the proportional increase in the older population (14,15). In Korea, a large sample cohort study based on nationwide health insurance data showed that the prevalence of AF increased from 0.36% in 2003 to 0.89% in 2013 (16). In addition, hospitalization rates and costs for AF have increased exponentially over the past 10 years, with a decrease in mortality associated with AF hospitalizations (17). AF will represent a signi cant public health burden in the near future. Thus, strategies for AF prevention are of paramount importance to prevent morbidity, mortality, and complications associated with AF.
Studies have reported that the independent risk factors for incident AF include aging, hypertension, congestive heart failure, coronary artery disease, valvular heart disease, diabetes mellitus, male sex, obesity, and excessive alcohol use (8, 18). Furthermore, increased numbers of unhealthy lifestyle factors, including current smoking, heavy drinking (30 g/day) and lack of regular exercise, were associated with a higher risk of incident AF (19). These risk factors play a crucial role in abnormal atrial remodelling, disease progression, and recurrence (20). Similarly, the current study showed that an SBP of more than 140 mmHg, obesity with central obesity, inactivity for leisure time, aging, male sex, and CVD were signi cant risk factors for AF incidence. Particularly, 81.2% of all participants had at least one MRF, and 72.2% (n = 6535) had leisure-time inactivity. In addition, 28.7% of incident AF appears to be attributable to these three MRF, with inactivity during leisure time being the greatest contributing risk factor, indicating the importance of MRF management, especially increasing LTPA.
Although studies for risk prediction and treatment of AF have been extensive, AF prevention has received relatively little attention (7)(8)(9)(10). Current international guidelines recommend modifying an inappropriate diet, quitting smoking, abstaining from alcohol and recreational drugs, and participating in regular physical activity programs as key health behaviours to prevent the development of AF (10). However, many earlier studies also usually assessed risk factors and risk prediction of AF using a single measurement for risk factors (baseline cox model) and provided no information regarding risk factor changes during the follow-up period. These results may have immortal time bias, which can lead to overestimation of the outcome event rate in the unexposed group, underestimation of the event rate in the exposed group, or both (21,22). When incorporating time-updated assessments of directly signi cant MRF, there is limited information on the association between optimal levels of risk factors and AF risk reduction. Our results showed that maintaining or achieving MRF ≤1 signi cantly reduced the risk of AF.
In particular, we identi ed a consistent decrement in AF risk with progressively optimal risk factor pro les, with a striking 71% lowering of AF risk with optimal levels of MRF (reversing SBP of more than 140 mmHg, obesity with central obesity and inactivity for leisure time). Our study suggests that risk factor improvement may decrease AF risk in general population. Similarly, Du X et al. reported that a high proportion of AF can be prevented by combining strategies, focusing on the high-risk population for better risk factor management, and emphasizing healthy lifestyle choices in the whole population (6). Our ndings indicate that it is possible to prevent approximately 29% of AF cases through risk factor modi cation. Unfortunately, we did not nd signi cant associations between MRF combinations and AF risk, as the sample size was too small and statistical power was too low to analyse these outcomes. However, we believe that our ndings provide rmer evidence to establish strategies for AF prevention in the general population.
Although the Framingham Heart Study reported that the risk factor burden, comprising modi able risk factors, and having multiple morbidities play a crucial role in the lifetime risk of AF, associations between MRF burden and incidence of AF have not been previously reported (23). Our ndings showed that the risk of AF progressively decreases according to the decrease in the proportion of visits with more than two MRF during the follow-up period. These results indicate that even if the durations of exposure to MRF are the same, the risk of AF may be lower in those who have a longer period of non-exposure to MRF. We suggest that lengthening the period of non-exposure to MRF (especially when the number of MRF is one or less) during a lifetime could help reduce the risk of AF. Moreover, a log-linear association of high SBP burden with AF incidence suggests that there are cumulative effects of high SBP on the risk of AF. Although we did not nd a progressive decrease according to the decreases in the burdens of obesity with central obesity and that of inactivity, an MRF burden of < 72% for high SBP, 89% for obesity with central obesity, and 88% for inactivity lowered the risk of incident AF more than 50%, compared with an MRF burden of 100%. Therefore, minimizing the MRF burden by early intervention and control could reduce the incidence of AF. There is also potential for the burden and costs of AF to be reduced. Our ndings provide a necessary evidence base to support future investment in intervention trials aimed at modi cation of risk factors for AF in the general population without AF.

Strengths and limitations
This study had several limitations that need to be addressed. First, the study population comprised healthy and middle-aged subjects recruited from two speci c communities in Korea (Ansan and Ansung).
Thus, the PAF estimated in this study, which is population-speci c, may not be applicable to the general Korean population. Second, self-reporting questionnaires may not have accurately re ected the level of LTPA. In addition, the LTPA was divided into only two groups (inactive vs. active), because the majority of the study population had 0 min/week of LTPA (inactivity) and a small event size when categorizing the active group. Thus, we could not assess the association between LTPA intensity and incidence of AF.
Third, information regarding some AF risk factors (e.g., smoking, drinking and sleep apnoea) were not available, due to missing data. We also conducted multivariate model analysis for these factors after excluding missing data, but signi cant associations between these factors and incident AF were not found. Fourth, AF was identi ed biennially using a standard 12-lead ECG, therefore some cases of paroxysmal AF could have been missed. Instead, we additionally analysed the incidence of AF using the Korean Classi cation of Diseases-7 (KCD-7) codes, which is similar to the International Classi cation of Diseases-10 (ICD-10), for 7,620 participants who consented to data linkage between KoGES and the Korean National Health Insurance Service (NHIS) database. We con rmed that the overall incidence rate of AF using KCD-7 codes was 2.5%, which was similar to our results (2.0%).
However, the study had several strengths. The main strengths were its community-based prospective design and the long follow-up period. In addition, to our knowledge, this is the rst study to investigate the association between MRF for incident AF using a time-updated model and assessing the cumulative effects of MRF burden on AF risk in South Korea.

Conclusions
In a prospective cohort study in Korea, our ndings provide support for the concept that targeting MRF, including high SBP, obesity with central obesity, and inactivity for leisure-time, has the potential to signi cantly reduce the individual risk and population burden of AF. Future studies on appropriate level of control of MRF may provide insights into an e cient approach to reduce AF risk or burden in general population. Presently, this needs to be scrutinized by prospective intervention trials to nd suitable level of control of MRF.

Data source and study population
Data were obtained from the Ansan-Ansung cohort within the Korean Genome and Epidemiology Study (KoGES) which is conducted by the Korea National Institute of Health. The Ansan-Ansung cohort is an ongoing, prospective, community-based cohort study that was initiated in 2001-2002. The aim of this cohort study is to ascertain the relationships between genetic, environmental, and lifestyle determinants of chronic diseases such as diabetes mellitus, cerebrovascular disease, and hypertension in Korean people (24). The participants are residents of both urban (Ansan) and rural (Ansung) areas. Enrolment in the study was based on the characteristics of the communities and the most e cient method of recruitment of a representative sample of the Korean population. Initially, the cohort comprised 10,030 participants that were 40-69 years old between 2001 and 2002. Follow-up examinations were conducted biennially between 2003 and 2018. All the participants provided written informed consent. The study protocol was approved by the Institutional Review Board of the Korea Centers for Disease Control and Prevention. All research procedures were performed in accordance to the relevant guidelines and regulations.
In the current study, participants with AF at baseline (n = 39), those with missing values for electrocardiography (ECG) at baseline (n = 12), those without ECG tests during follow-up (n = 55) and those lost to follow-up (n = 875) were excluded from the analysis ( Figure S1). The nal study cohort included 9,049 subjects.
Case ascertainment and follow-up All participants underwent the 12-lead standard ECG at every visit to identify AF. The mortality data for participants lost to follow-up were ascertained by examination of National Death Records. The participants in this study cohort were followed until the index date (AF, death, or end of the study period [December 31, 2018], whichever came rst). The primary outcomes were incident AF or AF-related mortality.

Risk factor ascertainment
Information on the presence of cardiovascular disease (CVD), including myocardial infarction, coronary artery disease, congestive heart failure and stroke/transient ischemic attack, were obtained using a questionnaire at every visit. The questionnaires were administered by trained interviewers according to a speci ed protocol. Blood pressure was measured using a mercury sphygmomanometer, by trained examiners, at least twice at the level of the heart, in a sitting position, and averaged. Systolic blood pressure (SBP) was classi ed into three categories: <120 mmHg, 120-139 mmHg, and ≥140 mmHg. Body mass index (BMI) was calculated as body mass in kilograms divided by height in meters squared. Waist circumference (WC) was measured at the mid-point between the lower ribs and the top of the iliac crest in the standing position. Central obesity was de ned as a WC ≥85 cm in women and ≥90 cm in men (25). Obesity was de ned as a BMI ≥ 25 kg/m 2 , in accordance with the World Health Organization criteria for individuals of Asian descent (26). Estimated glomerular ltration rate (eGFR) was calculated using the Chronic Kidney Disease Epidemiology Collaboration Study (CKD-EPI) equation, and chronic kidney disease (CKD) was de ned as an eGFR ≤60 mL/min/1.73 m 2 . Leisure time physical activity (LTPA); including aerobics, jogging, swimming, tennis, golf, bowling, tness club exercise, walking, and climbing, was assessed using a questionnaire to quantify activities in the leisure time domain. All participants were asked about the types, duration, and frequency of their LTPA. LTPA was then categorized into no physical activity (inactivity) and > 0 min/weeks (active).

Statistical analysis
Baseline characteristics, according to incident AF at the index date, were captured. Continuous variables were expressed as mean ± standard deviation and compared using t-tests; categorical variables were expressed as frequency (percentage) and compared using chi-square tests. Incidence rates of AF were reported as number of patients per 100,000 person-years.
To reduce the possibility of immortal biased effect estimates, we constructed Cox models with timevarying assessment of risk factors to identify signi cant MRF for incident AF. Models were initially adjusted for sex, area, and time-updated assessment of age, SBP, combinations of obesity and central obesity, LTPA, CKD, CVD, HbA1c and total cholesterol. To estimate the joint risk reduction associated with risk factor modi cation, we constructed a risk factor score composed of signi cant MRF (one point each for: SBP ≥140 mmHg, obesity with central obesity and inactivity). Using Cox models with and without time-varying assessment of risk factors ( Figure S2), we compared multivariable-adjusted hazard ratios (HR) for participants with the least favourable risk factor pro le (three points) to those with progressively favourable pro les (two, one, and zero points). The proportional hazards assumption was tested using the scaled Schoenfeld residuals (27). For all the above-mentioned outcomes, we calculated the population attributable fractions (PAF), which re ect the fraction of the event rate or risk, in a given period, attributable to the exposure of interest (assuming a causal relationship). The PAF were computed using indirect standardization using the SAS procedure STDRATE.
To investigate the cumulative effect of MRF on AF risk, we assessed the association with each participant's MRF burden during follow-up. In this study, MRF burden was de ned as the proportion of times MRF appeared during follow-up, based on the number of visits ( Figure S3). We constructed restricted cubic spline curves to identify the association between MRF burden and AF risk.
All statistical tests were two-tailed and P-values < 0.05 were considered statistically signi cant. All statistical analyses were performed using SAS software (ver. 9.4; SAS Institute, Cary, NC, USA) and R 3.5.3 (R Foundation, Vienna, Austria). Figure 1 Modi able risk factors and AF risk in general population using baseline and time-updated models.

Declarations
Modi able risk factors include time-updated systolic blood pressure 140 mmHg, obesity with central obesity, and inactivity. Hazard ratios (HR) for the baseline model were adjusted for age, sex, area, chronic kidney disease, cardiovascular disease, HbA1c and total cholesterol at baseline. Time-updated models were adjusted for sex and area at baseline, and time-updated assessment of age, chronic kidney disease, cardiovascular disease, HbA1c and total cholesterol. MRF=modi able risk factor; HR=hazard ratio; AF=atrial brillation Associations between MRF burden and risk of AF. The solid black line and shaded gray areas represent hazard ratio and 95% con dence bands. Restricted cubic splines for hazard ratios were calculated with a burden of 100% as a reference. MRF=modi able risk factor; SBP=systolic blood pressure; LTPA=leisure