The dataset included patients with T2DM who received routine care at the affiliated hospital of Qingdao University between January 2014 and December 2023. We extracted anthropometric, demographic, laboratory, diagnostic, and prescription data from the electronic medical records (EMRs) of these patients. The study design and protocol received written approval from the Institutional Review Boards of the affiliated hospital of Qingdao University (QYFY WZLL 28911). Individual informed consent was not required as we utilized anonymized data sourced from EMRs.

The definition of T2DM in this study was based on the SUPREME-DM criteria (14), which include the following: (a) at least one International Classification of Disease, Tenth Revision, Clinical Modification (ICD-10-CM) for T2DM related to inpatient stays; (b) two or more ICD codes associated with outpatient visits on separate days within a two-year period; and (c) a combination of two or more of the following variables related to outpatient visits on different days within two years: ICD codes, fasting glucose level ≥7.0 mmol/L, 2-hour glucose level ≥11.1 mmol/L, random glucose level ≥11.1 mmol/L, HbA1c ≥ 6.5%, and prescription of an glucose-lowering medication. A total of 13,958 patients with T2DM were identified. Patients were further excluded if they were older than 80 years, had a diagnosis of atrial fibrillation or heart failure at baseline, had incomplete data at baseline, or had severe liver or kidney dysfunction. Additionally, patients who were lost to follow-up were also excluded from the final analysis. After applying these criteria, a total of 9,930 patients remained eligible for the final analysis ( Figure 1 ).

We compiled an analytic database using the unique Chinese Personal Identification Numbers. The database included patients’ information such as date of birth, sex, age at diabetes diagnosis, smoking status, and medication use, including antihypertensive drugs, glucose-lowering drugs, lipid-lowering drugs, and antiplatelet or anticoagulant drugs. Additionally, patients’ height, weight, blood pressure, plasma glucose, C-peptide, lipid profiles including triglycerides (TG), total cholesterol (TC), low-density lipoprotein (LDL) cholesterol, and high-density lipoprotein (HDL) cholesterol, glycated hemoglobin A1c (HbA1c), serum creatinine, and C-reactive protein levels were measured or collected either using standardized methods during inpatient stays or self-report questionnaires during outpatient visits. Body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared (kg/m²). The estimated glomerular filtration rate (eGFR) was determined using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation. Information on diabetic complications and comorbidities was extracted from the database and defined according to the International Classification of Diseases, 10th Revision (ICD-10) codes. Patients with a history of atrial fibrillation or heart failure were also identified by ICD-10 codes.

The primary exposure in this study was the calculated remnant cholesterol. The equation for calculating remnant cholesterol is as follows: RC = TC - LDL-C - HDL-C.

The primary outcome of this study was the incidence of atrial fibrillation. Atrial fibrillation was identified either using the ICD-10-CM codes I48.x or based on electrocardiogram (EKG) findings, characterized by the absence of discrete P waves and the presence of irregularly irregular R-R intervals on a 12-lead EKG or a 24-hour Holter monitoring. The incident atrial fibrillation events included both paroxysmal atrial fibrillation and persistent atrial fibrillation.

Data were presented differently based on their characteristics: normally distributed data as mean ± standard deviation, skewed data as median with interquartile ranges, and categorical variables as frequencies and percentages. For intergroup comparisons, Student’s t-tests were utilized for normally distributed data, Wilcoxon rank-sum tests for skewed data, and χ2 tests for categorical variables. Fisher’s exact test for categorical variables were further used to verify the results of χ2 tests. Hazard ratios (HRs) for the incidence of atrial fibrillation were estimated using Cox proportional hazards regression, considering RC either as a categorical variable or a continuous variable. RC levels and other categorical variables were treated as dummy variables in the models, with linearity significance tested by assigning ordinal numeric values to each dummy variable. The proportional hazards assumption for the Cox model was evaluated using graphical methods and models incorporating time-by-covariate interactions, and overall, these assumptions were found to be appropriate. Initially, all analyses were performed with adjustments for age and sex, followed by additional adjustments for factors including age, sex, systolic blood pressure, TG, HbA1c, smoking status, eGFR, antiplatelet or anticoagulant, lipid-lowering medications, antihypertensive medications, glucose-lowering medications. Kaplan–Meier analysis was employed to generate survival curves, and event-free survival rates among patients with varying RC levels were compared using the log-rank test. Stratified analyses were conducted based on various patient characteristics, including age, sex, BMI, HbA1c levels, estimated glomerular filtration rate (eGFR), smoking status, and baseline medication use. A P-value of less than 0.05 was considered statistically significant. All statistical analyses were executed using R software version 4.4.1 (R Foundation for Statistical Computing, Vienna, Austria).

The baseline characteristics of participants according to different RC levels are summarized in Table 1 . Participants were divided into four quartiles based on RC levels: Q1 (≤0.47 mmol/L), Q2 (0.48–0.64 mmol/L), Q3 (0.65–0.88 mmol/L), and Q4 (>0.88 mmol/L). The mean age was similar across Q1 to Q3, ranging from 66.1 to 66.4 years, but was slightly younger in Q4 at 63.4 years (between-group P<0.05). The proportion of males increased with higher RC levels, from 42.8% in Q1 to 48.8% in Q4 (P for trend < 0.05). Both systolic and diastolic blood pressures were consistent across all quartiles, with systolic around 133 mmHg and diastolic around 76–77 mmHg. BMI increased with higher RC levels, from 23.2 ± 2.41 kg/m² in Q1 to 25.5 ± 4.10 kg/m² in Q4 (P for trend < 0.05). TC, LDL-C, and TG all increased with higher RC levels, whereas HDL-C decreased (all P for trend < 0.05). The eGFR showed a slight decrease from Q1 (94.0 ± 24.9 mL/min/1.73 m²) to Q4 (91.2 ± 24.8 mL/min/1.73 m²) (P for trend < 0.05). HbA1c levels increased from 7.1 ± 1.5% in Q1 to 7.7 ± 1.7% in Q4 (P for trend < 0.05). The percentage of current smokers varied slightly among the quartiles, with the lowest in Q2 (7.3%) and the highest in Q1 (10.4%). The use of antihypertensive medications increased from 71.4% in Q1 to 77.0% in Q4, lipid-lowering medications from 52.9% in Q1 to 66.8% in Q4, and glucose-lowering medications from 67.5% in Q1 to 82.6% in Q4 (all P for trend < 0.05). The use of antiplatelet or anticoagulant medications remained relatively stable across the quartiles.

During a mean follow up of 6.12 ± 2.75 years, a total of 1,028 AF events occurred. The HRs for the incidence of atrial fibrillation by different RC levels at baseline are presented in Table 2 . Compared to Q1, the age- and sex-adjusted HRs for coronary artery disease were 0.89 (95% CI: 0.80-1.28) in Q2, 0.88 (95% CI: 0.80-0.98) in Q3, and 0.76 (95% CI: 0.67-0.84) in Q4. Further Kaplan–Meier analysis confirmed the above findings ( Supplementary Figure S1 , Q4 had the lowest incidence of AF). As a continuous variable, RC had an HR of 0.77 (95% CI: 0.64-0.93). After adjusting for age, sex, systolic blood pressure, triglycerides, HbA1c, smoking status, eGFR, and the use of antiplatelet or anticoagulant, lipid-lowering, antihypertensive, and glucose-lowering medications, the multivariable-adjusted HRs were 0.92 (95% CI: 0.82-1.32) in Q2, 0.97 (95% CI: 0.88-1.08) in Q3, and 0.88 (95% CI: 0.79-0.98) in Q4. As a continuous variable, RC had an HR of 0.83 (95% CI: 0.66-0.99) ( Figure 2 ).

The association between baseline RC levels and AF was evaluated through subgroup analysis according to various baseline characteristics, as shown in Supplementary Table S1 . The inverse association between RC and incident AF events were more evident among patients aged 65 years or older, maple patients, patients with HbA1c < 7.0%, patients with eGFR < 60 mL/min/1.73 m², patients without smoking as well as patients with current use of antiplatelet or anticoagulant, lipid-lowering, antihypertensive and glucose-lowering medications. The P for interaction values indicated no significant interactions for age, sex, smoking status, antiplatelet or anticoagulant use, lipid-lowering medication use, antihypertensive medication use, or glucose-lowering medication use. There was a significant interaction for BMI (P for interaction = 0.001) and eGFR (P for interaction = 0.073).