ETNA-AF-China (Clinicaltrials.gov: NCT04747496) is a multicentre, prospective, observational study designed to evaluate the effectiveness and safety of edoxaban for stroke prevention in 89 sites across the Chinese Mainland, enrolling 5,001 patients with a total of 2-year follow-up. ETNA-AF-China included three predefined data cut-off points for analysis: baseline, 1-year outcomes, and final 2-year outcomes, based on scheduled follow-up data collection. Here, we present the first 1-year follow-up outcomes based on the interim data cut-off on 1 February 2024; data are subject to change as data collection was ongoing at the time of interim data cut-off. Final data cleaning for 2-year follow-up is ongoing. The detailed study design and rationale of the Global ETNA-AF programme, and baseline characteristics of ETNA-AF-China were previously published (De Caterina et al., 2019b; De Caterina et al., 2019a; Guo et al., 2024).

Patients with a history of NVAF who were prescribed edoxaban at enrolment or baseline (≤1 week after baseline) and did not simultaneously participate in an interventional trial were eligible upon providing written consent. No exclusion criteria were present to avoid any possible selection bias.

The primary objective of the study was to evaluate the real-world safety of edoxaban by evaluating bleeding events (including major, major or clinically relevant non-major [CRNM] bleeding according to the International Society on Thrombosis and Haemostasis’s [ISTH] definition) (Schulman and Kearon, 2005) intracranial haemorrhage [ICH], all-cause and CV mortality in patients with AF. Stroke (ischaemic, haemorrhagic, and unknown), SEEs, transient ischaemic attack (TIA), and myocardial infarction (MI) were evaluated as secondary outcomes.

The CHA₂DS₂-VASc and HAS-BLED scores were both centrally calculated based on the investigator-reported baseline clinical characteristics of patients. The DOAC score was calculated based on the development and validation by R. Aggarwal (Aggarwal et al., 2023). The ‘recommended dose’ for edoxaban treatment is 60 mg OD as labelled and 30 mg OD for those who met at least one of the following criteria: (1) moderate-severe renal impairment (CrCl ≤50 mL/min), (2) body weight ≤60 kg and (3), concomitant use of label-listed P-gp inhibitors. “Non-recommended dose” is referred to as not in line with label recommendation, that is, 60 mg was considered non-recommended if the patient fulfilled at least one dose reduction criterion, or 30 mg was non-recommended if the patient did not meet any dose reduction criterion.

Statistical analyses were performed on the full analysis set (FAS). FAS was defined as all patients with any documentation (such as post-baseline ADRs, AF-related clinical events, or hospitalisations) from at least one follow-up data collection or at final assessment (any information except “lost to follow-up”) from BAS.

Baseline analysis set (BAS) was defined as: (1) the patient with a confirmed history of NVAF by an electrical test (e.g., electrocardiogram, Holter monitoring, pacemaker, or implantable cardioverter-defibrillator). (2) The investigator confirmed that the patient was receiving treatment with edoxaban after providing informed consent. In addition, those patients who were judged by the investigator as not receiving edoxaban treatment after providing informed consent were additionally checked whether edoxaban initiation date was at least within 7 days after baseline visit date. (3) The patient not concurrently participating in any other interventional studies was included (Supplementary Figure S1).

The baseline characteristics were summarised descriptively as frequencies (“n” and percentage) and mean value ± standard deviation (SD) rounded to an integer. Patient characteristics between the 60 mg and 30 mg dose groups were compared using the chi-square test for the categorical data and the Mann–Whitney U test for the continuous data. The normality of continuous data was checked using the Shapiro test. Adherence to label-recommendation (recommended dose/non-recommended dose) was calculated by dose prescription at baseline.

Based on the reported events of the investigators, annualised event rates of the safety and effectiveness outcomes were presented. No preplanned central adjudication of this real-world study was carried out; all events were reported by qualified investigators. Annualised event rates were estimated by censoring those lost to follow-up or premature termination. Annual rates were based on time exposure of whole population regardless of the discontinuation of edoxaban or edoxaban dose. The annualised event rates were calculated as 100*[number of patients with at least one event within 2 years after baseline] divided by [the sum of all individual patient observation time]. Individual patient observation time was defined as time in years from baseline data collection point (i.e., reference start date) until first event date (within 2 years after baseline), or if patient completed/discontinued from the study without experiencing an event (within 2 years after baseline), until the date of premature study discontinuation, or death, or last available follow-up date, or censoring after 24 months from baseline data collection point whichever occurred first.

Kaplan–Meier curves were used to present the cumulative event rates of all stroke/SEE/TIA, major bleeding, and all-cause mortality. Two or more time-to-event data points of the curves were compared using the log-rank test. Univariable Cox regression analysis identified the hazard ratio (HR) and 95% confidence interval (CI) for recommended or non-recommended doses and 30 mg versus 60 mg related outcomes.

Multivariable Cox model was conducted using a stepwise regression method by the Akaike Information Criterion for all covariables while fixing edoxaban dosing (60 mg and 30 mg) and recommended/non-recommended doses. The prespecified variables included edoxaban dose, age category, gender, body mass index (BMI) category, creatinine clearance category, economic region, CHA₂DS₂-VASc score, modified HAS-BLED score, AF type, frailty, chronic obstructive pulmonary disease, prior TIA, hypertension, heart failure, diabetes, dyslipidaemia, valvular heart disease, prior stroke at baseline, and prior major bleeding.

Statistical analyses were performed using R Statistical Software (version 4.3.2, Boston, Massachusetts, United States). No adjustment for multiple testing was made; therefore, all P values must be interpreted in a purely exploratory or descriptive way.

The study was approved by the institutional review boards and independent Ethics Committees from Beijing Anzhen Hospital (No. 2019-058D) and all participating centres in compliance with the Declaration of Helsinki and Guidelines for Good Pharmacoepidemiology Practice (GPP) (Public Policy Committee, International Society of Pharmacoepidemiology, 2016). All participants provided written informed consent.

Of the overall 5,001 patients enrolled in the study, 124 patients were excluded because of either receiving unspecified doses, starting edoxaban more than 1 week after baseline, participating simultaneously in other interventional studies, or being lost to follow-up without any post-baseline documentation (Supplementary Figure S1). The baseline demographics and clinical characteristics of the FAS, consisting of 4,877 patients, with 2,650 patients (54.3%) initially received edoxaban 60 mg and 2,227 (45.7%) received 30 mg, are presented in Table 1. The overall mean (SD) patient age was 70.3 ± 9.5 years, and 2,780 patients (57.0%) were male. The majority of the patient population belonged to the Eastern economic region (49.9%), followed by the Middle (33.5%), Northeastern (8.6%), and Western economic regions (8.1%) in China. The mean (SD) body weight was 67.9 ± 12.5 kg, and the mean (SD) body mass index (BMI) was 25.0 ± 3.7 kg/m². The calculated mean (SD) CrCl by the Cockcroft–Gault formula was 71.0 ± 27.2 mL/min, and the mean (SD) estimated glomerular filtration rate (eGFR) by CrCl normalized to 1.73 m² of body surface area using the Du Bois and Du Bois formula (Michels et al., 2010) was 58.7 ± 19.8 mL/min/1.73 m². The calculated mean (SD) CHA₂DS₂-VASc and modified HAS-BLED scores were 2.9 ± 1.4 and 1.8 ± 1.0, respectively.

Most of the patients presented with paroxysmal (1780; 36.5%) and persistent AF (1,548; 32.5%), with only 17% and 13.7% of them having long-standing persistent and permanent AF, respectively. The majority of the patients (2,379; 93.6%) had a left ventricular ejection fraction of ≥40%. The extent of perceived frailty was minimal (309; 6.3%).

Patients receiving 30 mg edoxaban OD at baseline were older (mean 73.4 years vs. 67.6 years) had higher CHA₂DS₂-VASc score (3.3 vs. 2.6), CHA₂DS₂-VA score (3.1 vs. 2.6), modified HAS-BLED score (2.0 vs. 1.5), DOAC score (5.1 vs. 3.6), and more percentage of impaired renal function (CrCl <80 mL/min: 70.5% vs. 42.8%) than those receiving 60 mg (all P < 0.001). When compared with the 60 mg group, the 30 mg group had a higher proportion of patients with perceived frailty (9.9% vs. 3.4%; P < 0.001), higher proportion of HF (17.5% vs. 12.3%; P < 0.001), and valvular disease (6.6% vs. 4.5%, P = 0.001), whereas a smaller percentage of hypertension (70.5% vs. 75.8%, P < 0.001) and dyslipidaemia (22.5% vs. 27.3%, P < 0.001) (Supplementary Table S1; Supplementary Figure S2).

The overall adherence of edoxaban to the label recommendation was high (3,269/4,192 [with evaluable data]; 78.0%; P < 0.001), with a numerically higher adherence rate of recommended 60 mg (1790; 83.8%) than recommended 30 mg (1,479; 71.9%) (Table 1).

The baseline demographics and clinical characteristics of the patients based on the dosing recommendation in accordance with label, with 1790 patients (36.7%) initially received recommended 60 mg edoxaban, 578 (11.9%) received non-recommended 30 mg, 1,479 patients (30.3%) received recommended 30 mg, and 345 patients (7.1%) received non-recommended 60 mg, are presented in Table 2.

When compared with the patients receiving the recommended 60 mg, the patients receiving non-recommended 30 mg edoxaban OD were older (mean 71.4 years vs. 67.0 years), had a higher proportion of patients with perceived frailty (10.0% vs. 2.8%) and HF (19.9% vs. 14.0%), had higher CHA₂DS₂-VASc score (3.0 vs. 2.5), CHA₂DS₂-VA score (3.1 vs. 2.7), modified HAS-BLED score (1.9 vs. 1.5), DOAC score (4.4 vs. 3.5), and a greater percentage of impaired renal function (CrCl <80 mL/min: 65.2% vs. 48.6%; all P < 0.001).

Similarly, patients receiving the recommended 30-mg edoxaban OD at baseline were older (mean 74.2 years vs. 71.5 years; P < 0.001), had higher CHA₂DS₂-VASc score (3.4 vs. 3.1; P < 0.001), DOAC score (5.4 vs. 4.7; P < 0.001), and a greater percentage of impaired renal function (CrCl <80 mL/min: 80.1% vs. 75.1%; P = 0.002) than those receiving the non-recommended 60 mg (Table 2).

Overall, a majority of patients (60 mg: 1,938 [73.1%], 30 mg: 1,626 [73.0%]) completed the 1-year follow-up while continuing to use edoxaban (Table 3; Supplementary Figure S3). A lesser proportion of patients treated with edoxaban 30 mg OD initially at baseline switched to 60 mg OD (32; 1.4%) than from 60 mg at baseline to 30 mg OD (158; 6.0%).

In the 60 mg group, 277 (10.5%) patients suspended edoxaban and 380 patients (14.3%) discontinued the drug during 1-year follow-up. Similarly, in the 30 mg group, edoxaban was suspended in 218 patients (9.8%), whereas 397 patients (17.8%) discontinued. Within each group, the percentages of patients who discontinued edoxaban and immediately switched to another OAC were less than or ∼1% (60 mg: warfarin 16 [0.6%], dabigatran 9 [0.3%] and rivaroxaban 23 [0.9%]; 30 mg: warfarin 16 [0.7%], apixaban 1 [0.04%], dabigatran 4 [0.2%] and rivaroxaban 24 [1.1%]). Suspension or discontinuation of edoxaban was observed in 37 patients (1.4%) in the 60 mg group and in 39 patients (1.8%) in the 30 mg group.

The overall annualised event rate of all-cause death was 2.30%/y in the total population (Figure 1). The overall annualised rates were low for CV death (0.62%/y), major bleeding (1.04%/y), ICH (0.21%/y), and major GI bleeding (0.44%/y). Stroke/SEE occurred in 49 patients (1.13%/y), of whom 28 patients (0.65%/y) had ischaemic stroke and seven patients (0.16%/y) had haemorrhagic stroke. The incidences of TIA and MI were 0.35%/y and 0.25%/y, respectively, in the overall patient population.

Unadjusted annualised event rates, which should be interpreted with caution due to significant baseline differences between dose groups (Table 1), showed that during the 1-year follow-up period, the overall annualised rates of all-cause death in the 30 mg and 60 mg edoxaban remained constant over time (Figure 2A). Patients receiving 30 mg edoxaban had a higher risk of all-cause death as well as CV death when compared with patients receiving 60 mg edoxaban (all-cause death: 3.73%/y vs. 1.13%/y, log-rank P<0.0001; CV deaths: 0.92%/y vs. 0.38%/y, P = 0.024; Figures 2A,B). A larger proportion of patients in the 30 mg edoxaban group than 60 mg edoxaban group experienced major bleeding (1.59%/y vs. 0.59%/y, P = 0.001; Figure 2C). No significant differences in the annualised rates of stroke/SEE, or TIA between the dose groups were observed (Figure 2D; P > 0.05).

The factors associated with all-cause death, CV death, all stroke/SEE/TIA, and major bleeding were significant if P < 0.05 of univariable analysis (Supplementary Tables S2–S5). The selected factors were further adjusted by fixing edoxaban dosing (60 and 30 mg) in a stepwise multivariable Cox regression model (Figure 3). A CrCl <30 mL/min (HR: 5.79, 95% CI: 2.53–12.71; P<0.0001), underweight (BMI < 18.5; HR: 2.94, 95% CI: 1.52–5.69; P = 0.001), CHA₂DS₂-VASc scores (HR: 1.33, 95% CI: 1.11–1.58; P = 0.002), long-standing persistent AF (HR: 3.14, 95% CI: 1.64–5.99; P<0.0001) and permanent AF (HR: 3.30, 95% CI: 1.76–6.19; P<0.0001) were the main clinically significant factors related to all-cause death (Figure 3A). Male gender (HR: 3.09, 95% CI: 1.93–4.96; P<0.0001), perceived frailty (HR: 2.14, 95% CI: 1.20–3.83; P<0.0001), and economic regions of Northeast (HR: 3.10, 95% CI: 1.66–5.80; P<0.0001) were further identified as the predictors of all-cause death. Similarly, CrCl <30 mL/min (HR: 10.85, 95% CI: 2.46–47.90; P = 0.002) and perceived frailty (HR: 2.94, 95% CI: 1.16–7.46; P = 0.023) were also associated with CV death (Figure 3B). The persistent AF type (HR: 4.18, 95% CI: 1.70–10.30; P = 0.002) was a strong predictor of major bleeding events, in addition to permanent AF (HR: 3.61, 95% CI: 1.28–10.19; P = 0.015), BMI of <18.5 (HR: 3.75, 95% CI: 1.38–10.17; P = 0.009), HAS-BLED scores (HR: 1.60, 95% CI: 1.12–2.30; P = 0.011) and CHA₂DS₂-VASc scores (HR: 1.35, 95% CI: 1.05–1.73; P = 0.020) (Figure 3C). For stroke/SEE/TIA, HAS-BLED score (HR: 1.69, 95% CI: 1.31–2.18; P<0.0001) and HF (HR: 1.96, 95% CI: 1.12–3.44; P = 0.019) were relevant predictors (Figure 3D).

Upon performing the step-wise multivariable cox regression analysis of recommended/non-recommended doses, patients on non-recommended doses (both under- and over-dosing) often had worse outcomes, likely reflecting their higher baseline risk profile rather than a direct effect of the dosing itself (Supplementary Table S6).

After 1-year follow-up, patients who were prescribed the non-recommended 30 mg dose of edoxaban had higher risks of all-cause mortality (HR: 2.49, 95% CI: 1.27–4.91; P = 0.006), major bleeding (HR: 3.54, 95% CI: 1.37–9.17, P = 0.005), and major GI bleeding (HR: 5.22, 95% CI: 1.25–21.85, P = 0.01) compared with the recommended 60 mg dose (Supplementary Figure S4). Patients receiving the non-recommended 60 mg versus the recommended 30 mg had lower annualised event rates of all-cause mortality (HR: 0.40, 95% CI: 0.16–0.99; P = 0.04). The rates of stroke/SEE/TIA and ischaemic stroke were similar between non-recommended 30 mg versus recommended 60 mg or non-recommended 60 mg versus recommended 30-mg dose groups (all P > 0.05).

The strengths of the ETNA-AF-China study were the inclusion of 4,877 patients and the presentation of comprehensive 1-year follow-up data by interim cut-off, which provided significant insights into the demographic, clinical characteristics, safety, and effectiveness outcomes of patients with AF prescribed edoxaban at the beginning of the study. This study fully collected dose reduction criteria of edoxaban (including renal function), which can provide information for judgment of recommended and non-recommended dosing regimens of NOACs in clinical management of AF. The reflection of real-world conditions, including the extensive sample size, its prospective study design, and the autonomy of physicians in making therapeutic choices, enhances the reliability and applicability of the gathered data in the ETNA-AF-China study.

This study, however, has certain limitations. Owing to its observational and open-label design, detailed data on laboratory tests and other assessments were not collected. The single-arm study design only reported the effectiveness and safety of edoxaban, not a comparison or control against other anticoagulants such as NOAC or VKA. In addition, patients were assigned to the 30 mg or 60 mg edoxaban dose based on clinical characteristics rather than randomization, making direct comparisons between dose groups potentially confounded by indication and underlying health differences rather than the drug’s effect alone. Therefore, the reported P values should be interpreted in an exploratory or descriptive manner, since no adjustment for multiple testing was made. The subgroup analyses are further limited by their small sample size, multiple testing, and potential instability of the model due to small event numbers. Moreover, the annualised event rates were estimated by censoring those lost to follow-up or premature termination (11.2%), which may lead to underreported events or co-medications. As this is a real-world, observational study, there was no mandatory requirement for participants or investigators to complete follow-up. This inherent lack of compulsion may have led to incomplete data capture. Furthermore, the current data of ETNA-AF China are not mature enough to be pooled with the global ETNA-AF database, as analyses are still undergoing alignment to standardize results across ETNA-AF studies.