Epidemiology, comorbidity profiles, and oral anticoagulation patterns among patients with atrial fibrillation in Turkey: a meta-analysis of national registry-based and multicenter observational studies

Abstract

Background

Atrial fibrillation (AF) is a major public health burden globally, associated with increased risk of stroke and mortality. While real-world data on AF management have expanded globally, comprehensive syntheses from developing countries such as Turkey remain limited. The objective of this study was to systematically evaluates the epidemiological characteristics, comorbidity profiles, and patterns of oral anticoagulant (OAC) use among patients with AF in Turkey based on national registry-based and multicenter observational studies.

Methods

A systematic search of PubMed, Scopus, Web of Science, and Embase was conducted on 15 November 2025, and subsequently updated on 8 July 2026 to identify studies of Turkish AF patients. Included registries were conducted between 2013 and 2024. Data were extracted on demographics, clinical risk scores (CHA₂DS₂-VASc, HAS-BLED), comorbidities, AF type, and OAC patterns. Pooled estimates were calculated using random-effects meta-analyses with restricted maximum likelihood (REML) estimation and Knapp–Hartung adjustments.

Results

Six registries comprising 18,219 Turkish AF patients were included. The pooled mean age was 69 years (95% CI: 66.9–71.2), with 58% (95% CI: 0.56–0.60) female representation. Hypertension (0.76, 95% CI: 0.71–0.80) and diabetes mellitus (0.25, 95% CI: 0.21–0.28) were the most common comorbidities. The mean CHA₂DS₂-VASc and HAS-BLED scores were 3.67 (95% CI: 2.99–4.35) and 1.72 (95% CI: 1.36–2.08), respectively, reflecting moderate-to-high thromboembolic risk and moderate bleeding risk. Persistent/permanent AF accounted for 0.73 (95% CI: 0.57–0.89) of cases, although the very high heterogeneity warrants caution. Because three registries enrolled patients on the basis of NOAC therapy by design, treatment-pattern estimates were stratified by registry type. Among all-comer registries, pooled NOAC use was 0.60 (95% CI: 0.20–0.90) and overall OAC use was 0.88 (95% CI: 0.49–0.98), whereas pooling all registries produced inflated estimates (NOAC 0.97, OAC 0.99); VKA use was correspondingly higher in all-comer registries (0.24, 95% CI: 0.10–0.47). Procedural interventions such as ablation (3%) and cardioversion (9%) remained infrequent.

Conclusions

This meta-analysis of registry-based and multicenter observational studies suggests progress in pharmacological stroke prevention among AF patients in Turkey, with greater NOAC use in later and all-comer registries. However, the low use of rhythm-control strategies and the substantial inter-registry heterogeneity indicate ongoing gaps in AF management that warrant cautious interpretation and system-level improvements.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12872-026-06342-2.

Introduction

Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia worldwide, with its global prevalence rising from approximately 19 million in 1990 to 52.6 million in 2021; this number is projected to continue increasing through 2046 [1, 2]. AF has been described as a growing epidemic, as its incidence and prevalence continue to increase in tandem with aging populations [3]. This arrhythmia contributes significantly to cardiovascular morbidity and mortality, being a leading cause of stroke, heart failure, and other adverse outcomes [1, 3]. Notably, AF confers approximately a five-fold higher risk of ischemic stroke, and strokes related to AF tend to be more severe or fatal than other ischemic strokes [4–6].

Stroke prevention with appropriate oral anticoagulation (OAC) is the cornerstone of AF management and has markedly reduced stroke incidence in AF patients [3]. Current clinical guidelines recommend chronic OAC for most AF patients with elevated stroke risk (e.g., those above specific CHA₂DS₂-VASc score thresholds) to prevent thromboembolic events [7, 8]. In the past decade, non-vitamin K antagonist oral anticoagulants (NOACs) have largely supplanted warfarin as first-line therapy for stroke prevention in non-valvular AF, owing to their non-inferior efficacy and improved safety and convenience profiles [9]. AF is also an increasingly prevalent and consequential condition in Turkey, mirroring global trends [10]. Patients in Turkish AF cohorts tend to be elderly (mean age ~ 70 years) with a high burden of comorbidities; for instance, hypertension is present in about 70–80% of AF patients in several registries [11, 12]. Such demographics emphasize a predominantly high-risk AF population in Turkey, similar to those observed in other aging societies.

The management of AF in Turkey has evolved markedly over the last decade, particularly regarding the use of OAC therapy [13]. More recent data indicate substantial improvement: the 2022 AFTER-2 study demonstrated significantly higher OAC usage compared to prior Turkish registries, and it highlighted that persistently low time-in-therapeutic-range with warfarin remains a problem, reinforcing the national shift toward NOAC therapy [13]. Nevertheless, to date, there has been no comprehensive synthesis of these disparate Turkish studies. In light of this gap and the limited epidemiological information previously available from developing countries, a meta-analysis of registry-based and multicenter observational studies was undertaken to better characterize the epidemiology of AF and patterns of oral anticoagulation use in Turkey.

Methods

This systematic review and meta-analysis was designed and conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2020) guidelines [14]. This review was not prospectively registered in a public database (e.g., PROSPERO). Although a protocol specifying the objectives, eligibility criteria, and analysis plan was defined a priori before data extraction and analysis, it was not published on a public registration platform. Our primary objective was to evaluate the epidemiological characteristics and real-world oral anticoagulant use among patients with AF in Turkey, based on registry-and multicenter observational studies. The PRISMA checklist is also presented in Supplementary Table S1.

The search strategy targeted multiple electronic databases, including PubMed/MEDLINE, Scopus, Embase (Ovid), and Web of Science Core Collection, to ensure comprehensive coverage of relevant literature. Eligible studies were observational, registry-based, or cross-sectional investigations reporting on the prevalence, treatment patterns, or anticoagulation practices in adult patients diagnosed with atrial fibrillation in Turkey. No language restrictions were applied.

Literature search and data sources

A comprehensive literature search was conducted to identify all relevant studies reporting on the epidemiology and oral anticoagulant treatment patterns among patients with atrial fibrillation in Turkey. The search strategy was developed a priori in accordance with PRISMA 2020 recommendations, with the aim of maximizing sensitivity and minimizing selection bias. The search combined Medical Subject Headings (MeSH) and free-text terms related to atrial fibrillation, oral anticoagulant therapy, and the Turkish population. Boolean operators (“AND,” “OR”) were used to combine terms within and between concepts. The detailed search strategies for each database are provided in Supplementary Table S2.

To ensure comprehensive coverage, the reference lists of all included studies and relevant review articles were also manually screened to identify additional eligible studies not captured by the electronic searches. No language or publication status restrictions were applied. Duplicate records were removed using Mendeley reference management software. The deduplicated records were then screened in two stages. First, two reviewers independently screened all titles and abstracts against the eligibility criteria. Second, the full texts of potentially eligible records were retrieved and independently assessed by the same two reviewers. At both stages, disagreements were resolved by discussion until consensus was reached; when consensus could not be reached, a third senior reviewer made the final decision.

Study selection of interest and eligibility criteria

Studies were considered eligible if they met predefined inclusion criteria established a priori. Eligible studies included observational, registry-based, prospective or cross-sectional investigations that reported on the epidemiology, treatment patterns, or oral anticoagulant use among adult patients (≥ 18 years) with a diagnosis of atrial fibrillation in Turkey. We included studies published between January 1, 2010 and November 15, 2025, to capture the evolution of oral anticoagulant prescribing patterns over time and subsequently updated on 8 July 2026. The included registries were published between 2013 and 2024.

To be eligible, a study was required to report at least one of the following outcomes: (i) epidemiological characteristics (prevalence, demographic profile, or comorbidities); (ii) oral anticoagulant treatment patterns (NOAC or VKA use, dosing, or adherence); (iii) risk-stratification scores (CHA₂DS₂-VASc or HAS-BLED); or (iv) clinical outcomes (bleeding events, thromboembolic events, or treatment persistence), where reported. These outcome categories correspond directly to those specified in the PICOS framework below. We excluded: (i) case reports, reviews, editorials, and conference abstracts without primary data; (ii) randomized controlled trials; (iii) studies of paediatric populations; (iv) studies conducted outside Turkey; and (v) articles without sufficient extractable data relevant to the meta-analysis objectives. When duplicate reports from the same registry or cohort were identified, the most comprehensive and up-to-date version of the dataset was retained to avoid double-counting. All inclusion and exclusion decisions were made independently by two investigators, with disagreements resolved by discussion and consensus and, where necessary, adjudication by a third senior reviewer.

Data extraction and quality assessment

Data extraction was performed independently by two investigators using a standardized, pre-piloted data extraction form. From each eligible study, we extracted the following information: (i) study identifiers (first author, publication year, journal); (ii) study design and setting (number of centers, recruitment method, population characteristics); (iii) sample size and demographic data (age, sex distribution); (iv) clinical characteristics (comorbidities, risk scores such as CHA₂DS₂-VASc and HAS-BLED); (v) type and pattern of oral anticoagulant use (NOAC or VKA, dosing, adherence, or treatment persistence); and (vi) outcome data where available (bleeding events, thromboembolic complications, or discontinuation rates). All data were cross-checked for accuracy and completeness. When information was incomplete or unclear, attempts were made to retrieve additional data from the original publication or supplementary materials. If multiple publications reported data from the same registry or cohort, the most recent and comprehensive dataset was included to avoid duplication.

Because this study synthesised prevalence and proportion data from cross-sectional and prospective registries, methodological quality was assessed using the Joanna Briggs Institute (JBI) Critical Appraisal Checklist for Studies Reporting Prevalence Data [15], a tool specifically designed for this study type and applicable across the included designs. The checklist comprises nine items addressing the appropriateness of the sampling frame and sampling method, sample size, description of subjects and setting, coverage of the analysed sample, the validity and reliability of atrial fibrillation ascertainment, the appropriateness of the statistical analysis, and the adequacy of the response rate. Each item was rated as ‘yes,’ ‘no,’ or ‘unclear,’ and a summary score (0–9) was derived, with higher scores indicating lower risk of bias. Each study was independently appraised by two reviewers, and disagreements were resolved by discussion and, where necessary, adjudication by a third senior reviewer. The detailed appraisal is presented in Table 1.

Table 1.

Methodological quality of the included studies assessed with the Joanna Briggs Institute (JBI) critical appraisal checklist for studies reporting prevalence data

Study	First author, year	1	2	3	4	5	6	7	8	9	Total /9
AFTER	Ertaş et al., 2013 [11]	Y	Y	Y	Y	Y	Y	Y	Y	N	8
RAMSES	Başaran et al., 2016 [16]	Y	Y	Y	Y	Y	Y	Y	Y	Y	9
NOAC-TURK	Altay et al., 2017 [12]	N	Y	Y	Y	Y	Y	Y	Y	N	7
NOAC-TR	Emren et al., 2018 [17]	N	Y	Y	Y	Y	Y	Y	Y	N	7
ROTA	Kocabaş et al., 2023 [18]	Y	Y	Y	Y	Y	Y	Y	Y	N	8
ANATOLIA-AF	Kocabaş et al., 2024 [19]	N	Y	Y	Y	Y	Y	Y	Y	N	7

Each item scored 1 (yes) or 0 (no/unclear) or Not applicable; total score range 0–9. Higher scores indicate lower risk of bias. All included studies scored ≥ 7, corresponding to low risk of bias. The three NOAC-only registries (NOAC-TURK, NOAC-TR, ANATOLIA-AF) scored ‘No’ on item 1 because their sampling frame was restricted by design to patients receiving NOAC therapy and therefore did not represent the general atrial fibrillation population. Item 9 was scored ‘Yes’ only for RAMSES, which reported a predefined sample-size calculation and enrolment/response handling

Y yes, N no

Endpoints of interest for meta-analysis and PICOS framework

The primary aim of this meta-analysis was to synthesize epidemiological data and treatment patterns regarding oral anticoagulant use among patients with atrial fibrillation in Turkey. Endpoints were pre-specified to ensure consistency in data collection and analysis. The primary endpoints were: Prevalence and demographic characteristics of patients with atrial fibrillation; Oral anticoagulant treatment patterns, including NOAC and VKA use, dosing strategies, and treatment selection over time. The secondary endpoints included: Clinical risk stratification using CHA₂DS₂-VASc and HAS-BLED scores; Adherence and persistence rates with oral anticoagulation; Clinical outcomes, including ischemic stroke, systemic embolism, and major bleeding events when reported; Temporal trends in anticoagulant prescribing practices.

To ensure methodological clarity, the PICOS (Population, Intervention, Comparator, Outcomes, Study design) framework was applied as follows:

Population

Adults (≥ 18 years) with a diagnosis of atrial fibrillation in Turkey.

Exposure/intervention

Oral anticoagulant therapy (NOACs or VKAs), where reported; studies reporting epidemiological or risk-stratification data without anticoagulant information were also eligible, consistent with the descriptive aim of this synthesis.

Comparator

Not applicable. This was a single-group prevalence/proportion synthesis; no comparator group was required.

Outcomes

At least one of the following—(i) epidemiological characteristics (prevalence, demographic profile, comorbidities); (ii) oral anticoagulant treatment patterns (NOAC or VKA use, dosing, adherence); (iii) risk-stratification scores (CHA₂DS₂-VASc, HAS-BLED); or (iv) clinical outcomes (bleeding, thromboembolic events, treatment persistence), where reported.

Study design

Multicenter observational, registry-based, prospective, or cross-sectional studies.

Statistical analysis

All quantitative syntheses were conducted in Stata/MP 18.0 (StataCorp, College Station, TX). Proportion and mean outcomes were synthesised using the Stata meta suite (meta set, meta summarize, and meta forestplot). Analyses followed a pre-specified plan and were executed with random-effects models throughout to account for expected between-study variability across registries, settings, and calendar periods. For prevalence and treatment-pattern outcomes, proportions were pooled on the logit scale (with a 0.5 continuity correction) and back-transformed to the proportion scale for presentation. Accordingly, the adjusted proportion was calculated as (x + 0.5)/(n + 1), and the transformed effect size was defined as logit[(x + 0.5)/(n + 1)], with a standard error of √[1/(x + 0.5) + 1/(n − x + 0.5)], where x denotes the number of events and n denotes the total sample size. For continuous variables (e.g., age, CHA₂DS₂-VASc, HAS-BLED), the pooled mean with its 95% confidence intervals (CIs) was estimated. When studies reported median (IQR) or median (range) instead of mean (SD), we estimated the mean and SD from summary quantiles using established methods [20, 21]. Standard errors (SEs) were then derived as SE = SD / √n. All conversions were performed prior to meta-analysis. The primary model was a random-effects meta-analysis using a restricted maximum likelihood (REML) estimator for the between-study variance τ². CIs for pooled estimates were derived using the Knapp–Hartung small-sample adjustment for outcomes with a sufficient number of studies. For subgroup analyses containing fewer than three studies per stratum, the Knapp–Hartung adjustment was not estimable, and standard random-effects (REML-based) CIs were used instead. Between-study heterogeneity was quantified with I² (with thresholds of ~ 25%, 50%, and 75% interpreted as low, moderate, and high heterogeneity, respectively). Assessment of publication bias (small-study effects) using funnel plot asymmetry and Egger’s regression test was pre-specified but not performed, because fewer than ten studies contributed to each outcome [22, 23]. Two-sided p < 0.05 defined statistical significance without multiplicity adjustment for secondary/exploratory analyses.

Results

A total of 215 records were identified through database searching (PubMed/MEDLINE, n = 47; Scopus, n = 63; Web of Science, n = 71; Embase, n = 34). After removal of 116 duplicate records, 99 unique records underwent title and abstract screening, of which 94 were excluded based on the predefined eligibility criteria. The remaining five full-text reports were retrieved and assessed for eligibility, and all five met the inclusion criteria. In addition, four reports were identified through citation searching of reference lists; of these, three were excluded because they involved an ineligible study population, and one met the inclusion criteria. In total, six studies [11, 12, 16–19] were included in the final qualitative synthesis and meta-analysis. The study selection process is summarized in the PRISMA flow diagram (Fig. 1).

Fig. 1.

PRISMA 2020 flow diagram for study selection

Baseline characteristics of studies included in the systematic review

A total of six registry-based and multicenter observational studies were included in this meta-analysis, encompassing a combined population of 18,219 patients with atrial fibrillation across multiple centers in Turkey. The sample sizes of individual studies ranged from 1,745 (AFTER registry) to 6,273 (RAMSES registry). The mean or median age across studies was consistently high, ranging from 69.2 to 72 years, with a predominance of female patients in all cohorts (range: 984–3,504 individuals per study). Hypertension (ranging from 1,274 to 4,305 patients) and diabetes (418 to 1,389 patients) were the most frequent comorbidities. Notably, heart failure and prior cerebrovascular events were also common, reflecting a high-risk population (Table 2).

Table 2.

Baseline characteristics of patients with atrial fibrillation in included studies

Characteristics	First author/year
	Ertaş et al. 2013 [11] (AFTER study)	Başaran et al. 2016 [16] (RAMSES study)	Emren et al. 2018 [17] (NOAC study)	Kocabaş et al. 2023 [18] (ROTA study)	Kocabaş et al. 2024 [19] (ANATOLIA‑AF study)	Altay et al. 2017 [12] (NOAC-TURK study)
Study design	Prospective, multicenter	Cross-sectional, multicenter	Cross-sectional, multicenter	Prospective, multicenter	Cross-sectional, multicenter	Cross-sectional, multicenter
Sample Size (n)	1745	6273	2738	2597	2004	2862
Age (Mean/Median)	69.2	69.6	70	72 (median)	72	70.3
Sex, female (n)	984	3504	1623	1491	1163	1731
Most common DOAC used	N/A	Rivaroxaban	N/A	Rivaroxaban	Rivaroxaban	Rivaroxaban & Dabigatran
CHA2DS2-VASc Mean	N/A	3.3	N/A	4.0	3.72	3.4
HAS-BLED Mean	N/A	1.6	N/A	1.0	1.49	1.8
Smoking (n)	227	1023	440	433	320	534
AF Persistent/Permanent (n)	1365	5066	N/A	1815	1025	2385
AF Paroxysmal (n)	289	859	N/A	782	N/A	325
AF First attack (n)	91	290	N/A	N/A	N/A	N/A
Hypertension (n)	1274	4305	2141	1994	1552	2320
Diabetes (n)	418	1389	674	743	572	568
Dyslipidemia (n)	N/A	N/A	787	710	568	1070
Heart Failure (n)	537	1386	689	984	768	765
Previous Stroke/TIA (n)	266	845	332	382	296	326
Vascular disease/MI (n)	522	1828	745	842	683	764
Anemia (n)	N/A	N/A	N/A	513	399	N/A
Cardioversion (n)	N/A	N/A	N/A	221	174	N/A
AF Ablation (n)	N/A	N/A	N/A	80	69	N/A
Major bleeding (n)	N/A	N/A	135	88	64	315
OAC use (n)	N/A	4513	N/A	2474	2004	2862
NOAC (n)	N/A	2340	N/A	2062	100	2847
VKA (n)	N/A	2173	N/A	412	0	13
No anticoagulant (n)	N/A	1760	N/A	113	0	2
Beta blocker (n)	N/A	3889	N/A	1936	1321	1302
Non-DHP CCB (n)	N/A	1451	N/A	320	207	369
Digoxin (n)	N/A	1259	N/A	383	248	323
Amiodarone (n)	N/A	295	N/A	123	62	N/A
Propafenone (n)	N/A	174	N/A	106	85	N/A
Sotalol (n)	N/A	56	N/A	14	N/A	N/A

AF Atrial Fibrillation, OAC Oral Anticoagulant, NOAC / DOAC Non–Vitamin K Antagonist Oral Anticoagulant / Direct Oral Anticoagulant, VKA Vitamin K Antagonist, LMWH Low Molecular Weight Heparin, CHA₂DS₂-VASc Congestive heart failure, Hypertension, Age ≥ 75 (doubled), Diabetes, Stroke (doubled), Vascular disease, Age 65–74, Sex category (female), HAS-BLED Hypertension, Abnormal renal/liver function, Stroke, Bleeding, Labile INR, Elderly (> 65), Drugs/alcohol, COPD Chronic Obstructive Pulmonary Disease, MI Myocardial Infarction, TIA Transient Ischemic Attack, CCB Calcium Channel Blocker, Non-DHP CCB Non-Dihydropyridine Calcium Channel Blocker, SD Standard Deviation, IQR Interquartile Range

Two studies (AFTER and ROTA) employed prospective, multicenter registry designs, whereas the remaining four were cross-sectional, nationwide studies. CHA₂DS₂-VASc scores were reported in four studies, with mean values ranging between 3.3 and 4.0, indicating moderate to high thromboembolic risk across cohorts. Similarly, HAS-BLED scores ranged from 1.0 to 1.8, suggesting a relatively low to moderate bleeding risk profile. Rivaroxaban emerged as the most commonly prescribed NOAC across studies, followed by dabigatran in some cohorts. In total, NOACs were preferred over VKAs, with 2,847 patients receiving NOACs in the NOAC-TURK registry compared to minimal VKA use in more recent studies (e.g., 0 patients in ANATOLIA-AF). Earlier registries, including RAMSES, reported balanced use of NOACs (n = 2,340) and VKAs (n = 2,173), whereas later registries reported proportionally greater NOAC use.

Quality assessment of included studies

The methodological quality of the six included studies was assessed using the JBI Critical Appraisal Checklist for prevalence studies (Table 1). Overall, the studies were of high methodological quality, with total scores ranging from 7 to 9 out of 9, corresponding to a low risk of bias across all included registries. RAMSES achieved the highest score (9/9), as it was the only registry to report a predefined sample-size calculation together with explicit handling of enrolment and response. AFTER and ROTA each scored 8/9, and the three NOAC-selected registries (NOAC-TURK, NOAC-TR, and ANATOLIA-AF) each scored 7/9. Across all studies, atrial fibrillation was ascertained by electrocardiographic documentation, sample sizes were large, and study populations and settings were described in detail (items 3, 4, 6, and 7 satisfied by all studies). Two sources of downgrading were identified: the three NOAC-only registries did not satisfy item 1, because their sampling frame was restricted by design to patients receiving NOAC therapy and therefore did not represent the general AF population; and the response rate (item 9) was not reported in five of the six registries. These findings indicate that, although the included registries were methodologically robust, the NOAC-only designs constrained the representativeness of their sampling frames—consistent with the stratified treatment-pattern analyses reported above.

Meta-analysis outcomes

Across the five eligible included studies, the pooled mean age of patients with atrial fibrillation was 69.0 years (95% CI: 66.9–71.2) (Fig. 2A). The proportion of female patients was 0.58 (95% CI: 0.56–0.60) (Fig. 2B). The prevalence of current smoking was estimated at 0.16 (95% CI: 0.14–0.18) across studies (Fig. 2C). In addition, the proportion of patients presenting with AF as a first attack was 0.05 (95% CI: 0.04–0.05), based on data from two national registries (AFTER and RAMSES) (Fig. 2D). Heterogeneity ranged from low to very high across different endpoints (I² = 0.04–98.8%).

Fig. 2.

Pooled baseline demographic and clinical characteristics of patients with atrial fibrillation in Turkey. A Pooled mean age (years), B proportion of female patients, C smoking prevalence, and D proportion of patients presenting with AF as a first attack. Random-effects meta-analysis using REML estimation and Knapp–Hartung standard errors

Across the included studies, the pooled mean CHA₂DS₂-VASc score was 3.67 (95% CI: 2.99–4.35) (Fig. 3A). Similarly, the mean HAS-BLED score was 1.72 (95% CI: 1.36–2.08) (Fig. 3B). Regarding AF type, the prevalence of persistent/permanent atrial fibrillation was estimated at 0.73 (95% CI: 0.57–0.89) across five national registries; however, given the very high heterogeneity (I² ≈ 99%), this pooled estimate should be interpreted with caution (Fig. 3C). In contrast, the proportion of paroxysmal AF was 0.18 (95% CI: 0.05–0.31) (Fig. 3D). All analyses were performed using random-effects models with REML estimation and Knapp–Hartung standard errors. Substantial heterogeneity was observed for all endpoints (I² ranging from 99.3% to 99.6%), reflecting variations in study periods, registry designs, and patient risk profiles.

Fig. 3.

Pooled estimates of risk stratification scores and AF type among patients with atrial fibrillation in Turkey. A pooled mean CHA₂DS₂-VASc score, B pooled mean HAS-BLED score, C proportion of persistent/permanent AF, and (D) proportion of paroxysmal AF. Random-effects meta-analysis using REML estimation and Knapp–Hartung standard errors

The pooled prevalence of hypertension among patients with atrial fibrillation in Turkey was 0.76 (95% CI: 0.71–0.80) (Fig. 4A). In contrast, the proportion of patients who had undergone AF ablation was markedly low at 0.03 (95% CI: 0.03–0.04) (Fig. 4B). Similarly, the prevalence of anemia was 0.20 (95% CI: 0.19–0.21) (Fig. 4C), while cardioversion was reported in only 0.09 (95% CI: 0.08–0.09) of patients (Fig. 4D). Heterogeneity was negligible in the ablation, anemia, and cardioversion estimates, but high for hypertension. The pooled prevalence of diabetes mellitus among patients with atrial fibrillation was 0.25 (95% CI: 0.21–0.28) (Fig. 5A). Similarly, dyslipidemia was reported in 0.30 (95% CI: 0.23–0.38) of patients (Fig. 5B). The prevalence of heart failure was 0.30 (95% CI: 0.23–0.37) (Fig. 5C). Additionally, previous stroke or transient ischemic attack was observed in 0.14 (95% CI: 0.12–0.15) of patients across the included registries (Fig. 5D). All estimates were derived using random-effects meta-analysis with REML estimation and Knapp–Hartung adjustments. Heterogeneity was high for most comorbidities (I²: 83–98%).

Fig. 4.

Pooled estimates of key comorbidities and interventional procedures among patients with atrial fibrillation in Turkey. A Prevalence of hypertension, B prevalence of AF ablation, C prevalence of anemia, and (D) prevalence of cardioversion. Random-effects meta-analysis using REML estimation and Knapp–Hartung standard errors

Fig. 5.

Pooled prevalence of major cardiometabolic comorbidities among patients with atrial fibrillation in Turkey. A Diabetes mellitus, B dyslipidemia, C heart failure, and (D) previous stroke or transient ischemic attack. Random-effects meta-analysis was performed using REML estimation with Knapp–Hartung standard errors

The pooled prevalence of vascular disease among patients with atrial fibrillation was 0.30 (95% CI: 0.27–0.33) (Fig. 6A). In contrast, the use of amiodarone was low, with a pooled prevalence of 0.04 (95% CI: 0.02–0.07) (Fig. 6B). The use of beta-blockers was high, with a pooled proportion of 0.62 (95% CI: 0.43–0.81) (Fig. 6C). Conversely, digoxin was prescribed less frequently, with a pooled prevalence of 0.15 (95% CI: 0.08–0.21) (Fig. 6D). Heterogeneity was high for all comparisons. Heterogeneity was very high, reflecting variations in registry periods and patient risk profiles. The pooled prevalence of propafenone use among patients with atrial fibrillation was 0.04 (95% CI: 0.02–0.06), based on data from three registries (Fig. 7A). The prevalence of sotalol use was even lower, with a pooled estimate of 0.01 (95% CI: 0.00–0.01) across two studies (Fig. 7B). Heterogeneity was high for propafenone use (I² = 83.8%) and high for sotalol (I² = 72.2%). Because three registries (NOAC-TURK, NOAC-TR, ANATOLIA-AF) enrolled patients on the basis of NOAC therapy by design, all treatment-pattern outcomes were stratified by registry type (Fig. 8). Among all-comer registries (RAMSES, ROTA), the pooled proportion of NOAC use was 0.60 (95% CI: 0.20–0.90; panel A) and overall OAC use was 0.88 (95% CI: 0.49–0.98; panel C), compared with inflated full-dataset estimates of 0.97 (0.42–1.00) and 0.99 (0.77–1.00), respectively. Conversely, VKA use among all-comer registries was 0.24 (95% CI: 0.10–0.47; panel D) versus a deflated overall estimate of 0.02 (0.00–0.38). The between-group differences were significant for every outcome (Qb = 11.5–20.6, all p < 0.001). A design-robust metric, NOAC share among anticoagulated patients, was 0.70 (95% CI: 0.34–0.91) in all-comer registries (panel B). The pooled prevalence of major bleeding events among patients receiving oral anticoagulation was 0.06 (95% CI: 0.00–0.11) (Fig. 9A). The use of non-dihydropyridine calcium channel blockers was less frequent, with a pooled prevalence of 0.15 (95% CI: 0.06–0.24) (Fig. 9B).

Fig. 6.

Pooled prevalence of vascular disease and commonly used medications among patients with atrial fibrillation in Turkey. A Vascular disease, B amiodarone use, C beta-blocker use, and (D) digoxin use. Random-effects meta-analysis was performed using REML estimation with Knapp–Hartung standard errors

Fig. 7.

Pooled prevalence of propafenone and sotalol use among patients with atrial fibrillation in Turkey. A Propafenone use and (B) sotalol use. Random-effects meta-analysis was performed using REML estimation with Knapp–Hartung standard errors

Fig. 8.

Treatment-pattern estimates among patients with atrial fibrillation in Turkey, stratified by registry type (all-comer vs. NOAC-only). (A) NOAC use as a proportion of all AF patients; (B) NOAC share among anticoagulated patients; (C) overall oral anticoagulant (OAC) use; (D) vitamin K antagonist (VKA) use. Random-effects meta-analysis was performed using REML estimation

Fig. 9.

Pooled prevalence of major bleeding events and non-dihydropyridine calcium channel blocker use among patients with atrial fibrillation in Turkey. (A) Major bleeding; (B) Non-dihydropyridine calcium channel blocker use. Random-effects meta-analysis using REML estimation with Knapp–Hartung standard errors and 95% prediction intervals

Discussion

This meta-analysis provides a comprehensive synthesis of epidemiological characteristics and anticoagulation patterns among Turkish patients with AF. In summary, Turkish AF patients were predominantly elderly with a high comorbidity burden and a moderate-to-high thromboembolic risk profile, and later all-comer registries reported greater NOAC use, while rhythm-control interventions remained infrequent. The data are consistent with a greater NOAC use in the more recent all-comer registry (ROTA, 2021) than in the earlier one (RAMSES, 2015). However, this inference rests on a small number of registries spanning different years, and pooled NOAC estimates were strongly influenced by registry design; among all-comer registries, NOAC use was 0.60. This trend should therefore be regarded as indicative rather than definitive. Because registry year is strongly confounded by registry design, the NOAC-only registries were conducted contemporaneously with the earlier all-comer registry (2015–2016) yet enrolled only NOAC-treated patients, we did not perform a formal temporal (meta-regression) analysis, which would be underpowered and confounded with only two all-comer registries available. The data therefore indicate that later and all-comer registries reported greater NOAC use, but they do not establish a temporal trend. Notably, after excluding the NOAC-only registries, these primary all-comer estimates were derived from only two registries (RAMSES and ROTA) and were correspondingly imprecise, with wide confidence intervals (NOAC use 0.60, 95% CI: 0.20–0.90). Moreover, the pooled estimates that combined NOAC-only and all-comer registries were upwardly biased by design and must not be interpreted as representative population-level estimates of anticoagulant use in Turkish AF patients. Rhythm-control procedures were infrequent (ablation 3%, cardioversion 9%). Whether this represents underuse cannot be established without comparator data on the proportion of guideline-indicated candidates, and these rates may also reflect differences in patient selection, access, or clinical practice. Overall, these findings are consistent with improvements in pharmacologic stroke prevention alongside persistent deficiencies in comprehensive AF care in Turkey, indicating a need to better characterise the use of, and access to, rhythm-control therapies in Turkey.

Previously published studies, including those from Western populations, have reported that the mean age of patients with AF exceeds 70 years [24, 25]. In our meta-analysis based on Turkish data, we found that the mean age among patients with AF was 69.0 years (95% CI: 66.9–71.2). In recent years, studies involving AF populations have shown higher rates of comorbid conditions such as acute coronary syndrome, cerebrovascular events, heart failure, and coronary artery disease compared with earlier reports. The variations in demographic characteristics among AF patients included in our meta-analysis may be attributed to several factors, including the recent increase in the elderly population in Turkey, differences in study design and patient recruitment, and the selection of participating centers.

Although more than 55% of participants were male in phase III randomized controlled trials of NOACs and in previous reports from Western healthcare systems [7, 26], 58% of the study population in this meta-analysis were female. The reasons for this female predominance are uncertain. It may partly reflect the older age structure of women in Turkish AF cohorts, as AF prevalence rises steeply with age and women in these registries were on average older. A higher burden of metabolic risk factors such as obesity and metabolic syndrome among Turkish women has been reported and could contribute, although our data do not permit direct assessment of these factors, and this explanation remains speculative [7, 10]. The role of female sex in AF has been actively reconsidered in recent years. Although female sex was historically associated with increased thromboembolic risk and is assigned one point in the CHA₂DS₂-VASc score, contemporary data indicate that this incremental risk has attenuated over time and appears to act as a risk modifier rather than an independent risk factor [27]. On this basis, the 2024 ESC guidelines introduced the CHA₂DS₂-VA score, which omits the sex category, to simplify stroke-risk stratification and standardise anticoagulation decisions across sexes [7]. At the same time, real-world European data show that women with AF may still experience differences in oral anticoagulant prescription and a residual risk of thromboembolism and stroke [28]. Given that women comprised 58% of our pooled Turkish AF population, these developments are directly relevant: our finding of a marked female predominance underscores the importance of applying contemporary, sex-neutral risk-stratification (CHA₂DS₂-VA) while remaining attentive to potential sex-related disparities in anticoagulant use and outcomes. Future Turkish registries should report sex-stratified anticoagulation and outcome data to clarify whether such disparities exist in this population.

In real-world studies, the mean CHA₂DS₂-VASc score has generally been reported to range between 3.2 and 4, while the mean HAS-BLED score varies between 1.0 and 1.6 [7, 24]. In the Turkish AF population analyzed in our meta-analysis, the mean CHA₂DS₂-VASc and HAS-BLED scores were 3.67 (95% CI: 2.99–4.35) and 1.72 (95% CI: 1.36–2.08), respectively, reflecting a moderate-to-high thromboembolic risk and a moderate bleeding risk profile. These findings suggest that the Turkish AF population tends to present with a relatively higher burden of both thromboembolic and hemorrhagic risk factors compared to some Western cohorts. This may be explained by the higher prevalence of hypertension, diabetes mellitus, and heart failure in Turkey, as well as suboptimal management of modifiable risk factors such as obesity and physical inactivity. Additionally, differences in patient selection, healthcare access, and clinical practice patterns may contribute to these elevated risk scores. Overall, these results underscore the importance of individualized anticoagulation strategies and comprehensive risk assessment tailored to the characteristics of the Turkish AF population.

Unlike randomized controlled trials (RCTs), observational studies aim to capture a broader spectrum of disease characteristics and real-world patient profiles. The findings of this meta-analysis, which included observational data, indicate that patients with AF in Turkey are generally at a moderate-to-high risk of thromboembolic events while exhibiting a low-to-moderate risk of bleeding. In large international RCTs, the incidence of major bleeding among patients receiving anticoagulant therapy for AF has been reported to range between 3.3% and 6.04% [7]. In contrast, our pooled analysis demonstrated that the overall prevalence of major bleeding events among Turkish patients receiving oral anticoagulants for AF was 6% (95% CI: 0.00–0.11), suggesting comparable or slightly higher rates than those observed in controlled settings. Moreover, the use of antiarrhythmic agents such as amiodarone and rate-control drugs such as digoxin appears to be lower in Turkey compared with RCT populations. This discrepancy may be attributed to differences in clinical practice patterns, prescribing behaviors, and patient characteristics across healthcare settings, emphasizing the need for localized management strategies that reflect real-world clinical conditions.

Rhythm control (including antiarrhythmic drugs and catheter ablation procedures) and rate control strategies constitute the cornerstone of atrial fibrillation (AF) management. In a recent meta-analysis evaluating which strategy provides superior clinical outcomes, data from 18 studies involving a total of 17,536 patients (mean age: 68.6 ± 9.7 years; 37.9% female) were analyzed. Among these patients, 31.9% had paroxysmal AF. The rhythm control strategy was found to significantly reduce cardiovascular mortality (HR: 0.78; 95% CI: 0.62–0.96), stroke (HR: 0.80; 95% CI: 0.64–0.99), and hospitalizations due to heart failure (HR: 0.80; 95% CI: 0.69–0.94), compared with rate control [29]. In contrast, our meta-analysis demonstrated that in the Turkish AF population, the proportion of patients undergoing catheter ablation for rhythm control was markedly lower—only 3% (95% CI: 0.03–0.04). This low rate may reflect limited access to advanced electrophysiological interventions, resource constraints, or variability in clinical decision-making processes across centers. Additionally, it may indicate a preference for pharmacological management over interventional approaches in real-world practice. These low observed rates warrant further evaluation, including comparison with the proportion of guideline-indicated candidates, to determine whether they reflect genuine underuse, differences in patient selection, or limited access to advanced rhythm-control services. Beyond procedural rhythm control, the pharmacological antiarrhythmic profile of the Turkish AF population also warrants comment, particularly given its advanced mean age. In our pooled analysis, antiarrhythmic drug (AAD) use was generally low: amiodarone was used in only 4% of patients, propafenone in 4%, and sotalol in 1%, while rate-controlling agents such as beta-blockers were used in 62% and digoxin in 15%. This pattern suggests a predominant reliance on rate control and a limited role for membrane-active AADs in routine Turkish practice. Such restraint may be clinically appropriate in an elderly population, in which the safety and tolerability of AADs become increasingly important. Ageing alters the pharmacokinetics and pharmacodynamics of antiarrhythmic agents and narrows their therapeutic window, and AAD use in older AF patients has been associated with a higher risk of hospitalization [30]. Class IC agents such as propafenone are contraindicated in the presence of structural heart disease and ischaemia, which is common in elderly cohorts; sotalol carries a dose- and renal-function-dependent risk of QT prolongation and torsades de pointes; and although amiodarone is among the most effective agents, its cumulative thyroid, pulmonary, hepatic, and neurological toxicity limits long-term use, especially in older patients. Digoxin, still prescribed in 15% of our population, has a narrow therapeutic index that is particularly hazardous in the setting of the reduced renal function and polypharmacy typical of elderly patients. At the same time, contemporary evidence indicates that a rhythm-control strategy may confer benefits beyond symptom relief, including a possibly lower risk of new-onset dementia, and that catheter ablation is a reasonable option in selected, fit elderly patients [29, 30]. Taken together, the low AAD and procedural rhythm-control rates observed in Turkish registries likely reflect a combination of appropriate caution in a high-risk elderly population and limited access to advanced rhythm-control infrastructure. These findings underscore the need for individualised, age-adapted rhythm-management strategies that carefully weigh the efficacy of AADs and ablation against their safety in older AF patients.

Several limitations should be considered when interpreting our findings. First, all included data sources were observational and registry-based studies. Although this reflects routine clinical practice and was an explicit inclusion criterion, registry populations are subject to selection bias: patients captured in Turkish AF registries are frequently those managed at tertiary or specialist cardiology centres and those with more complete documentation, and may differ systematically from the wider AF population in disease severity, comorbidity burden, and access to anticoagulation. The pooled estimates may therefore not be fully generalisable to patients managed in primary care or underserved regions. Second, only six registries met the inclusion criteria, and several key estimates were derived from a small subset of these. In particular, the treatment-pattern analyses were affected by the design of the source registries: three registries (NOAC-TURK, NOAC-TR, ANATOLIA-AF) enrolled patients on the basis of NOAC therapy by design and could not contribute unbiased estimates of NOAC-versus-VKA utilisation. Although we addressed this by stratifying by registry type and restricting the primary treatment-pattern estimates to all-comer registries, the latter analyses rested on only two registries (RAMSES and ROTA), which limits their precision and the strength of any temporal inference. Consequently, the primary treatment-pattern estimates carried wide confidence intervals and should be interpreted with corresponding caution; the full-sample pooled estimates, which combine all-comer and NOAC-only registries, are inflated by the enrolment design of the latter and do not represent population-level anticoagulant use. Third, there was substantial clinical and methodological heterogeneity across the included registries, and between-study heterogeneity was very high for most outcomes (I² frequently exceeding 95%). The contributing registries differed in their underlying purpose, geographic and temporal coverage, and case-ascertainment strategies. AF-specific definitions were not standardised between sources: the classification of AF type (paroxysmal versus persistent/permanent), the calculation and reporting of CHA₂DS₂-VASc and HAS-BLED scores, and the categorisation of oral anticoagulant use varied across registries, raising the possibility that ostensibly similar categories captured clinically distinct groups of patients. This heterogeneity limits the degree to which the individual data sources can be considered directly comparable. Fourth, outcome reporting was incomplete and inconsistent. Not all registries reported every outcome of interest, and several clinically important parameters, including time-in-therapeutic-range for VKA therapy, adherence and persistence with NOACs, and long-term clinical outcomes such as stroke and major bleeding, were not uniformly available. The absence of individual patient data precluded subgroup analyses by sex, age, or comorbidity burden and prevented harmonised handling of missing data, which was managed differently across the original studies. These factors may bias the pooled estimates in either direction and reduce their precision and comparability. Fifth, the predominantly cross-sectional design of most included registries limits temporal and causal inference; accordingly, the patterns reported here should be interpreted as descriptive of real-world practice rather than as evidence of causal effect. As the source registries did not routinely collect all variables influencing treatment allocation and prognosis, the potential for residual and unmeasured confounding, including indication bias, cannot be excluded. Finally, temporal changes in treatment availability and clinical guidelines over the period spanned by the registries may have influenced the pooled estimates, and this review was not prospectively registered in a public database. Although we retained the most comprehensive dataset when the same cohort was reported more than once, the registries partially overlapped in participating centres and recruitment periods, so patient-level overlap across sources cannot be entirely excluded. Taken together, these limitations indicate that our results should be interpreted with appropriate caution and highlight the need for prospective studies using harmonised, AF-specific protocols and standardised outcome definitions to confirm and extend the present observations.

Conclusions

This meta-analysis, which incorporates registry-based and multicenter observational studies, highlights notable progress in pharmacological stroke prevention among patients with AF in Turkey, as reflected by greater NOAC use in more recent all-comer registries. A moderate-to-high thromboembolic risk and a low-to-moderate bleeding risk profile characterize patients with AF in Turkey. Nonetheless, the low observed rates of rhythm-control strategies and the persistent variability observed across registries highlight ongoing deficiencies in AF management. In particular, the design-adjusted (all-comer) treatment-pattern estimates are based on only two registries and are imprecise, whereas the pooled estimates combining all registry types do not represent population-level anticoagulant use; larger, uniformly designed national registries are needed to provide precise and representative estimates. These findings emphasize the need for system-level improvements, enhanced infrastructure for rhythm management, and broader implementation of evidence-based clinical guidelines to optimize outcomes for AF patients in Turkey.

Authors' contributions

MK, MEA, BS, MZ: Conceptualization; Literature search; Study selection; Data extraction; Risk-of-bias assessment; Quality appraisal; Validation; Writing—original draft; Writing—review & editing. MEA and MK: Conceptualization; Methodology; Formal analysis and meta-analysis; Risk-of-bias assessment; Investigation; Data curation oversight; Heterogeneity analyses; Interpretation of results; Supervision; Project administration; Writing—review & editing. All authors meet ICMJE authorship criteria, agree to be accountable for all aspects of the work, and approved the final version for submission.

Funding

None.

Data availability

The datasets used and/or analyzed in this study are available upon reasonable request from the corresponding author.

Declarations

Ethics approval and consent to participate

Ethical approval is not required for this study.

Consent for publication

Not Applicable.

Competing interests

The authors declare no competing interests.