In the Korean NHIS data, 1,034,099 patients with prevalent AF who were aged ≥18 years were identified during the period from January 1, 2002 to December 31, 2016. The following were exclusion criteria: 1) those who ever received treatment with OAC before AF diagnosis (n=115,997), to exclude patients who had taken OAC for reasons other than AF such as venous thrombosis and 2) those with any of following conditions such as mitral valve stenosis and prosthetic valve replacement (ICD-10 codes I050, I052, I342; n=29,562). Finally 888,540 patients were included in the analysis (Figure 1). The annual prevalence of AF was calculated by dividing the number of AF patients of each year with exception for AF patients who died in previous year by the number of total Korean residents of that year. The annual incidence of AF was the number of incident cases of AF divided by the number of person-years at risk among all Korean residents of that year who had never been diagnosed as AF. AF was identified with ICD-10 codes; I48 (AF and atrial flutter), I48.0 (AF), and I48.1 (atrial flutter). To ensure accuracy, diagnosis was established based on one inpatient or 2 outpatient records of ICD-10 codes in the database.13)14) To evaluate the accuracy of our definition of AF, we conducted a validation study in 2 hospitals with 628 randomly chosen patients with the ICD-10 code I48. The patients were ascertained to have AF if it was documented by electrocardiogram examinations. The positive predictive value was found to be 94.1%.

Baseline comorbidities were identified from medical claims according to ICD-10 codes and prescription codes and all comorbidities were established based on one inpatient or 2 outpatient records of ICD-10 codes in the database, similar to previous studies based with NHIS cohort. Hypertension, diabetes mellitus, heart failure, peripheral arterial disease, a history of myocardial infarction, a history of stroke and/or transient ischemic attack (TIA), chronic obstructive pulmonary disease (COPD), chronic kidney disease (CKD), and end-stage renal disease (ESRD) were all assessed. The CHA₂DS₂-VASc score (2 points each for age 75 years or older and previous TIA or stroke; and 1 point each for heart failure, hypertension, age 65 years or older, diabetes mellitus, vascular disease, and female sex) for each subject was estimated at the end of the screening period, and every year during the follow-up period. Definitions of comorbidities are presented in Supplementary Table 1. Economic status was categorized into 3 groups based on the total amount of national health insurance premiums paid by the insured individual in each year, which is proportional to the individual's income: low, intermediate, and high economic status.15)

The primary outcome was prescription of any OAC (warfarin, dabigatran, rivaroxaban, or apixaban). The rates of OAC use calculated by the continued use of OAC in each period with considering prescription date and amount. The difference in the rates of OAC use according to the different criteria was presented in Supplementary Figure 1. Secondary outcomes included prescription of any NOAC (dabigatran, rivaroxaban, or apixaban) and of the individual NOACs. Prescription data were obtained from NHIS outpatient encounter documentation. For patients with multiple encounters, the last encounter was used in the analysis.

In Korea, NOACs were partially and fully reimbursed by the national insurance system in January 2013 and July 2015, respectively. We then examined temporal trends in the rates of use of any OAC, warfarin, and NOACs. For each month during the study period, we determined the proportion of patients who received therapeutic OAC with warfarin, dabigatran, rivaroxaban, or apixaban. If patients had >1 visit per month, the last study visit from each patient in each month was used for the analysis. A Cochrane-Armitage test for trend analysis was then performed to evaluate for changes in rates of use of any OAC over time.

We then examined the association between patient-level variables and the use of OAC. We used separate hierarchical, multivariable logistic regression models to better understand the individual contributions of patient factors to OAC use. Patient-level variables for this model included demographics (age and sex), comorbid conditions and risk factors (hypertension, diabetes mellitus, dyslipidemia, heart failure, previous stroke or TIA, vascular diseases, CKD or ESRD, COPD, and the use of antiplatelet agents). We also calculated the total CHA₂DS₂-VASc score.

To assess the effect of practice level on OAC use, we divided medical institutions by scale as follows: tertiary referral hospital, secondary care hospital, primary care hospital, and other healthcare facilities such as public health center, nursing hospital, etc. Then we compared the changes in the utilization rates of any OAC, warfarin, and NOACs within one year after the AF diagnosis of each practice level by year, among patients with more than 2 points of CHA₂DS₂-VASc score. Statistical analyses were performed using SPSS version 23.0 statistical package (SPSS Inc., Chicago, IL, USA).

Patients were categorized by reimbursement, and their characteristics were compared (Table 1). In general, patients after full reimbursement of NOAC had higher rates of comorbidities, and higher CHA₂DS₂-VASc scores.

In all population, overall OAC use increased from 13.2% to 23.4% (p for trend <0.001), and NOAC use increased from 0% to 14.6% (p for trend <0.001). OAC use increased from 13.2% to 18.0% before the full reimbursement of NOAC (p for trend <0.001), and further increased to 23.4% after the full reimbursement of NOAC (p for trend <0.001). Compared with pre-reimbursement (0.48%), the annual increase of OAC was significantly higher after partial (1.16%, p<0.001) and full reimbursement of OAC (3.72%, p<0.001) (Figure 2A). In the subgroup of AF patients with CHA₂DS₂-VASc score ≥2, the rate of use of overall OAC increased from 14.6% in 2010 to 28.1% in 2016 (p for trend <0.001). The annual increase of OAC was 0.60%, 1.38% and 4.85% at before, after partial and full reimbursement of NOACs. The annual increase of OAC significantly increased after partial (p<0.001) and full reimbursement of OAC (p<0.001) (Figure 2B).

Warfarin use also increased from 13.0% to 15.6% before the full reimbursement of NOAC (p for trend <0.001), but finally decreased to 8.8% during the study period. Any NOAC use increased from 0% to 14.6% (p for trend <0.001).

In all patients, patient characteristics associated with OAC use are shown in Figure 3A. In the multivariate models, patient factors associated with lower rates of use of OAC included CKD or ESRD, COPD, vascular disease, use of antiplatelet agents, female sex and diabetes mellitus. Patient factors associated with higher rates of use of OAC included heart failure, hypertension, previous stroke or TIA, dyslipidemia and age older than 65 years. Partial (adjusted odds ratio [OR], 1.58; 95% confidence interval [CI], 1.56–1.60) and full reimbursement of NOAC (adjusted OR, 2.04; 95% CI, 2.00–2.09) were independently associated with higher OAC use in total population.

In patients with CHA₂DS₂-VASc score ≥2, patient factors associated with lower rates of use of any OAC included use of antiplatelet agents, CKD or ESRD, COPD, vascular disease, female sex, diabetes mellitus and hypertension (Figure 3B). Higher rates of use of any OAC included heart failure, previous stroke or TIA, dyslipidemia and age older than 65 years. Partial (adjusted OR, 1.60; 95% CI, 1.58–1.63) and full reimbursement of NOACs (adjusted OR, 2.10; 95% CI, 2.04–2.15) was independently associated with higher OAC use.

Significant practice-level variation was evident in the use of OAC. Figure 4 illustrates the temporal trends in OAC initiation within 1 year after AF diagnosis according to the practice level. The use of any OAC in eligible patients ranged from 4.9% to 22.8% in 2010, and from 7.2% to 44.0% in 2016 according to the practice level. The difference of OAC usage between tertiary referral hospital and nursing or public health centers increased from 17.9% in 2010 to 36.8% in 2016 (Figure 4A). The rate of OAC use significantly increased from 22.8% to 44.0% in tertiary referral hospitals (p<0.001), from 17.9% to 36.7% in secondary care hospitals (p<0.001), and from 11.0% to 25.2% at primary care hospitals (p<0.001). OAC use remained without improvement in other healthcare facilities such as nursing hospital or public health center.

Warfarin use increased from 2010 to 2014, before the full reimbursement of NOAC, but finally decreased in 2016 compared to 2010 at all practice-levels (Figure 4B). In 2016, the rate of NOAC use ranged from 5.5% to 37.2% according to the practice level (Figure 4C). NOAC use increased dramatically from 0% to 37.2% at the tertiary referral hospital, 31.8% at secondary care hospitals, 20.9% at primary care hospitals, and 5.5% at other healthcare facilities (for all, p for trend <0.001). Practice level variation in the prescription rate of individual NOAC among AF patients along time elapse is presented in Supplementary Figure 2.

In addition, we further analyzed OAC usage according to the patients' economic status (Figure 5). The increase in NOAC use was slightly higher in patients with high economic status compared to the others, but the use of NOAC showed an even trend across the 3 economic status groups, without significant difference.