Sex Differences in the Effect of Long-Term Blood Pressure Trajectories and Variability on the Atrial Fibrillation Risk in a Korean Cohort Authors: Hee Jin Kwon, Seon-Ah Jin, Jae-Hyeong Park, Soon-Ki Ahn Published: 2026 ABSTRACT BackgroundLong-term blood pressure (BP) patterns and variability may influence atrial fibrillation (AF) risk. However, whether these associations differ according to sex is unclear.MethodsWe analyzed 45,663 adults (age ≥ 40 years) from a Korean National Health Insurance cohort. Over the past decade, systolic BP (SBP) trajectories have been classified into two distinct trajectory classes: Class 1 (stable-low BP) and Class 2 (sustained-elevated BP). Visit-to-visit SBP variability, quantified as variability independent of the mean (VIM), was stratified into quartiles. The participants were followed up for incident AF between 2012 and 2019. Using sex-stratified Cox proportional hazards models, we assessed the association of SBP trajectory class and VIM with AF risk.ResultsMale sex (adjusted hazard ratio [HR], 1.43; P < 0.001), elevated SBP trajectory (Class 2, adjusted HR, 1.42; P = 0.015), and VIM 4Q (adjusted HR, 1.25; P = 0.007) were associated with a higher risk of AF. Additionally, older age, higher body mass index, alcohol consumption, diabetes, congestive heart failure, and myocardial infarction were independently associated with a higher risk of AF. In men, SBP trajectory Class 2 (adjusted HR, 1.47; P = 0.025) and VIM 4Q (adjusted HR, 1.25; P = 0.038) were significantly associated with an increased incidence of AF, whereas these associations were not significant in women.ConclusionLong-term SBP trajectories and variability demonstrated sex-specific effects on AF risk. An elevated SBP trajectory and high variability conferred an increased risk of AF in men, but not in women. This novel finding may provide new insights into sex-specific differences in the pathogenesis of AF. INTRODUCTION Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia. It is associated with substantial morbidity and mortality, including the risk of heart failure, embolic stroke, and cardiovascular death.1 Its prevalence will double in the coming decades because of several factors, including the aging population, rise in risk factors, heightened awareness among healthcare professionals and the public, and advancements in detection technology.1 Recently updated clinical guidelines emphasize the prevention of AF through a multidisciplinary approach and early management of lifestyle and risk factors.12 Among the increased risk factors for AF, hypertension is a well-estimated risk factor, increasing incident AF risk by 60% in men and 70% in women.3 Because hypertension is a major modifiable contributor to the development of AF, its control can reduce the incidence of AF. In addition to the presence of hypertension, the severity, duration, and variability of elevated blood pressure (BP) may also contribute to AF development.4 A large meta-analysis has revealed that patients with hypertension had an approximately 50% higher risk of developing AF than patients without hypertension and that the risk of developing AF increased by 18% for every 20 mmHg increase in systolic BP (SBP).5 Long-term BP patterns, such as whether an individual’s BP remains controlled or progressively increases over time, may better capture the true cumulative BP burden on AF development than a single measurement. Recent studies have examined BP trajectories and visit-to-visit BP variability as predictors of AF.67 Furthermore, epidemiological studies have identified sex-based disparities in the prevalence and etiology of AF. A higher incidence of AF was observed in men across most age groups. Furthermore, the onset of AF tends to occur at a younger age in men than in women.8 Because women tend to have longer lifespans, their lifetime risk of developing AF eventually approaches that of men. However, the sex differences in BP trajectories and variability in AF development remain unclear. In this study, we investigated the relationship between long-term BP profiles assessed by trajectory and variability and the development of AF, with a particular focus on sex differences, using a large population-based cohort from the Korean National Health Insurance Service (NHIS). The present study hypothesized that individuals with adverse BP profiles, characterized by persistently elevated or rising BP and high BP variability, would have a higher risk of incident AF and that this association might vary between sexes. This study aimed to characterize distinct BP trajectory patterns over a decade, evaluate the association between incident AF, BP trajectory classes, and BP variability, and determine whether these associations differ between men and women. METHODS Data source We used data from the Korean NHIS-National Sample Cohort 2.0 (NHIS-NSC 2.0), a population-based dataset comprising approximately 2% of the total Korean population as of 2006. The Korean NHIS-NSC 2.0 cohort includes retrospective and prospective data collected from 2002 to 2019. To ensure representativeness, the Korean government stratified the entire population into 2,142 strata based on age, sex, region, insurance eligibility, and income level, of which 2.1% (n = 1,021,208) were randomly selected. Given that the Korean NHIS-NSC 2.0 covers roughly 97% of the population, the cohort reflects the national demographic structure. The dataset comprises four key databases: 1) insurance eligibility, which includes socioeconomic information and vital status; 2) medical treatments that capture healthcare utilization and diagnoses coded using the International Classification of Diseases, 10th revision (ICD-10); 3) medical institutions, detailing facility characteristics such as type and location; and 4) general health examinations, which include self-reported health behaviors, physical measurements (e.g., BP and anthropometrics), chest radiographs, and laboratory tests. These health check-ups are recommended biennially for adults in South Korea. Mortality data were obtained from death registration records maintained by Statistics Korea, a national statistical authority. Study population This study included adults aged ≥ 40 years who underwent more than one national health check-up between 2002 and 2011, among whom individuals who underwent longitudinal BP assessments across five waves of biennial checkups within the study period were eligible for enrollment. Hypertension was identified using the relevant ICD-10 codes (I10, I11, I12, I13, and I15), prescription records of antihypertensive agents, and self-reported histories from the health screening questionnaire. Diabetes mellitus (DM) was defined using the ICD-10 codes E10, E11, E12, E13, and E14, prescriptions of antidiabetic medications, or a questionnaire-based history. Dyslipidemia was determined using the ICD-10 codes E78, E780, E781, E782, E783, E784, and E785, lipid-lowering therapy records, and self-reported diagnoses. Congestive heart failure (CHF) was identified using the relevant ICD-10 codes I50.1, I50.2, I50.3, I50.4, I50.8, and I50.9. Myocardial infarction (MI) was identified using the ICD-10 codes I21, I21.4, I21.01, I21.9, I22, and I25.2. Chronic kidney disease (CKD) diagnosis required two outpatient hospital visits or one admission using the following ICD-10 codes: N18, N18.1, N18.2, N18.3, N18.4, N18.5, N18.9, Z49, Z94, Z94.0, O7020, O7021, O7031, O7032, O7033, O7034, and R3280. Stroke was determined using the ICD-10 codes I60.0-I60.9, I61, I63, I64, and I69.3. Baseline lifestyle and behavioral factors were assessed as follows: smoking status was categorized as never, past, or current smoker; alcohol consumption frequency as never, 1–2 times/week, or ≥ 3 times/week. Physical activity was classified according to frequency (none, 1–4 days/week, or ≥ 5 days/week). Family income was stratified into low (1st to 3rd deciles), middle (4th to 7th deciles), and high (8th to 10th deciles), based on household insurance premium levels. Height and weight were measured by trained personnel following standardized protocols, and the body mass index (BMI) was calculated as weight (kg) divided by height in meters squared. Baseline laboratory evaluations included fasting glucose, total cholesterol, and hemoglobin levels. Individuals were excluded if they met any of the following predefined criteria: 1) a prior diagnosis of AF at baseline, 2) death during the screening period, or 3) incomplete regular BP measurements. Outcome The primary outcome of this study was the development of incident AF during the follow-up period from 2012 to 2019, stratified according to BP trajectory and variability. AF was identified using medical claims data and defined as either 1) hospital admission with a primary diagnosis of AF or 2) outpatient visits for AF recorded on at least two separate days. Diagnoses were confirmed using relevant ICD-10 codes, including I48.0, I48.1, I48.2, I48.3, I48.4, and I48.9. Follow-up was censored at the time of AF onset or the end of the study period (December 31, 2019), whichever occurred first. Classification of SBP trajectories and variability SBP values measured in the right upper arm during biennial national health checkups were used to construct BP trajectories. The SBP trajectories were determined using measurements obtained from biennial health examinations conducted during the study period. As most participants underwent health checkups with BP readings every 1–2 years, we organized the SBP data into five waves, each representing a 2-year interval. The average SBP was used for individuals with BP measurements in both years of the wave. If only one measurement was available within a wave, the value was used as the representative SBP value. The participants included in the trajectory analysis underwent at least three SBP measurements over a 10-year period. To identify distinct patterns of SBP change over time, we applied latent class mixed modeling (LCMM) using the LCMM package (version 2.0.2) in R (version 4.3.0; R Foundation for Statistical Computing, Vienna, Austria). This approach models longitudinal data as a finite mixture of latent subgroups with similar trajectory patterns estimated via maximum likelihood.9101112 The modeling process follows a three-step procedure. In the first step (initial hypothesis), we assumed that SBP trajectories could be classified into two to four latent groups. During model development, we employed a censored normal model appropriate for continuous outcomes such as SBP. We began with a linear polynomial function and evaluated the model fit across different group numbers using the Bayesian Information Criterion (BIC), with lower values indicating a better fit while penalizing model complexity. In the final step (model selection), the optimal model was selected based on several criteria: 1) the lowest BIC, 2) average posterior probabilities of group membership exceeding 0.6, 3) odds of correct classification > 5 for each group, 4) each class representing more than 5% of the sample, and 5) the clinical interpretability of the trajectories.1314 The final selected model comprised two linear SBP trajectory groups that provided the most parsimonious and clinically meaningful fit (Supplementary Table 1). The first group (trajectory Class 1) included individuals who maintained relatively low SBP throughout the study period. The second group (trajectory Class 2) included individuals with a sustained elevated SBP. SBP variability during the screening period was assessed using variability independent of the mean (VIM), which quantifies intraindividual BP variability while minimizing dependence on mean SBP. VIM was calculated as follows: where SD and mean are the standard deviation and average SBP for each individual, respectively, β is derived from the regression of log(SD) on log(mean), and the population mean is used for standardization. A higher VIM indicates significant BP variability, although the BP itself is not necessarily elevated. This method allows for an unbiased comparison of SBP variability across individuals with different average BP levels. Statistical analysis Participant characteristics are presented as means with standard deviations for continuous variables and as counts with percentages for categorical variables. Group differences were evaluated using the Student’s t-test or χ2 test, as appropriate. Multivariable Cox proportional hazards models were used to examine the associations among SBP trajectories, SBP VIM quartiles, and AF incidence. The results are expressed as hazard ratios (HRs) with corresponding 95% confidence intervals (CIs). All statistical analyses were conducted using SAS version 9.2 (SAS Institute, Cary, NC, USA) and R version 4.3.0 (R Foundation for Statistical Computing). Ethics statement The study protocol was approved by the Institutional Review Board (IRB) of the Chungnam National University Hospital (CNUH 2024-03-057), and this research was conducted in good observance of ethics following the latest version (2013) of the Code of Ethics of the World Medical Association (Declaration of Helsinki) for research involving humans. This study used a sample cohort obtained from the Korean National Health Checkup. The IRB waived the requirement for consent from the participants. RESULTS A total of 554,320 adults aged ≥ 40 years who underwent at least one health examination through the Korean NHIS between January 1, 2002, and December 31, 2011, were initially screened (Fig. 1). After excluding 21,931 individuals, 10,290 who died during the screening period, 10,759 with preexisting AF, and 882 with missing BP data, 532,389 participants were deemed eligible for further analysis. A total of 45,663 participants completed all five BP assessments and were included in the final study cohort. These individuals were followed up for incident AF from January 1, 2012 to December 31, 2019. The study flowchart delineates the participant selection process from the Korean National Health Insurance Service health checkup database from 2002 to 2019. From the initial population of 554,320 adults aged ≥ 40 years, 21,931 were excluded from the study. The remaining 532,389 individuals were eligible for longitudinal BP assessment across five waves of biennial checkups from 2002 to 2011. A total of 45,663 participants with complete BP measurements were stratified by sex and SBP trajectory and variability classes using VIM. The AF incidence during the FU period (2012–2019) is reported for each subgroup. AF = atrial fibrillation, BP = blood pressure, SBP = systolic blood pressure, VIM = variability independent of the mean, FU = follow-up. The participants were divided into categories based on their SBP trajectories and variability using VIM. Among the 27,085 men, 1,747 were classified as SBP trajectory Class 1 (stable low SBP) and 25,338 as Class 2 (sustained elevated SBP). During the follow-up period, AF developed in 41 patients (2.3%) in Class 1 and 854 (3.4%) in Class 2. Among the 18,578 women, 637 were categorized as Class 1 and 17,941 as Class 2. Incident AF occurred in 15 (2.4%) patients in Class 1 and 467 (2.6%) in Class 2. To further characterize longitudinal SBP trends, we plotted individual SBP trajectories by age and wave for each sex. Among men (Fig. 2A and B), those in Class 1 exhibited relatively stable SBP patterns across ages and waves. In contrast, those in Class 2 demonstrated greater dispersion and instability in SBP values, with higher overall mean SBP values and a tendency toward progressive elevation with advancing age. When stratified by the measurement wave (Fig. 2B), men in Class 1 maintained consistent SBP readings across all five time points, whereas those in Class 2 exhibited marked fluctuations with wide inter-individual variation. Similar trends were observed in men and women (Fig. 2C and D). Individual SBP trajectories by variability class and sex. The values of SBP are displayed according to the trajectory group, with the classification of age (A) and waves (B) in men. The SBP trajectory Class 1 (left) demonstrates a typical and relatively stable SBP during the follow-up duration (A). The following investigation will examine the utilization of SBP patterns when plotted across a series of five examination waves in the male participants. Individuals classified as Class 1 exhibit relatively stable SBP values, whereas those classified as Class 2 demonstrate significant variability in their SBP measurements. When SBP values are shown by trajectory group according to age (C) and waves (D) in women, similar findings are shown like in men. The blue lines in the figure represent smoothed locally weighted scatterplot smoothing curves, which summarize the overall trend patterns. SBP = systolic blood pressure. Baseline characteristics according to sex This study included 18,578 women and 27,085 men (Table 1). The women were older than the men (51.5 ± 8.6 years vs. 49.6 ± 8.4 years, P < 0.001). A higher percentage of men were in the younger age group, with 34.9% of men and 25.7% of women aged < 45 years. Conversely, women constituted a larger proportion of older age groups, with 25.8% of women and 18.8% of men in the 55- to 64-year-old category, and 9.2% of women and 7.2% of men were ≥ 65 years old. Table 1Comparison of baseline characteristics according to the sexVariablesFemales (n = 18,578)Males (n = 27,085)P valueAge, yr51.5 ± 8.649.6 ± 8.4< 0.001Age group, yr< 0.001< 454,775 (25.7)9,461 (34.9)45–547,303 (39.3)10,578 (39.1)55–644,793 (25.8)5,083 (18.8)≥ 651,707 (9.2)1,963 (7.2)Average SBP, mmHgWave 1123.3 ± 17.7127.3 ± 15.7< 0.001Wave 2123.3 ± 16.9127.2 ± 15.1< 0.001Wave 3123.6 ± 16.3126.4 ± 14.3< 0.001Wave 4123.4 ± 15.6126.1 ± 13.6< 0.001Wave 5123.6 ± 15.0126.1 ± 13.5< 0.001Trajectory class< 0.001Class 1637 (3.4)1,747 (6.5)Class 217,941 (96.6)25,338 (93.5)SBP VIM1.0001Q4,646 (25.0)6,773 (25.0)2Q4,644 (25.0)6,770 (25.0)3Q4,644 (25.0)6,772 (25.0)4Q4,644 (25.0)6,770 (25.0)BMI group, kg/m2 < 0.001< 18.5362 (2.0)485 (1.8)18.5–22.97,304 (39.3)9,018 (33.3)23.0–24.94,993 (26.9)8,068 (29.8)≥ 25.05,905 (31.8)9,506 (35.1)Income status< 0.001Low5,163 (28.1)3,428 (13.0)Middle5,828 (31.7)8,944 (33.9)High7,414 (40.3)13,997 (53.1)Smoking status< 0.001Nonsmoker17,483 (97.5)11,199 (41.9)Ex-smoker170 (0.9)4,946 (18.5)Current smoker270 (1.5)10,568 (39.6)Alcohol consumption< 0.001Almost none14,936 (82.2)8,676 (32.4)1–2 times per week2,949 (16.2)13,717 (51.2)Over 3 times per week286 (1.6)4,389 (16.4)Physical activity< 0.001Almost none11,602 (64.2)11,787 (44.8)1–4 days per week4,816 (26.6)12,198 (46.3)Over 5 days per week1,664 (9.2)2,354 (8.9)Hypertension8,579 (48.4)11,855 (44.7)< 0.001Diabetes mellitus5,106 (27.6)7,126 (26.6)0.012Dyslipidemia9,451 (50.9)11,385 (42.1)< 0.001CKD116 (0.6)247 (0.9)< 0.001CHF1,664 (9.0)1,797 (6.6)< 0.001Stroke1,802 (9.7)2,103 (7.8)< 0.001MI543 (2.9)938 (3.5)< 0.001Incident AF, 2012–2019482 (2.6)895 (3.3)< 0.001Values are presented as mean ± standard deviation or number (%).SBP = systolic blood pressure, VIM = variability independent of mean, BMI = body mass index, CKD = chronic kidney disease, CHF = congestive heart failure, MI = myocardial infarction, AF = atrial fibrillation. Comparison of baseline characteristics according to the sex Values are presented as mean ± standard deviation or number (%). SBP = systolic blood pressure, VIM = variability independent of mean, BMI = body mass index, CKD = chronic kidney disease, CHF = congestive heart failure, MI = myocardial infarction, AF = atrial fibrillation. The men consistently had higher average SBP across all five waves than the women (e.g., Wave 1: 127.3 ± 15.7 mmHg in men vs. 123.3 ± 17.7 mmHg in women; all P < 0.001). Regarding the SBP trajectory class, significantly more men (6.5%) were in Class 1 than women (3.4%), whereas women had a higher percentage in Class 2 (96.6% vs. 93.5% for men) (P < 0.001). The BMI group distribution also differed significantly between sexes (P < 0.001). Women had a higher prevalence in the 18.5–22.9 kg/m2 BMI group (39.3%) than men (33.3%). Conversely, men had a higher percentage in the ≥ 25.0 kg/m2 BMI group (35.1%) than women (31.8%). Income status also differed significantly by sex (P < 0.001), with a higher proportion of women in the low-income group (28.1% vs. 13.0%) and a higher proportion of men in the high-income group (53.1% vs. 40.3%). Regarding smoking status, the men had a significantly higher percentage of current smokers (39.6% vs. 1.5%) and ex-smokers (18.5% vs. 0.9%) than women (P < 0.001). The frequency of alcohol consumption was also higher among men. Finally, the prevalence of comorbidities differed significantly, with women having a higher prevalence of hypertension (48.4% vs. 44.7% for men, P < 0.001) and dyslipidemia (50.9% vs. 42.1% for men, P < 0.001). However, men had a higher prevalence of CKD (0.9% vs. 0.6%, P < 0.001), MI (3.5% vs. 2.9%, P < 0.001), and AF incidence than women (3.3% vs. 2.6%, P < 0.001). The incidences of CHF (9.0% vs. 6.6%, P < 0.001) and stroke (9.7% vs. 7.8%, P < 0.001) were higher among women. The prevalence of DM was significantly higher among women than among men (27.6% vs. 26.6%, P = 0.012). Baseline characteristics according to SBP trajectory classes Participants classified into the SBP trajectory Class 1 exhibited significantly lower baseline SBP across all time points than those in Class 2 (e.g., Wave 1: 109.4 ± 10.7 mmHg vs. 126.6 ± 16.5 mmHg, P < 0.001, Table 2). Class 1 individuals were older on average (53.0 ± 7.7 years vs. 50.2 ± 8.6 years, P < 0.001), more likely to be male (73.3% vs. 58.5%, P < 0.001), and more often in the lower BMI categories. A significantly lower proportion of Class 1 participants had hypertension (19.0% vs. 46.2%, P < 0.001), DM (22.3% vs. 27.0%, P < 0.001), dyslipidemia (37.9% vs. 46.1%, P < 0.001), CHF (4.8% vs. 7.7%, P < 0.001), and stroke (7.3% vs. 8.6%, P = 0.027). AF was slightly less frequent in Class 1 but did not reach statistical significance (2.3% vs. 3.1%, P = 0.058). Table 2Comparison of baseline characteristics according to the presence of systolic blood pressure trajectory classVariablesClass 1 (n = 2,384)Class 2 (n = 43,279)P valueSex, female637 (26.7)17,941 (41.5)< 0.001SBP VIM1.0001Q597 (25.0)10,822 (25.0)2Q596 (25.0)10,818 (25.0)3Q597 (25.0)10,819 (25.0)4Q594 (24.9)10,820 (25.0)Age, yr53.0 ± 7.750.2 ± 8.6< 0.001Age group, yr< 0.001< 45313 (13.1)13,923 (32.2)45–541,149 (48.2)16,732 (38.7)55–64705 (29.6)9,171 (21.2)≥ 65217 (9.1)3,453 (8.0)Average SBP, mmHgWave 1109.4 ± 10.7126.6 ± 16.5< 0.001Wave 2108.5 ± 10.0126.6 ± 15.7< 0.001Wave 3107.1 ± 9.6126.2 ± 14.8< 0.001Wave 4106.7 ± 9.3126.0 ± 14.2< 0.001Wave 5106.2 ± 9.3126.2 ± 13.7< 0.001BMI group, kg/m2 < 0.001< 18.5122 (5.1)725 (1.7)18.5–22.91,268 (53.2)15,054 (34.8)23.0–24.9580 (24.3)12,481 (28.9)≥ 25.0413 (17.3)14,998 (34.7)Income status0.005Low369 (15.7)8,222 (19.4)Middle669 (28.5)14,103 (33.2)High1,310 (55.8)20,101 (47.4)Smoking status< 0.001Nonsmoker1,312 (56.3)27,370 (64.7)Ex-smoker288 (12.4)4,828 (11.4)Current smoker730 (31.3)10,108 (23.9)Alcohol consumption< 0.001Almost none1,336 (57.1)22,276 (52.3)1–2 times per week808 (34.5)15,858 (37.2)Over 3 times per week197 (8.4)4,478 (10.5)Physical activity0.355Almost none1,254 (54.0)22,135 (52.6)1–4 days per week869 (37.4)16,145 (38.3)Over 5 days per week198 (8.5)3,820 (9.1)Hypertension454 (20.2)19,980 (47.6)< 0.001Diabetes mellitus532 (22.6)11,700 (27.2)< 0.001Dyslipidemia903 (37.9)19,933 (46.1)< 0.001CKD19 (0.8)344 (0.8)1.000CHF115 (4.8)3,346 (7.7)< 0.001Stroke174 (7.3)3,731 (8.6)0.027MI81 (3.4)1,400 (3.2)0.706Incident AF, 2012–201956 (2.3)1,321 (3.1)0.058Values are presented as number (%) or mean ± standard deviation.SBP = systolic blood pressure, VIM = variability independent of mean, BMI = body mass index, CKD = chronic kidney disease, CHF = congestive heart failure, MI = myocardial infarction, AF = atrial fibrillation. Comparison of baseline characteristics according to the presence of systolic blood pressure trajectory class Values are presented as number (%) or mean ± standard deviation. SBP = systolic blood pressure, VIM = variability independent of mean, BMI = body mass index, CKD = chronic kidney disease, CHF = congestive heart failure, MI = myocardial infarction, AF = atrial fibrillation. Baseline characteristics according to SBP variability (VIM) quartile With increasing SBP variability, participants became progressively older (mean age from 48.8 to 52.4 years across quartiles, P < 0.001), and the prevalence of cardiovascular risk factors increased significantly (Table 3). Hypertension increased from 39.3% in the lowest VIM quartile to 54.3% in the highest quartile (P < 0.001), as did DM (P < 0.001), dyslipidemia (P < 0.001), alcohol consumption (P < 0.001), and physical inactivity (P < 0.001). The prevalence of CKD, CHF, stroke, and MI increased significantly from the lowest to the highest VIM quartile. The incidence of AF increased with increasing VIM class (from 2.3% to 3.8%; P < 0.001). However, BMI and smoking distributions were largely stable across the VIM classes, with modest shifts. Table 3Comparison of baseline characteristics according to the presence of systolic blood pressure variability classVariables1Q (n = 11,419)2Q (n = 11,414)3Q (n = 11,416)4Q (n = 11,414)P valueSex, female4,646 (40.7)4,644 (40.7)4,644 (40.7)4,644 (40.7)1.000SBP trajectory1.000Class 1597 (5.2)596 (5.2)597 (5.2)594 (5.2)Class 210,822 (94.8)10,818 (94.8)10,819 (94.8)10,820 (94.8)Age, yr48.8 ± 7.849.6 ± 8.250.7 ± 8.652.4 ± 9.2< 0.001Age group, yr< 0.001< 454,183 (36.6)3,836 (33.6)3,420 (30.0)2,797 (24.5)45–544,667 (40.9)4,617 (40.5)4,455 (39.0)4,142 (36.3)55–642,002 (17.5)2,222 (19.5)2,586 (22.7)3,066 (26.9)≥ 65567 (5.0)739 (6.5)955 (8.4)1,409 (12.3)BMI group, kg/m2 < 0.001< 18.5180 (1.6)167 (1.5)224 (2.0)276 (2.4)18.5–22.93,990 (35.0)4,115 (36.1)3,995 (35.0)4,222 (37.0)23.0–24.93,386 (29.7)3,224 (28.3)3,285 (28.8)3,166 (27.8)≥ 25.03,860 (33.8)3,902 (34.2)3,906 (34.2)3,743 (32.8)Income status0.017Low2,111 (19.7)2,231 (20.0)2,070 (18.5)2,079 (18.5)Middle3,637 (32.5)3,638 (32.6)3,709 (33.1)3,788 (33.8)High5,348 (47.8)5,301 (47.5)5,418 (48.4)5,344 (47.7)Smoking status0.012Nonsmoker7,111 (63.8)7,154 (64.1)7,239 (64.7)7,178 (64.4)Ex-smoker1,342 (12.0)1,301 (11.7)1,296 (11.6)1,177 (10.6)Current smoker2,695 (24.2)2,704 (24.2)2,654 (23.7)2,785 (25.0)Alcohol consumption< 0.001Almost none5,693 (50.6)5,880 (52.3)5,964 (53.0)6,075 (54.2)1–2 times per week4,456 (39.6)4,238 (37.7)4,133 (36.8)3,839 (34.2)Over 3 times per week1,094 (9.7)1,129 (10.0)1,148 (10.2)1,304 (11.6)Physical activity< 0.001Almost none5,599 (50.4)5,731 (51.6)5,836 (52.6)6,223 (56.0)1–4 days per week4,554 (41.0)4,378 (39.4)4,252 (38.3)3,830 (34.5)Over 5 days per week950 (8.6)1,005 (9.0)1,011 (9.1)1,052 (9.5)Hypertension4,350 (39.3)4,783 (43.3)5,292 (47.8)6,009 (54.3)< 0.001Diabetes mellitus2,671 (23.6)2,941 (26.0)3,134 (27.7)3,486 (30.7)< 0.001Dyslipidemia4,765 (41.8)5,119 (44.9)5,319 (46.6)5,633 (49.4)< 0.001CKD71 (0.6)78 (0.7)105 (0.9)109 (1.0)0.007CHF632 (5.5)808 (7.1)875 (7.7)1,146 (10.0)< 0.001Stroke772 (6.8)907 (7.9)967 (8.5)1,259 (11.0)< 0.001MI265 (2.3)317 (2.8)363 (3.2)432 (3.8)< 0.001Incident AF, 2012–2019265 (2.3)317 (2.8)363 (3.2)432 (3.8)< 0.001Values are presented as number (%) or mean ± standard deviation.SBP = systolic blood pressure, BMI = body mass index, CKD = chronic kidney disease, CHF = congestive heart failure, MI = myocardial infarction, AF = atrial fibrillation. Comparison of baseline characteristics according to the presence of systolic blood pressure variability class Values are presented as number (%) or mean ± standard deviation. SBP = systolic blood pressure, BMI = body mass index, CKD = chronic kidney disease, CHF = congestive heart failure, MI = myocardial infarction, AF = atrial fibrillation. Baseline characteristics according to incident AF Participants who developed AF were older (56.1 ± 8.5 years vs. 50.2 ± 8.5 years, P < 0.001), more likely to be male (65.0% vs. 59.1%, P < 0.001), and had a higher prevalence of cardiovascular comorbidities, including hypertension (70.1% vs. 45.4%, P < 0.001), DM (38.7% vs. 26.6%, P < 0.001), dyslipidemia (53.2% vs. 45.4%, P < 0.001), CHF (17.3% vs. 7.3%, P < 0.001), stroke (17.6% vs. 8.3%, P < 0.001), and MI (6.1% vs. 3.2%, P < 0.001, Table 4). AF incidence was positively associated with increasing SBP variability (VIM Q4: 31.4% vs. Q1: 19.2%, P < 0.001), whereas no significant difference in SBP trajectory class distribution was observed (P = 0.058). A higher proportion of AF patients were in the highest BMI category (≥ 25.0 kg/m2: 41.1% vs. 33.5%, < 0.001). Table 4Comparison of baseline characteristics according to the incident atrial fibrillation during the follow-up duration (2012–2019)VariablesAF (+) (n = 1,377)AF (−) (n = 44,286)P valueSex, female482 (35.0)18,096 (40.9)< 0.001SBP trajectory class0.058Class 156 (4.1)2,328 (5.3)Class 21,321 (95.9)41,958 (94.7)SBP VIM< 0.0011Q265 (19.2)11,154 (25.2)2Q317 (23.0)11,097 (25.1)3Q363 (26.4)11,053 (25.0)4Q432 (31.4)10,982 (24.8)Age, yr56.1 ± 8.550.2 ± 8.5< 0.001Age group, yr< 0.001< 45185 (13.4)14,051 (31.7)45–54416 (30.2)17,465 (39.4)55–64488 (35.4)9,388 (21.2)≥ 65288 (20.9)3,382 (7.6)BMI group, kg/m²< 0.001< 18.522 (1.6)825 (1.9)18.5–22.9403 (29.3)15,919 (35.9)23.0–24.9386 (28.0)12,675 (28.6)≥ 25.0566 (41.1)14,845 (33.5)Income status0.905Low254 (18.7)8,337 (19.2)Middle452 (33.4)14,320 (33.0)High649 (47.9)20,762 (47.8)Smoking status0.084Nonsmoker843 (62.4)27,839 (64.3)Ex-smoker180 (13.3)4,936 (11.4)Current smoker328 (24.3)10,510 (24.3)Alcohol consumption< 0.001Almost none700 (51.7)22,912 (52.6)1–2 times per week433 (32.0)16,233 (37.2)Over 3 times per week222 (16.4)4,453 (10.2)Physical activity< 0.001Almost none727 (54.0)22,662 (52.6)1–4 days per week444 (33.0)16,570 (38.5)Over 5 days per week175 (13.0)3,843 (8.9)Hypertension938 (70.1)19,496 (45.4)< 0.001Diabetes mellitus530 (38.7)11,702 (26.6)< 0.001Dyslipidemia732 (53.2)20,104 (45.4)< 0.001CKD15 (1.1)348 (0.8)0.274CHF238 (17.3)3,223 (7.3)< 0.001Stroke242 (17.6)3,663 (8.3)< 0.001MI84 (6.1)1,397 (3.2)< 0.001Values are presented as number (%) or mean ± standard deviation.SBP = systolic blood pressure, VIM = variability independent of mean, BMI = body mass index, CKD = chronic kidney disease, CHF = congestive heart failure, MI = myocardial infarction, AF = atrial fibrillation. Comparison of baseline characteristics according to the incident atrial fibrillation during the follow-up duration (2012–2019) Values are presented as number (%) or mean ± standard deviation. SBP = systolic blood pressure, VIM = variability independent of mean, BMI = body mass index, CKD = chronic kidney disease, CHF = congestive heart failure, MI = myocardial infarction, AF = atrial fibrillation. SBP trajectories, VIM, and the future development of AF Fig. 3 shows the Kaplan–Meier analyses of the development of AF according to the SBP trajectory class and SBP VIM. In the overall cohort, individuals in the sustained elevated SBP group (SBP trajectory Class 2) exhibited a higher cumulative incidence of AF than those in the stable low BP group (SBP trajectory Class 1), although the difference was not statistically significant (log-rank P = 0.057; Fig. 3A). Cumulative incidence of AF stratified by SBP trajectory class and VIM quartiles. Comparison of AF incidence between SBP trajectory Class 1 (normal BP values) and Class 2 (high BP values). A trend toward higher AF incidence was observed in Class 2 (P = 0.057) (A). When we plot cumulative incidence of AF across VIM quartiles in the entire cohort (B), a significant stepwise increase in AF risk was noted with increasing SBP variability (P < 0.001). Subgroup analysis of the SBP trajectory Class 1 individuals revealed no significant difference in AF incidence across VIM quartiles (P = 0.500) (C). Among the SBP trajectory Class 2 individuals, those in higher VIM quartiles exhibited significantly greater AF incidence (P < 0.001), indicating an interaction between SBP trajectory instability and visit-to-visit BP variability (D). The number at risk for each group is presented below each panel. SBP = systolic blood pressure, VIM = variability independent of the mean, AF = atrial fibrillation. However, when participants were stratified into VIM quartiles, a stepwise increase in AF risk was observed with increasing variability levels (Fig. 3B). The highest quartile of VIM (4Q) was associated with the greatest cumulative incidence of AF, whereas the lowest quartile (1Q) had the lowest incidence (log-rank P < 0.001), indicating a dose–response relationship between SBP variability and AF development. Subgroup analyses according to trajectory class revealed distinct patterns. Among the Class 1 individuals (Fig. 3C), no significant differences in AF incidence were observed across the VIM quartiles (log-rank P = 0.500), suggesting that in individuals with a stable low BP, the additive effect of VIM-defined variability may be minimal. Conversely, in the Class 2 subgroup (Fig. 3D), the cumulative incidence of AF increased progressively with higher VIM quartiles (log-rank P < 0.001), confirming that greater within-individual BP fluctuations over time further compounded the risk of AF in the sustained elevated SBP trajectory group. These findings highlight the independent and synergistic effects of long-term BP trajectory patterns and beat-to-beat SBP variability on AF development. In the multivariable Cox regression analysis adjusted for demographic, clinical, and socioeconomic variables (Fig. 4), several factors were independently associated with the incidence of AF. Male sex was associated with a significantly higher risk of incident AF than female sex (HR, 1.43; 95% CI, 1.23–1.65; P < 0.001). Similarly, participants classified in the high BP trajectory variability group (Class 2) had a higher AF risk than those in Class 1 (HR, 1.42; 95% CI, 1.07–1.88; P = 0.015). Multivariable Cox proportional hazards analysis of factors associated with incident AF. Forest plot showing adjusted HRs and 95% CIs for clinical and demographic variables in the entire cohort. Male sex, higher SBP trajectory class (Class 2), upper VIM quartile (4Q), older age, higher BMI (≥ 25.0 kg/m2), and diabetes mellitus were independently associated with higher AF risk. Variables such as income, smoking, and dyslipidemia were not statistically significant predictors after adjustment. HR = hazard ratio, CI = confidence interval, SBP = systolic blood pressure, VIM = variability independent of the mean, BMI = body mass index, DM = diabetes mellitus, CKD = chronic kidney disease, CHF = congestive heart failure, MI = myocardial infarction, AF = atrial fibrillation. When SBP variability was further stratified into VIM quartiles, the highest quartile (4Q) was associated with a significantly increased risk of AF compared with the lowest quartile (1Q) (HR, 1.25; 95% CI, 1.06–1.47; P = 0.007), with a trend toward increased risk in the 3Q group. Age was the most potent risk factor, with individuals aged ≥ 65 years showing a nearly 5.5-fold higher risk (HR, 5.45; 95% CI, 4.46–6.69; P < 0.001) than those aged < 45 years. A higher BMI (≥ 25.0 kg/m2) was also significantly associated with AF (HR, 1.29; 95% CI, 1.13–1.48; P < 0.001), whereas DM conferred a modest but statistically significant elevation in risk (HR, 1.15; 95% CI, 1.02–1.30; P = 0.021). Frequent alcohol consumption (≥ 3 times per week) was also associated with AF (HR, 1.39; 95% CI, 1.17–1.65; P < 0.001) compared with no alcohol consumption. Income level, smoking status, and dyslipidemia were not significantly associated with AF incidence after adjustment. The sex-stratified analyses are shown in Supplementary Fig. 1. Among men, both SBP trajectory Class 2 (HR, 1.47; 95% CI, 1.05–2.05; P = 0.025) and VIM 4Q (HR, 1.25; 95% CI, 1.01–1.54; P = 0.038) remained significant predictors of AF onset. The effect of age was especially prominent, with those aged ≥ 65 years exhibiting a 4.93-fold higher risk than those aged < 45 years (P < 0.001). Elevated BMI, DM, CHF, and stroke were independently associated with AF in men. In contrast, in women, neither SBP trajectory Class 2 nor VIM quartiles were significantly associated with AF risk. Age remained a strong predictor of AF, with a similar stepwise increase in risk across the age groups. DM, CHF, and stroke were significantly associated with AF in women, whereas BMI, income, smoking status, and dyslipidemia were not independent predictors. Therefore, the prognostic effect of SBP variability may be more prominent in men than in women. DISCUSSION In this large population-based cohort with a 10-year screening period using BP data and an 8-year follow-up period for incident AF, long-term SBP trajectories and variability were identified as important predictors of AF development; however, their impact differed markedly according to sex. Specifically, men who exhibited sustained-elevated SBP (trajectory Class 2) or the highest SBP variability (VIM 4Q) were significantly more likely to develop AF. In contrast, we observed no significant association between SBP trajectory or variability and incidence of AF in women. These novel findings contribute to the growing evidence on how BP influences cardiac arrhythmias over time, which can vary between sexes. Elevated BP is a well-established and modifiable risk factor for the development of AF.4 Prior studies have demonstrated that BP trajectories, as well as single BP measurements, are associated with the risk of AF.6 In our cohort, an individual’s long-term BP pattern provides prognostic information beyond a single BP reading. In our cohort comprising many healthy individuals, those classified as Class 1 had normal BP, whereas those classified as Class 2 had relatively elevated BP. Even within the normal BP range, individuals with a persistently elevated SBP had a higher incidence of AF than those with a stable low SBP. This is consistent with the notion that chronically elevated afterload leads to atrial remodeling (e.g., left atrial [LA] enlargement and fibrosis), which fosters AF.12 Moreover, we demonstrated that SBP variability was an independent risk factor for AF, in line with recent findings in other populations.7 Mechanistically, intermittent surges in BP could provoke transient hemodynamic stress in the LA and left ventricle, promoting inflammation and triggering arrhythmogenic cardiac autonomic responses, all of which could contribute to AF initiation over time. Sustained elevated BP induces functional and structural cardiac remodeling, including ventricular and atrial remodeling. LA remodeling is a critical mechanism in the initiation and perpetuation of AF.2 Cumulative higher BP exposure at a young age is independently associated with adverse LA remodeling. Animal models have demonstrated that an elevated BP is associated with atrial electrical and structural remodeling, including LA conduction abnormalities, LA hypertrophy, fibrosis, and inflammation. These changes are favorable for AF initiation and maintenance.2 Furthermore, BP variability represents an individual’s inability to maintain homeostasis and reflects the underlying vascular dysfunction, autonomic dysregulation, or other pathophysiological processes that contribute to AF pathogenesis.15 Several major cohort studies have shown positive evidence of long-term BP patterns and their association with AF incidence.61617 Compared with consistently normotensive individuals, those with persistently elevated SBP trajectories have a significantly higher AF risk. In addition to mean BP, visit-to-visit BP variability has emerged as an independent risk factor for AF. A Korean population-based cohort study reported that higher BP variability is associated with a modestly increased risk of AF.18 The impact is stronger in high-risk individuals, including older adults (≥ 65 years), those with DM, or those with CKD.18 Similarly, the J-RHYTHM Registry demonstrated that visit-to-visit variability of SBP is significantly associated with adverse events, including thromboembolism, major hemorrhage, and all-cause death in patients with AF.19 These results reinforce the concept that the cumulative burden of elevated SBP can be a driver of AF. Moreover, prior studies have shown that hypertension not only predisposes patients to AF but also that the magnitude of BP elevation, duration of hypertension, and even visit-to-visit BP fluctuations can modulate this risk.12 The most striking aspect of our findings is the difference in the significance of BP trajectory/variability in AF risk between men and women. Male sex is a well-known risk factor for AF, and in our study, men had a 43% higher risk of developing AF than women of similar age and profile. A previous study using the Tromsø Study data has reported on the association of long-term BP trajectories with increased AF incidence in both sexes, although the association was stronger in women than in men.17 However, we observed that the excess AF risk conferred by an adverse SBP trajectory was essentially confined to men. Men with a long-term pattern of high SBP (versus those with a stable-normal BP) had a 1.47-fold greater risk of AF than those with a stable normal BP, whereas women with a similarly high BP trajectory did not show a significant increase in risk. Similarly, high SBP variability was detrimental in men but not in women. These observations are in contrast with those of previous studies. For instance, Sharashova et al.17 have reported in the Tromsø study that long-term SBP trajectories were associated with AF in both sexes, but with stronger relative risks in women than men. They also reported that women who maintained high SBP for around 15 years had approximately twice the risk of AF as women with low BP, whereas men with long-term high BP had a more modest (about 50%) increase in risk.17 Another analysis from a large-scale health checkup and administrative claims data from more than three million Japanese individuals demonstrated that hypertension was associated with a higher risk of developing AF in both sexes, and the relationship between hypertension and incident AF was more pronounced in women than in men.20 Our findings diverge from those reports, suggesting a scenario where men are more sensitive to BP-related AF risk. There are several possible explanations for discrepancy. First, the onset of AF is, on average, 10 years later in women than in men.21 Therefore, given exposure to the same BP loading, the effect may manifest later in women than in men. Second, estrogen may have a beneficial effect on AF development in women.22 Because 65% of the female population included in our study was < 54 years old (Table 1). The potential antiarrhythmic effects of estrogen include prolongation of atrial action potential duration, reduction of cardiac fibrosis, and favorable effects on diastolic function.2122 Third, given that baseline risk factors for AF were significantly lower in women than in men, we hypothesized that the influence of elevated SBP loading would be less evident in women when considered in conjunction with other risk factors. Finally, in our study, follow-up was conducted with participants in their 50s and 60s (median age: approximately 58 years, as of 2019). Thus, the full impact of a prolonged high BP burden on possible AF manifestation in women only at later ages was beyond our observation window. If women develop AF in their 70s or 80s, a higher BP trajectory in their 50s might lead to AF after a longer latency. In the Tromsø study, where women with high BP did show increased AF, follow-up duration was longer and the cohort was older.17 Consequently, a potential explanation for the observed outcomes is that the female participants in our study may develop AF in subsequent years. Thus, the results may indicate a timing issue, rather than a true null effect, among women. This finding aligns with the established knowledge that the AF risk in women reaches the same level as in men at later ages. However, the exact mechanisms underlying these sex differences remain unclear, and further research is required. This study had some limitations. First, incident AF was identified solely based on ICD-10 codes from administrative claims, which may have introduced misclassification owing to diagnostic inaccuracies. Second, our analysis was limited to variables routinely collected from standard health examinations, precluding adjustments for clinical markers that require more detailed or costly evaluations. In particular, thyroid disease is a major risk factor for the development of AF, but its influence was not evaluated. Third, although the longitudinal design supports temporal inference, the observational nature of the study limits causal interpretation, and residual confounding from unmeasured variables cannot be excluded. Fourth, our primary focus was on SBP trajectory and VIM without incorporating other potentially informative BP parameters, such as diastolic BP, pulse pressure, or combined BP patterns. Future studies should explore these factors in relation to AF risk. Fifth, because this cohort included only individuals who underwent regular BP assessments across the five waves of biennial checkups over a 10-year period, many patients were excluded from the study. Moreover, the sex ratios were not equal. Men participated in economic activities more frequently than women did. This has led men to undergo regular health checkups, resulting in sex-dependent differences. This reflects real-world health-screening completion rates, with a higher participation rate in repeated health-screening among men. Sixth, owing to our strict inclusion criteria, only 45,663 individuals were included out of 554,320. This was primarily because we calculated trajectories and VIM based on SBP records collected periodically over the course of a decade-long screening program. Therefore, our study may have been subject to selection bias, potentially excluding less health-conscious or lower-risk individuals. Finally, the study population comprised exclusively Korean adults, which may limit the generalizability of our findings to other ethnic and geographical populations. In our large Korean cohort, long-term SBP trajectories and high SBP variability were associated with an increased risk of AF, predominantly in men. Men who maintained an elevated SBP for over a decade or experienced greater SBP fluctuations had a significantly higher incidence of AF independent of other risk factors. In contrast, in women, high SBP trajectories and variability did not have a significant impact on AF risk during midlife follow-up, suggesting potential sex-specific thresholds or timings for the contribution of hypertension to AF. These findings highlight the importance of sustained BP control and stability, especially in men, as part of AF risk reduction. This novel finding provides new insights into sex-specific differences in the pathogenesis of AF. Elevated BP remains a significant modifiable risk factor for cardiovascular disease, predominantly through chronic hemodynamic stress and the resultant cardiac remodeling. Long-term trajectories and variability in SBP have been determined as valuable prognostic markers for identifying individuals at an increased risk of developing AF. Integrating monitoring approaches based on SBP trajectory and variability into regular health screening could facilitate early detection and timely intervention in at-risk populations, potentially preventing AF onset. However, studies addressing effective strategies for mitigating this risk in individuals with adverse BP profiles are scarce. Future studies with prompt designs should confirm these sex differences in diverse populations and evaluate whether personalized BP targets or interventions such as stricter variability control can differentially benefit men and women in preventing AF. Additionally, sex- and age-specific AF prevention strategies should be investigated.