This analysis was performed based on the Korean nationwide claims database from the Korean Health Insurance Review Agency (HIRA) database. In South Korea, all citizens are subscribed to the medical insurance system, called the Korean National Health Insurance Service (NHIS) provided by the Korean government (11). Information on subscribers’ medical use is collected for NHIS operation, and information on medical use, which becomes insurance coverage, is submitted from health care providers. The submitted information is reviewed by the Korean HIRA, which is a quality control department that provides a review of the medical costs incurred. The Korean HIRA database contains all medical expenses claim data of the entire Korean population, including subscribers’ demographic information, diagnoses, examinations, prescriptions, and procedures for both inpatient and outpatient services (11, 12). Diagnoses were coded based on the International Classification of Disease, Tenth Revision, Clinical Modification (ICD-10-CM) codes (11, 12).

Using the Korean nationwide claims database, we included AF patients who underwent PCI from January 1, 2009 to February 28, 2019. Considering dynamic changes in antithrombotic therapy according to the period after receiving PCI, the index antithrombotic treatment was independently defined at different times after receiving PCI and we constructed two cohorts: 1- and 3-year after PCI (Figure 1). Cohort 1 consisted of patients who had just passed 1 year after PCI. Patients with AF who underwent PCI between September 1, 2009 and June 30, 2017, were firstly identified. Patients who died before 1-year after PCI and underwent repeated PCI before 1-year after PCI were excluded. Among these, the OAC monotherapy group and OAC plus SAPT group were defined by identifying prescriptions between 12 and 15 months from PCI (Figure 1A). Cohort 2 was defined as patients 3 years after PCI. Patients with AF who underwent PCI between September 1, 2009 and June 30, 2015, were included. Similar to cohort 1, patients who died before 3-year after PCI and underwent repeated PCI before 3-year after PCI were excluded. OAC monotherapy group and OAC plus SAPT group were identified by the prescription between 36 and 39 months from PCI (Figure 1B).

The study design was approved by the Institutional Review Board of the Seoul National University Hospital (E-1911-052-1078). The review board waived informed consent since each patient is de-identified and encrypted in the HIRA database to ensure patient privacy.

Subjects’ age, sex, comorbidities including hypertension, diabetes, dyslipidemia, heart failure, prior myocardial infarction, peripheral artery disease, prior ischemic stroke/transient ischemic attack/systemic embolism, prior intracranial hemorrhage, prior gastrointestinal bleeding, renal disease, and liver disease were ascertained by the prespecified operational definitions summarized in Supplementary Table 1 (13, 14). Concomitant medications include renin-angiotensin-aldosterone system inhibitors, beta-blockers, calcium channel blockers, loop diuretics, statins, non-steroidal anti-inflammatory drugs, and proton-pump inhibitors were ascertained based on the prescription records. The type of OAC [warfarin or direct oral anticoagulant (DOAC) including rivaroxaban, dabigatran, apixaban, and edoxaban], dose of DOAC, type of antiplatelet agents among aspirin, clopidogrel, prasugrel or ticagrelor were also identified. CHA₂DS₂-VASc score and modified HAS-BLED score were calculated by the operational definitions of comorbidities and medical history including concomitant medication (Supplementary Table 1) (13, 14).

During the follow-up period, composites of ischemic stroke and myocardial infarction occurrence were identified for effectiveness evaluation. For safety evaluation, major bleeding was defined as a composite of intracranial hemorrhage, gastrointestinal bleeding, and extracranial/unclassified major bleeding. We identified the major bleeding that occurred during the follow-up period. To assess net clinical benefit, composite clinical outcomes of ischemic stroke, myocardial infarction, and major bleeding were ascertained. Furthermore, we reported each component of effectiveness and safety outcome as follows: ischemic stroke, myocardial infarction, intracranial hemorrhage, gastrointestinal bleeding, and gastrointestinal bleeding requiring transfusion. Clinical outcomes were defined by the ICD-10-CM codes and detailed definitions of clinical outcomes are summarized in Supplementary Table 1.

To evaluate the accuracy of the operational definitions of clinical outcomes including ischemic stroke, myocardial infarction, intracranial hemorrhage, and gastrointestinal bleeding, we conducted a validation study in a tertiary hospital with 200 randomly chosen patients with the relevant ICD-10-CM codes for each event (15). Patients’ medical records were reviewed by two physicians (JP and SK). The positive predictive values of the operational definitions were 91.2, 92, 95.1, and 91.7% for ischemic stroke, myocardial infarction, intracranial hemorrhage, and gastrointestinal bleeding (15). In each cohort, the index date was the first date of OAC monotherapy or OAC plus SAPT prescription. Patients were censored at the outcome events or the end of the study period (February 28, 2019), whichever came first.

Continuous variables are presented as mean (standard deviation) and median (interquartile ranges, IQR). Categorical variables are presented as number and percentage. For each clinical outcome, the crude incidence rate for each clinical outcome was estimated by dividing the number of incidents during the follow-up period by the number of 100 person-years at risk. Unadjusted hazard ratios (HRs) and 95% confidence intervals (CIs) were analyzed for estimation of the risk of clinical outcomes using the Cox proportional hazards regression models.

To compare the OAC monotherapy and OAC plus SAPT groups, propensity score methods were used (16). A logistic regression model with all baseline factors (except for DOAC dose) was used to determine the probability score for being in each treatment group. To balance baseline characteristics across the two treatment groups, time-to-event analyses were conducted using inverse probability of treatment weighting (IPTW) analysis with stabilized weights computed from the propensity score (17, 18). Following IPTW, the covariate balance between the two groups was determined using the absolute standardized difference (ASD) (19). In a covariate, an ASD of ≤0.1 (10%) indicated that the two groups were well-balanced, with a negligible difference. The weighted number of events throughout the follow-up period was divided by 100 person-years at risk to calculate the weighted incidence rates. Survival analysis with the Kaplan-Meier method (log-rank test) and weighted Cox proportional hazards regression models with IPTW were used to determine the risk of clinical outcomes for OAC monotherapy and OAC plus SAPT (reference).

To provide complementary results, we conducted multivariable Cox analyses for a sensitivity analysis. Age, sex, hypertension, diabetes mellitus, dyslipidemia, heart failure, prior myocardial infarction, peripheral artery disease, prior stroke/transient ischemic attack/systemic embolism, prior intracranial hemorrhage, prior gastrointestinal bleeding, renal disease, liver disease, CHA₂DS₂-VASc score, modified HAS-BLED score, and OAC type (warfarin or DOAC) were included for the multivariable-adjusted Cox analyses.

SAS software, version 9.3 (SAS Institute, Cary, NC, United States), was used for all statistical analyses, and a two-tailed p-value of 0.05 was considered statistically significant.

A median follow-up duration of cohort 1 was 2.3 (IQR, 1.2–4.2) years. Crude incidence rates of clinical outcomes and unadjusted HRs for clinical outcomes are presented in Supplementary Table 2. Figure 2A showed weighted cumulative incidence curves of effectiveness, safety, and composite clinical outcomes of cohort 1. Weighted incidence rates and weighted HRs are presented in Figure 3A. After IPTW, OAC monotherapy and OAC plus SAPT showed comparable risks for a composite of ischemic stroke and myocardial infarction, major bleeding, and composite clinical outcomes (Figures 2A, 3A). OAC monotherapy and OAC plus SAPT did not show any significant differences for the individual components of the effectiveness and safety outcomes.

A median follow-up duration of cohort 2 was 2.5 (IQR, 1.3–4.2) years. Crude incidence rates of clinical outcomes and unadjusted HRs for clinical outcomes are presented in Supplementary Table 2. Figure 2B showed weighted cumulative incidence curves of effectiveness, safety, and composite clinical outcomes of cohort 2. Weighted incidence rates and weighted HRs are presented in Figure 3B. In cohort 2 with 3-year after PCI, OAC monotherapy and OAC plus SAPT showed a comparable risk for a composite of ischemic stroke and myocardial infarction, however, OAC monotherapy was associated with a lower risk of composite clinical outcomes (HR 0.762, 95% CI 0.607–0.950), mainly driven by a reduction of major bleeding risk (HR 0.498, 95% CI 0.345–0.701) compared to OAC plus SAPT (Figures 2B, 3B).

For each component of effectiveness and safety outcomes, OAC monotherapy and OAC plus SAPT group showed comparable risks for both ischemic stroke and myocardial infarction (Figure 3B). OAC monotherapy was associated with lower risks of intracranial hemorrhage, gastrointestinal bleeding, and gastrointestinal bleeding requiring transfusion than OAC plus SAPT (Figure 3B).

Multivariable Cox analyses showed consistent results with the IPTW analyses in two cohorts (Supplementary Table 2).

First, there is a possibility of residual confounding, although we ascertained available variables and matched the balance between the two treatment groups. Among possible confounders, these data did not include information about the characteristics and numbers of coronary stents, the complexity of PCI procedure, and the presence of remaining significant coronary lesions. Second, this study is an observational study, which would include more comprehensive patients than RCTs in which patients are highly selected, but patients who died within 1 or 3 years or who received repeated PCI were excluded from the study design. However, if a physician considers prescribing patients without additional coronary events for several years after PCI, our data can be applied practically. Third, OAC monotherapy and OAC plus SAPT do not represent the majority of prescriptions in AF patients with stable CAD in Korea, who are often prescribed with antiplatelet agents only (22, 23). Therefore, the number of study subjects is limited, and it should be considered when interpreting the results that patients who received OAC prescriptions in real-world practice were selected by physicians. Fourth, the Korean HIRA database did not include laboratory findings such as serum creatinine. Therefore, to indirectly measure renal dysfunction, we included “renal diseases” as one of the baseline covariates defined using the operational definition adopted in previous observational studies based on the claims database (14, 22, 23). Fifth, among DOAC users in OAC plus SAPT group, a higher proportion of patients were prescribed reduced dose DOAC than those in the OAC monotherapy group. In previous observational studies and even in the RCT (6, 9), reduced dose DOAC was preferred in OAC plus SAPT group. In this dataset, patients’ body weight and creatinine clearance were not available, thus, DOAC dosing adherence could not be evaluated. Notwithstanding the higher proportion of reduced dose DOAC in the OAC plus SAPT group than in the OAC monotherapy group, a combination of OAC and SAPT still showed a higher risk of bleeding than OAC monotherapy. Sixth, two types of antiplatelet agents (aspirin and clopidogrel) were prescribed for the most of patients in the OAC plus SAPT group. Although which antiplatelet agents are better than others also can be an important question for clinical practice, the primary objective of this study was the comparison between OAC and OAC plus SAPT in patients with AF and stable CAD. The number of the study population was not sufficient to explore the better antiplatelet type or the better OAC type for these populations. Further clinical or observational studies are needed to answer this question.