The flowchart of the study population is shown in Figure 1. From the database, we extracted the data of patients diagnosed with AF during 2009 to 2016. We excluded the following populations: (1) patients with prevalent ESRD (n = 2091); (2) patients with mitral stenosis or prosthetic heart valves (n = 15,207); (3) patients who had no available health checkups within 2 years from the diagnosis of AF (n = 366,197); (4) patients aged <20 years (n = 29); (5) patients with missing values for study covariates (n = 1189); and (6) patients with a follow-up period <1 year (n = 4137). Consequently, 202,434 patients with AF without prior ESRD were investigated.

We defined the metabolic score range from 0 to 5 points, such that a patient received 1 point if he/she met each diagnostic criteria for metabolic syndrome. The diagnostic criteria of metabolic syndrome were defined based on an international guideline (9), with the adoption of the criteria for increased waist circumference according to the Korean Society for the Study of Obesity (12). The five diagnostic criteria are summarized in Table 1. The study population was then categorized into six groups (MS₀ to MS₅) according to their metabolic scores (0–5).

The study covariates were measured using data from the NHIS database. Individual covariates were obtained at the index health checkup, and Supplementary Table S1 summarizes their detailed definitions. General information regarding the population's characteristics, including age, sex, height, body weight, CHA₂DS₂-VASc scores, alcohol consumption (yes/no), smoking (yes/no), regular exercise (yes/no), and low-income status (yes/no) was collected. Comorbidities were investigated using established diagnostic codes, including diabetes mellitus, ischemic heart disease, heart failure, ischemic stroke, peripheral artery disease, dyslipidemia, diabetes mellitus, chronic obstructive pulmonary disease, CKD, and any malignancy. Diagnostic codes were encoded according to the International Classification of Diseases, Tenth Revision, Clinical Modification. Data about concomitant medication, including oral anticoagulants (warfarin or direct oral anticoagulants), antiplatelet agents (aspirin or P₂Y₁₂ inhibitors), antidiabetic drugs (sulfonylurea, meglitinide, metformin, thiazolidinedione, alpha-glucosidase inhibitors, dipeptidyl peptidase−4 inhibitors, and insulin), antihypertensive drugs (angiotensin receptor blockers, angiotensin-converting enzyme inhibitors, beta-blockers, calcium channel blockers, and diuretics), and statins were obtained from the claims database. Data for blood pressure, fasting blood glucose, total cholesterol, high- and low-density lipoprotein cholesterol (HDL-C, LDL-C), triglyceride, serum creatinine, and estimated glomerular filtration rate (eGFR) were obtained from the health checkup database.

The primary outcome was incident ESRD, which was defined as having a diagnostic code (N18.5 or Z49) with hemodialysis or peritoneal dialysis ≥2 times during the follow-up period. Individuals were right-censored when the primary outcome occurred and were followed up from the index health checkup to December 31st, 2018.

Baseline characteristics were compared across the six groups (MS₀–MS₅) using a one-way analysis of variance or Kruskal–Wallis H test according to the type of covariate. Survival analysis was performed using the Kaplan–Meier method, and a log-rank test was used to compare survival across the six groups. Crude incidence rates (IRs) of ESRD were calculated in 1000 person-years. The risk of incident ESRD was estimated by multivariate Cox regression analyses and reported as adjusted hazard ratios (HRs) with 95% confidence intervals (CIs). The final model used covariates, including age, sex, body mass index, low-income status, health habits (including alcohol consumption, smoking, and regular exercise), comorbidities (including ischemic heart disease, heart failure, stroke, peripheral artery disease, chronic obstructive pulmonary disease, and any malignancy), concomitant drug use (including oral anticoagulants and antiplatelet agents), the five metrics used in the definition of metabolic syndrome (including waist circumference, fasting blood glucose, blood pressure, triglyceride, and HDL-C), and renal function (eGFR). Subgroup analyses were performed for sex, strata of CHA₂DS₂-VASc scores (0–1 vs. ≥2), and strata of eGFR (≥60 vs. <60 ml/kg/1.73 m²) and concurrent use of oral anticoagulant (OAC) (any vs. none). All statistical analyses were performed using SAS version 9.3 (SAS Institute, Cary, NC, USA). Two-sided P < .05 were used to reject the null hypothesis.

To investigate the impact of each diagnostic criterion of metabolic syndrome on incident ESRD, the study population was divided according to the presence or absence of each diagnostic criterion of metabolic syndrome. The risk of incident ESRD was compared across the five groups based on patients meeting each diagnostic criterion. The impact of systolic blood pressure and fasting blood glucose levels on the incident ESRD risk was visualized using cubic spline curves.

We also calculated the area under the receiver operating characteristics curves (AUROCs) of metabolic scores and five components of metabolic syndrome (waist circumference, fasting blood glucose, systolic blood pressure, HDL-C, and triglyceride) to predict incident ESRD at 1-year.

A total of 11 statistical models were created using different sets of covariates for model adjustment, and their results were compared with those of the final model. A complete list of the statistical models is presented in Supplementary Table S2. We also compared the results of the final model with those of the other three models each with different covariates for renal function: eGFR, presence of CKD diagnosis (N18), and presence of decreased eGFR (<60 ml/kg/1.73 m²) for Model 8, 9, and 10, respectively.

In total, 202,434 patients with AF without prior ESRD were included in the analysis. The study population was divided into six groups (MS₀–MS₅) according to their metabolic scores (0–5), with n = 12,747 (6.3%), 31,059 (15.3%), 40,361 (19.9%), 48,068 (23.7%), 46,630 (23.0%), and 23,569 (11.6%) in each subgroup, respectively. The population's mean age, male proportion, and CHA₂DS₂-VASc score were 63.5 ± 12.1 years, 49.5%, and 2.8 ± 1.6, respectively.

As the metabolic score increased, the population's age, body mass index, and comorbidities (except malignancy) also increased (Table 2); mean ages increased from 52.5 years (MS₀) to 66.3 years (MS₅); body mass index from 22.0 kg/m² (MS₀) to 27.7 kg/m² (MS₅); all P < .001. Furthermore, the concomitant medication (oral anticoagulants, antiplatelet agents, antidiabetic drugs, antihypertensive drugs, and statins) also increased in relation to metabolic score (from MS₀ to MS₅, Table 2); all P < .001. Among the laboratory test results, blood pressure, fasting blood glucose, triglyceride, and serum creatinine increased as the metabolic score increased, while total cholesterol, HDL-C, LDL-C, and eGFR decreased (Table 2); all P < .001.

During a median follow-up of 3.5 (interquartile ranges, 1.7–5.6) years, the crude incidence rate of ESRD among AF patients gradually increased for higher metabolic scores; 0.16, 0.45, 0.70, 1.13, 1.87, and 2.48 per 1000 person-years for MS₀–MS₅, respectively. There was a significant difference in ESRD-free survival across the five groups (log-rank P < .001), although there was a comparable result between MS₀, MS₁, and MS₂ (pairwise log-rank P ≥ 0.05) (Figure 2).

Metabolic syndrome was associated with a 2.9-fold increase in the risk of ESRD [adjusted HR 2.94 (95% CI, 1.43–6.06)]. After multivariate adjustment, the final model showed a trend for higher risk of incident ESRD in relation to higher metabolic scores (P-for-trend <.001) (Figure 3). Compared to MS₀, all others (except MS₁) were associated with significantly increased risks of incident ESRD (adjusted HR, 1.60 [95% CI 0.78–3.48], 2.08 [1.01–4.31], 2.94 [1.43–6.06], 3.71 [1.80–7.66], and 4.82 [2.29–10.15] for MS₁–MS₅, respectively (Figure 3).

Among the five metabolic syndrome components, the increased risk of ESRD due to metabolic syndrome was primarily driven by elevated blood pressure; adjusted HRs (95% CI) in decreasing order, 2.20 (1.60–3.03), 1.66 (1.42–1.95), 1.61 (1.36–1.91), and 1.19 (1.02–1.40) for elevated blood pressure, impaired fasting blood glucose, low HDL-C, and elevated triglycerides, respectively (Supplementary Figure S1). In contrast, increased waist circumference did not significantly impact the risk of ESRD [adjusted HR 1.12 (95% CI, 0.90–1.38)]. The cubic spline curves showed that the systolic blood pressure and fasting blood glucose thresholds for increased ESRD risks were 125 mmHg and 113 mg/dl, respectively (Figure 4).

There was no significant interaction for strata of CHA₂DS₂-VASc scores (0–1 vs. ≥2) and sex (P-for-interaction = .966 and.838, respectively) (Supplementary Table S3). Compared to the subgroup with preserved eGFR (≥60 ml/kg/1.73 m²), the trend of increased risk for ESRD in relation to higher metabolic scores was accentuated in the subgroup with decreased eGFR (<60 ml/kg/1.73 m²) (Supplementary Table S3). For the subgroup without OAC use, there was a trend of increasing risks of incident ESRD with higher metabolic scores (Supplementary Table S3). Conversely, for the subgroup with OAC use, no definitive trend was observed. However, there was no significant interaction (P-for-interaction = .115).

The main result was compared with different multivariate Cox regression analyses (Models 1–10). Regardless of the statistical models, there was a consistent trend of increasing risk of ESRD in relation to higher metabolic scores. However, the magnitudes of HRs decreased as more covariates were adjusted for (all P-for-trend < .001) (Supplementary Table S4). Consistent results were observed across the models regardless of the covariates representing renal function: eGFR, the presence of CKD diagnosis, or the presence of decreased eGFR (<60 ml/kg/1.73 m²) for models 8, 9, and 10, respectively (Supplementary Table S4).

The AUROCs of metabolic scores and the five components of metabolic syndrome are presented in Table 3. Compared to the metabolic components, except for systolic blood pressure, metabolic scores showed a significantly higher AUROC (0.68 with a 95% CI of 0.65–0.72). Compared to the AUROC of systolic blood pressure, that of metabolic scores showed a higher value with marginal significance (AUROC = 0.68 [95% CI 0.65–0.72] vs. 0.64 [95% CI 0.61–0.68], P = .096).

Some limitations of this study need to be addressed. First, because this study was retrospective, a nationwide cohort is needed to ascertain the causal relationship between metabolic syndrome and incident ESRD. Second, the five metrics used for defining metabolic syndrome, especially blood pressure, may vary from time to time among patients with AF. Therefore, the reliability of the results may be a concern. Third, our results may not be applicable to the Western population because the definition of increased waist circumference was based on the Korean guideline (12). Fourth, although we observed consistent results across different multivariate Cox regression analyses, hidden confounders might have significantly affected the results. Fifth, the etiology of incident ESRD among patients is unknown in our study. We presume that most causes were hypertension or diabetic nephropathy, as they are the two major risk factors for ESRD. Sixth, there could be the potential influence of warfarin use on our results, considering its known impact on vascular calcification and renal function decline. According to Table 2, the proportion of warfarin use increased from 10.3% in MS₀ to 24.3% in MS₅. If the increased use of warfarin had a significant biasing effect on our results, we would expect to observe divergent outcomes between Model 4 and Model 3, because Model 4 incorporated the covariates from Model 3, along with the inclusion of oral anticoagulants and antiplatelet agents. However, Supplementary Table S4 demonstrates that both models yield comparable results. Based on these findings, we concluded that the potential bias arising from the increased use of warfarin might not be significant in our analysis. Seventh, this study could not analyze temporal trends in the associations between ESRD risks and metabolic scores. Metabolic scores could be dynamic and vary as patients age or receive medical management. However, this study utilized cross-sectional health check-up data, and therefore the dataset did not contain serial health check-up data for the study population. A further study is warranted to investigate the impact of temporal changes in metabolic status on the risk of ESRD. Eighth, the difference in the classes of antihypertensive and antidiabetic medications across groups could be potential bias in our study. To further investigate this issue, we analyzed the use of different drug classes among the groups, as presented in Supplementary Table S5. Our analysis revealed that the most used antihypertensive drug class was angiotensin receptor blockers, while the least used drug class was angiotensin-converting enzyme inhibitors, with similar patterns observed across the groups (excluding MS₀, where the use of antihypertensive drugs would not be expected). Furthermore, regardless of the metabolic score groups, the two most used antidiabetic drugs were metformin and sulfonylurea. Based on this analysis, it appears that the distribution of drug classes was similar among the different metabolic score groups. Ninth, although the use of OAC may prevent thromboembolic events, such as renal infarction, and potentially reduce the risk of incident ESRD, our study did not observe any significant interaction (Supplementary Table S3; P-for-interaction = .115). This lack of significant interaction could be attributed to the relatively low number of events among the subgroup with OAC use. Finally, this study has the potential for selection bias because it excluded patients who did not have health check-ups within two years of AF diagnosis. Furthermore, patients with longer AF durations may have results different from those of this study.