European Journal of Cardiovascular Medicine

Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia worldwide and represents a major contributor to cardiovascular morbidity, mortality, and healthcare utilization [1]. Its prevalence increases markedly with advancing age and the burden of comorbidities such as hypertension, diabetes mellitus, heart failure, and structural heart disease [2]. Importantly, AF is a potent risk factor for ischemic stroke, heart failure hospitalization, cognitive decline, and all-cause mortality. Early identification of individuals at risk for AF, particularly before the onset of overt or persistent arrhythmia, remains a critical unmet clinical need, as timely initiation of rhythm monitoring, anticoagulation, and risk factor modification can substantially reduce adverse outcomes [3].

Paroxysmal atrial fibrillation (PAF) poses a unique diagnostic challenge [4]. By definition, PAF is intermittent and often asymptomatic, with episodes that may be brief and self-terminating [5]. Standard 12-lead electrocardiography (ECG), which is typically recorded during sinus rhythm in routine clinical practice, frequently fails to capture these transient arrhythmic events [6]. Even extended rhythm monitoring strategies such as Holter monitoring or event recorders have limited sensitivity, particularly when AF burden is low [7]. Consequently, a substantial proportion of individuals with PAF remain undiagnosed until they present with complications such as stroke [8]. This diagnostic gap has driven growing interest in novel strategies that can identify latent susceptibility to AF even when the ECG demonstrates normal sinus rhythm.

In parallel with these clinical challenges, advances in artificial intelligence (AI), particularly machine learning (ML) and deep learning (DL), have transformed the analysis of biomedical signals [9]. The ECG, a low-cost, noninvasive, and ubiquitously available diagnostic tool, has emerged as a promising substrate for AI-based analysis [10]. Unlike traditional ECG interpretation, which relies on visually discernible features and predefined criteria, AI models can detect complex, high-dimensional patterns within raw ECG waveforms that may be imperceptible to human observers [11]. These latent signatures may reflect subtle atrial structural remodeling, conduction heterogeneity, autonomic influences, or electrophysiological alterations that precede the clinical manifestation of AF [12].

Over the past decade, multiple AI-enabled ECG models have been developed to identify AF or predict its future occurrence from ECGs recorded during sinus rhythm [13]. Initial studies demonstrated that deep neural networks, particularly convolutional neural networks, could accurately distinguish individuals with a history of AF from those without, even when analyzing sinus rhythm ECGs alone [14]. Subsequent investigations extended this paradigm to the prediction of incident or paroxysmal AF, raising the possibility that a single resting ECG could serve as a powerful screening and risk stratification tool [15]. Such an approach has substantial clinical appeal, as it could be seamlessly integrated into routine workflows across outpatient clinics, emergency departments, and population-level screening programs [16].

However, despite the growing volume of literature in this field, several important limitations remain. Many early AI-ECG studies were conducted in single-center cohorts, often using retrospective designs and internally derived validation datasets [17]. While these studies demonstrated promising discriminatory performance, internal validation alone is insufficient to establish clinical generalizability. Model performance may degrade substantially when applied to new populations with different demographic characteristics, comorbidity profiles, ECG acquisition systems, or AF ascertainment strategies [18]. External validation—defined as testing model performance in independent cohorts not used during model development—is therefore essential before such tools can be considered for widespread clinical adoption [19].

Furthermore, the reported performance metrics across studies are heterogeneous. Some investigations report area under the receiver operating characteristic curve (AUC) alone, whereas others provide sensitivity, specificity, predictive values, or complete 2×2 contingency data. Prediction horizons, reference standards for AF diagnosis, duration and intensity of rhythm monitoring, and definitions of paroxysmal or incident AF vary considerably [20]. Differences also exist in ECG characteristics, including the use of standard 12-lead versus single-lead recordings, sampling frequencies, and preprocessing techniques [21]. Collectively, this heterogeneity complicates direct comparison across studies and obscures the true diagnostic and predictive accuracy of AI-based ECG approaches in sinus rhythm [22].

To date, although narrative reviews and methodological commentaries have summarized the potential of AI-enabled electrocardiography for AF detection, a focused quantitative synthesis of external validation studies is lacking [23]. In particular, no comprehensive meta-analysis has systematically evaluated the diagnostic accuracy of AI-based ECG models applied to sinus rhythm ECGs for the prediction of paroxysmal or future AF, while explicitly restricting inclusion to externally validated cohorts [24]. Such an analysis is essential to provide pooled estimates of performance, explore sources of heterogeneity, and assess the robustness and clinical applicability of these models across diverse settings.

Accordingly, the present study aims to systematically review and meta-analyze external validation studies evaluating artificial intelligence–based electrocardiography in sinus rhythm for the prediction of paroxysmal or incident atrial fibrillation. By synthesizing available evidence using established diagnostic accuracy meta-analytic methods and rigorously assessing risk of bias and applicability, this review seeks to clarify the current state of the field, identify gaps in evidence, and inform future research and clinical translation of AI-enabled ECG technologies for AF risk prediction.

Literature Search A comprehensive literature search was conducted across PubMed/MEDLINE, Embase, Scopus, and Web of Science from database inception to the most recent update. Search terms combined controlled vocabulary and free-text keywords related to artificial intelligence, machine learning, deep learning, electrocardiography, and atrial fibrillation, with specific emphasis on sinus rhythm and prediction or detection of paroxysmal or incident atrial fibrillation. Reference lists of relevant reviews and included studies were manually screened to identify additional eligible articles. Only studies involving human participants and published in English were considered. The complete search strategy is provided in the Supplementary Appendix. Registered with Prospero CRD420251272967. Prisma Guideline was processed [25]. Study Selection and Data Extraction All records identified through the database search were imported into a reference management software, and duplicate citations were removed. Two reviewers independently screened titles and abstracts for eligibility based on predefined inclusion and exclusion criteria. Full-text articles were subsequently assessed for inclusion, with disagreements resolved by consensus or consultation with a third reviewer. Eligible studies included external validation cohorts evaluating artificial intelligence–based electrocardiography performed during sinus rhythm for prediction of paroxysmal or incident atrial fibrillation. Data were independently extracted using a standardized, prepiloted form capturing study design, population characteristics, ECG acquisition parameters, AI model architecture, reference standards for atrial fibrillation diagnosis, and performance metrics. When required data were unclear or incomplete, corresponding authors were contacted for clarification. Statistical Analysis and Risk of Bias Meta-analyses were performed using random-effects models to account for between-study heterogeneity. When available, 2×2 contingency data were extracted and pooled using bivariate or hierarchical summary receiver operating characteristic models. Otherwise, pooled estimates of area under the receiver operating characteristic curve with 95% confidence intervals were calculated. Statistical heterogeneity was assessed using the I² statistic. Prespecified subgroup analyses were conducted based on ECG type and atrial fibrillation ascertainment. All analyses were conducted using Stata version 18.0. Risk of bias was independently assessed using the ROBINS-I tool across seven domains, with each study categorized as low, moderate, serious, or critical risk of bias [26].

Demographics

A total of six studies [27-32] were included in the final quantitative synthesis, comprising an aggregate population of 441,593 participants. Of these, 232,491 (52.6%) were male and 208,736 (47.4%) were female. The pooled mean age of the study population was 66.9 years, reflecting an older cohort at increased risk for atrial fibrillation. The mean follow-up duration across studies was 12 months. Regarding baseline comorbidities, hypertension was the most prevalent condition, present in 91,057 participants, followed by diabetes mellitus in 49,280, heart failure in 26,365, and prior stroke in 71,080 individuals. Coronary artery disease was reported in 7,338 participants. Overall, the included studies represented a high-risk cardiovascular population undergoing artificial intelligence–based electrocardiographic assessment during sinus rhythm.

Pooled Accuracy of AI-Based ECG Models for Prediction of Paroxysmal Atrial Fibrillation

This forest plot summarizes the accuracy of artificial intelligence–enabled electrocardiographic models developed to predict future paroxysmal atrial fibrillation from sinus rhythm ECG recordings across external validation studies. Each study is represented by a square indicating the point estimate of accuracy, with the horizontal line denoting the corresponding 95% confidence interval; the size of the square reflects the relative weight of the study in the meta-analysis. The pooled estimate is depicted by the diamond and was calculated using a random-effects restricted maximum likelihood (REML) model to account for between-study variability. Overall, the meta-analysis demonstrated a pooled accuracy of 72.32% (95% CI: 59.96–84.67). Notably, substantial heterogeneity was observed among the included studies (I² = 99.96%), highlighting considerable variation in model performance across different populations, ECG acquisition settings, and AI architectures.

Pooled Precision of AI-Based ECG Models for Prediction of Paroxysmal Atrial Fibrillation

This forest plot illustrates the precision (positive predictive value) of artificial intelligence–enabled electrocardiographic models applied to sinus rhythm ECGs for the prediction of paroxysmal atrial fibrillation across external validation studies. Individual study estimates are represented by squares with corresponding 95% confidence intervals, while the size of each square reflects the relative statistical weight of the study. The pooled precision estimate, depicted by the diamond, was calculated using a random-effects restricted maximum likelihood (REML) model to account for between-study variability. Overall, the pooled precision was 72.32% (95% CI: 60.79–83.85). Substantial heterogeneity was observed among studies (I² = 99.99%), indicating marked variability in positive predictive performance across different populations, ECG acquisition methods, and AI model architectures.

Pooled Recall (Sensitivity) of AI-Based ECG Models for Prediction of Paroxysmal Atrial Fibrillation

This forest plot depicts the recall (sensitivity) of artificial intelligence–enabled electrocardiographic models applied to sinus rhythm ECGs for the prediction of paroxysmal atrial fibrillation across external validation studies. Each square represents the study-specific recall estimate with corresponding 95% confidence intervals, while the size of the square reflects the relative contribution of each study to the pooled analysis. The diamond represents the overall pooled recall, estimated using a random-effects restricted maximum likelihood (REML) model to account for between-study heterogeneity. The pooled recall was 77.53% (95% CI: 70.49–84.56), indicating a moderate-to-high ability of AI-based ECG models to correctly identify individuals who subsequently developed paroxysmal atrial fibrillation. Considerable heterogeneity was observed across studies (I² = 99.83%), underscoring variability in sensitivity related to differences in study populations, ECG acquisition protocols, and AI model architectures.

Pooled F1 Score of AI-Based ECG Models for Prediction of Paroxysmal Atrial Fibrillation

This forest plot presents the F1 score of artificial intelligence–enabled electrocardiographic models used for predicting paroxysmal atrial fibrillation from sinus rhythm ECGs across external validation studies. The F1 score represents the harmonic mean of precision and recall, providing a balanced measure of model performance, particularly in the context of class imbalance. Each square denotes the study-specific F1 score with corresponding 95% confidence intervals, and the size of the square reflects the relative weight assigned to each study in the meta-analysis. The pooled F1 score, represented by the diamond, was estimated using a random-effects restricted maximum likelihood (REML) model. The overall pooled F1 score was 67.22% (95% CI: 51.11–83.33). Marked between-study heterogeneity was observed (I² = 99.97%), indicating substantial variability in the balance between sensitivity and precision across different AI models, ECG datasets, and study populations.

This systematic review and meta-analysis evaluated the diagnostic performance of artificial intelligence (AI)–enabled electrocardiography applied to sinus rhythm ECGs for the prediction of future paroxysmal atrial fibrillation (PAF) across external validation studies [33]. By synthesizing evidence from multiple independent cohorts, the present analysis provides a comprehensive assessment of model generalizability and real-world performance, which remains a critical gap in the translation of AI-based cardiovascular diagnostics into routine clinical practice [34].

The pooled accuracy of approximately 72% indicates that AI-based ECG models can correctly classify a substantial proportion of individuals at risk for developing PAF using standard sinus rhythm ECG recordings [35]. This finding supports the underlying biological premise that subtle, subclinical electrical remodeling precedes clinically overt atrial fibrillation and can be detected through advanced machine-learning techniques. However, accuracy alone may mask clinically important trade-offs between false-positive and false-negative predictions, particularly in populations with low disease prevalence [36].

Precision, reflecting the proportion of correctly identified true positives among all positive predictions, was also moderate at approximately 72%. This suggests that a meaningful proportion of individuals flagged by AI algorithms may not subsequently develop PAF, which has implications for downstream testing, patient anxiety, and healthcare resource utilization [37]. In contrast, the pooled recall of nearly 78% demonstrates that AI-based ECG models exhibit a relatively strong ability to identify individuals who will eventually manifest PAF [38]. From a clinical screening perspective, higher sensitivity is desirable, particularly when the goal is early identification of patients who may benefit from closer rhythm monitoring or risk factor modification [39].

The pooled F1 score of approximately 67% highlights the inherent tension between sensitivity and precision across models and underscores the challenge of optimizing balanced performance in heterogeneous real-world datasets. The relatively lower F1 score compared with recall suggests that many models prioritize sensitivity at the expense of false-positive rates. While this strategy may be acceptable in screening contexts, it may be less suitable for definitive diagnostic decision-making without confirmatory testing.

A consistent and notable finding across all performance metrics was the presence of extreme between-study heterogeneity, with I² values exceeding 99% in all analyses. This heterogeneity likely reflects substantial variability in study populations, ECG acquisition methods, lead configurations, sampling frequencies, outcome definitions, and AI model architectures. Differences in prediction horizons and labeling strategies for PAF further complicate cross-study comparisons. Additionally, variation in prevalence of atrial fibrillation and baseline cardiovascular risk across cohorts may have significantly influenced performance metrics, particularly precision and F1 score.

These findings have important implications for clinical implementation. While AI-enabled ECG analysis demonstrates promising predictive capability, its performance is highly context-dependent. Models trained in specific populations or healthcare systems may not generalize well without recalibration or retraining. Therefore, external validation in diverse populations and transparent reporting of model development and evaluation, in line with TRIPOD-AI and PRISMA-AI recommendations, are essential before widespread adoption.

Future research should focus on standardizing outcome definitions, reporting thresholds, and performance metrics, as well as exploring the integration of AI-ECG outputs with clinical risk scores and imaging biomarkers. Prospective impact studies are also needed to determine whether AI-based prediction of PAF leads to improved clinical outcomes, such as reduced stroke incidence or earlier initiation of preventive strategies. Overall, AI-enabled ECG represents a promising but evolving tool for early atrial fibrillation risk stratification, warranting cautious interpretation and rigorous validation.

This systematic review and meta-analysis demonstrate that artificial intelligence–enabled electrocardiography applied to sinus rhythm ECGs shows moderate predictive performance for identifying individuals at risk of developing paroxysmal atrial fibrillation. The pooled accuracy, precision, recall, and F1 score indicate meaningful discriminatory capability, supporting the concept that subclinical electrical remodeling can be detected prior to overt arrhythmia. However, substantial between-study heterogeneity highlights considerable variability in model performance across populations, ECG acquisition settings, and AI architectures. Future studies should prioritize standardized reporting, robust external validation, and prospective outcome-driven evaluations before routine clinical implementation. Conflict of Interest The authors certify that there is no conflict of interest with any financial organization regarding the material discussed in the manuscript. Funding The authors report no involvement in the research by the sponsor that could have influenced the outcome of this work. Authors’ contributions. All authors contributed equally to the manuscript and read and approved the final version of the manuscript. Acknowledgement This paper is the collaborative work of all authors under the mentorship for the research work from BIR (Biomedical and International Research). We all authors acknowledge this mentorship for this meta-analysis.

1. Karatela MF, Calkins H. The Global Impact of Atrial Fibrillation. Arrhythm Electrophysiol Rev. 2025 Nov 8;14:e28. doi: 10.15420/aer.2025.33. PMID: 41346409; PMCID: PMC12673497.

2. Matheus AS, Tannus LR, Cobas RA, Palma CC, Negrato CA, Gomes MB. Impact of diabetes on cardiovascular disease: an update. Int J Hypertens. 2013;2013:653789. doi: 10.1155/2013/653789. Epub 2013 Mar 4. PMID: 23533715; PMCID: PMC3603160.

3. Schnabel RB, Marinelli EA, Arbelo E, Boriani G, Boveda S, Buckley CM, Camm AJ, Casadei B, Chua W, Dagres N, de Melis M, Desteghe L, Diederichsen SZ, Duncker D, Eckardt L, Eisert C, Engler D, Fabritz L, Freedman B, Gillet L, Goette A, Guasch E, Svendsen JH, Hatem SN, Haeusler KG, Healey JS, Heidbuchel H, Hindricks G, Hobbs FDR, Hübner T, Kotecha D, Krekler M, Leclercq C, Lewalter T, Lin H, Linz D, Lip GYH, Løchen ML, Lucassen W, Malaczynska-Rajpold K, Massberg S, Merino JL, Meyer R, Mont L, Myers MC, Neubeck L, Niiranen T, Oeff M, Oldgren J, Potpara TS, Psaroudakis G, Pürerfellner H, Ravens U, Rienstra M, Rivard L, Scherr D, Schotten U, Shah D, Sinner MF, Smolnik R, Steinbeck G, Steven D, Svennberg E, Thomas D, True Hills M, van Gelder IC, Vardar B, Palà E, Wakili R, Wegscheider K, Wieloch M, Willems S, Witt H, Ziegler A, Daniel Zink M, Kirchhof P. Early diagnosis and better rhythm management to improve outcomes in patients with atrial fibrillation: the 8th AFNET/EHRA consensus conference. Europace. 2023 Feb 8;25(1):6-27. doi: 10.1093/europace/euac062. PMID: 35894842; PMCID: PMC9907557.

4. Howlett PJ, Hatch FS, Alexeenko V, Jabr RI, Leatham EW, Fry CH. Diagnosing Paroxysmal Atrial Fibrillation: Are Biomarkers the Solution to This Elusive Arrhythmia? Biomed Res Int. 2015;2015:910267. doi: 10.1155/2015/910267. Epub 2015 Jul 1. PMID: 26229966; PMCID: PMC4502272.

5. Roche F, Gaspoz JM, Da Costa A, Isaaz K, Duverney D, Pichot V, Costes F, Lacour JR, Barthélémy JC. Frequent and prolonged asymptomatic episodes of paroxysmal atrial fibrillation revealed by automatic long-term event recorders in patients with a negative 24-hour Holter. Pacing Clin Electrophysiol. 2002 Nov;25(11):1587-93. doi: 10.1046/j.1460-9592.2002.01587.x. PMID: 12494616.

6. Rizzo C, Monitillo F, Iacoviello M. 12-lead electrocardiogram features of arrhythmic risk: A focus on early repolarization. World J Cardiol. 2016 Aug 26;8(8):447-55. doi: 10.4330/wjc.v8.i8.447. PMID: 27621772; PMCID: PMC4997525.

7. Lewalter T, Boriani G. Relevance of Monitoring Atrial Fibrillation in Clinical Practice. Arrhythm Electrophysiol Rev. 2012 Sep;1(1):54-58. doi: 10.15420/aer.2012.1.54. PMID: 26835031; PMCID: PMC4711515.

8. Turakhia MP, Guo JD, Keshishian A, Delinger R, Sun X, Ferri M, Russ C, Cato M, Yuce H, Hlavacek P. Contemporary prevalence estimates of undiagnosed and diagnosed atrial fibrillation in the United States. Clin Cardiol. 2023 May;46(5):484-493. doi: 10.1002/clc.23983. Epub 2023 Mar 1. PMID: 36855960; PMCID: PMC10189075.

9. Fahim YA, Hasani IW, Kabba S, Ragab WM. Artificial intelligence in healthcare and medicine: clinical applications, therapeutic advances, and future perspectives. Eur J Med Res. 2025 Sep 23;30(1):848. doi: 10.1186/s40001-025-03196-w. PMID: 40988064; PMCID: PMC12455834.

10. Siontis KC, Noseworthy PA, Attia ZI, Friedman PA. Artificial intelligence-enhanced electrocardiography in cardiovascular disease management. Nat Rev Cardiol. 2021 Jul;18(7):465-478. doi: 10.1038/s41569-020-00503-2. Epub 2021 Feb 1. PMID: 33526938; PMCID: PMC7848866.

11. Martínez-Sellés M, Marina-Breysse M. Current and Future Use of Artificial Intelligence in Electrocardiography. J Cardiovasc Dev Dis. 2023 Apr 17;10(4):175. doi: 10.3390/jcdd10040175. PMID: 37103054; PMCID: PMC10145690.

12. Iwamiya S, Ihara K, Nitta G, Sasano T. Atrial Fibrillation and Underlying Structural and Electrophysiological Heterogeneity. Int J Mol Sci. 2024 Sep 23;25(18):10193. doi: 10.3390/ijms251810193. PMID: 39337682; PMCID: PMC11432636.

13. Nasser A, Michalczak M, Żądło A, Tokarek T. AI-Powered Precision: Revolutionizing Atrial Fibrillation Detection with Electrocardiograms. J Clin Med. 2025 Jul 11;14(14):4924. doi: 10.3390/jcm14144924. PMID: 40725616; PMCID: PMC12295437.

14. Tsiartas E, Nayak D, Meade A. Artificial intelligence capabilities in identifying atrial fibrillation using baseline sinus rhythm ECG : a systematic review. Open Heart. 2025 Oct 31;12(2):e003657. doi: 10.1136/openhrt-2025-003657. PMID: 41173515; PMCID: PMC12581083.

15. Masumura M, Ohno A, Yoshinaga H, Sasaki T, Yamauchi Y, Hachiya H, Takahashi A, Imai Y, Fujita H, Ihara K, Ebana Y, Tanaka T, Furukawa T, Sasano T. AI-ECG for early detection of atrial fibrillation: First-year results from a stroke prevention study in Shimizu, Japan. J Arrhythm. 2025 Jul 4;41(4):e70132. doi: 10.1002/joa3.70132. PMID: 40621219; PMCID: PMC12227326.

16. Bowens FM, Frye PA, Jones WA. Health information technology: integration of clinical workflow into meaningful use of electronic health records. Perspect Health Inf Manag. 2010 Oct 1;7(Fall):1d. PMID: 21063545; PMCID: PMC2966355.

17. Niazai A, Jamil H, Hameed M, Sheikh S, Nisar MR. Artificial intelligence in cardiovascular diagnostics: a systematic review and descriptive analysis of clinical applications and diagnostic performance. BMC Cardiovasc Disord. 2025 Nov 28;25(1):849. doi: 10.1186/s12872-025-05327-x. PMID: 41315941; PMCID: PMC12661670.

18. Gulati G, Upshaw J, Wessler BS, Brazil RJ, Nelson J, van Klaveren D, Lundquist CM, Park JG, McGinnes H, Steyerberg EW, Van Calster B, Kent DM. Generalizability of Cardiovascular Disease Clinical Prediction Models: 158 Independent External Validations of 104 Unique Models. Circ Cardiovasc Qual Outcomes. 2022 Apr;15(4):e008487. doi: 10.1161/CIRCOUTCOMES.121.008487. Epub 2022 Mar 31. PMID: 35354282; PMCID: PMC9015037.

19. Arun S, Grosheva M, Kosenko M, Robertus JL, Blyuss O, Gabe R, Munblit D, Offman J. Systematic scoping review of external validation studies of AI pathology models for lung cancer diagnosis. NPJ Precis Oncol. 2025 Jun 7;9(1):166. doi: 10.1038/s41698-025-00940-7. PMID: 40483288; PMCID: PMC12145442.

20. Florkowski CM. Sensitivity, specificity, receiver-operating characteristic (ROC) curves and likelihood ratios: communicating the performance of diagnostic tests. Clin Biochem Rev. 2008 Aug;29 Suppl 1(Suppl 1):S83-7. PMID: 18852864; PMCID: PMC2556590.

21. Fabricius Ekenberg L, Høfsten DE, Rasmussen SM, Mølgaard J, Hasbak P, Sørensen HBD, Meyhoff CS, Aasvang EK. Wireless Single-Lead versus Standard 12-Lead ECG, for ST-Segment Deviation during Adenosine Cardiac Stress Scintigraphy. Sensors (Basel). 2023 Mar 9;23(6):2962. doi: 10.3390/s23062962. PMID: 36991673; PMCID: PMC10051714.

22. Gunjal A, Judgi T. A comprehensive survey of artificial intelligence methods for cardiovascular disease detection: Recent advances and future challenges. MethodsX. 2025 Oct 15;15:103678. doi: 10.1016/j.mex.2025.103678. PMID: 41216029; PMCID: PMC12596998.

23. Velandia H, Pardo A, Vera MI, Vera M. Systematic Review of Artificial Intelligence and Electrocardiography for Cardiovascular Disease Diagnosis. Bioengineering (Basel). 2025 Nov 14;12(11):1248. doi: 10.3390/bioengineering12111248. PMID: 41301204; PMCID: PMC12649912.

24. Elantary R, Othman S. Artificial Intelligence in Electrocardiography: From Automated Arrhythmia Detection to Predicting Hidden Cardiovascular Disease. Cureus. 2025 Oct 7;17(10):e94065. doi: 10.7759/cureus.94065. PMID: 41069568; PMCID: PMC12504586.

25. Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, Shamseer L, Tetzlaff JM, Akl EA, Brennan SE, Chou R, Glanville J, Grimshaw JM, Hróbjartsson A, Lalu MM, Li T, Loder EW, Mayo-Wilson E, McDonald S, McGuinness LA, Stewart LA, Thomas J, Tricco AC, Welch VA, Whiting P, Moher D. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021 Mar 29;372:n71. doi: 10.1136/bmj.n71. PMID: 33782057; PMCID: PMC8005924.

26. Sterne JA, Hernán MA, Reeves BC, Savović J, Berkman ND, Viswanathan M, Henry D, Altman DG, Ansari MT, Boutron I, Carpenter JR, Chan AW, Churchill R, Deeks JJ, Hróbjartsson A, Kirkham J, Jüni P, Loke YK, Pigott TD, Ramsay CR, Regidor D, Rothstein HR, Sandhu L, Santaguida PL, Schünemann HJ, Shea B, Shrier I, Tugwell P, Turner L, Valentine JC, Waddington H, Waters E, Wells GA, Whiting PF, Higgins JP. ROBINS-I: a tool for assessing risk of bias in non-randomised studies of interventions. BMJ. 2016 Oct 12;355:i4919. doi: 10.1136/bmj.i4919. PMID: 27733354; PMCID: PMC5062054.

27. Jeon KH, Kwon JM, Lee MS, Cho YJ, Oh IY, Lee JH. Deep learning-based electrocardiogram analysis detecting paroxysmal atrial fibrillation during sinus rhythm in patients with cryptogenic stroke: validation study using implantable cardiac monitoring. Eur Heart J Digit Health. 2022 Dec 22;3(4):ztac076.2777. doi: 10.1093/ehjdh/ztac076.2777. PMCID: PMC9779873.

28. Jabbour G, Nolin-Lapalme A, Tastet O, Corbin D, Jordà P, Sowa A, Delfrate J, Busseuil D, Hussin JG, Dubé MP, Tardif JC, Rivard L, Macle L, Cadrin-Tourigny J, Khairy P, Avram R, Tadros R. Prediction of incident atrial fibrillation using deep learning, clinical models, and polygenic scores. Eur Heart J. 2024 Dec 7;45(46):4920-4934. doi: 10.1093/eurheartj/ehae595. PMID: 39217446; PMCID: PMC11631091.

29. Schoels M, Krumm L, Nelde A, Olma MC, Nolte CH, Scheitz JF, Klammer MG, Leithner C, Meisel A, Scheibe F, Krämer M, Haeusler KG, Endres M, Meisel C. Artificial intelligence for prediction of atrial fibrillation in the stroke unit: a retrospective derivation validation cohort study. EBioMedicine. 2025 Aug;118:105869. doi: 10.1016/j.ebiom.2025.105869. Epub 2025 Aug 5. PMID: 40752407; PMCID: PMC12341230.

30. Khurshid S, Friedman SF, Al-Alusi MA, Kany S, Sommers T, Anderson CD, Ho JE, McManus DD, Borowsky LH, Ashburner JM, Lubitz SA, Atlas SJ, Maddah M, Singer DE, Ellinor PT. Artificial intelligence-enabled analysis of handheld single-lead electrocardiograms to predict incident atrial fibrillation: an analysis of the VITAL-AF randomized trial. NPJ Digit Med. 2025 Nov 26;8(1):776. doi: 10.1038/s41746-025-02164-2. PMID: 41299008; PMCID: PMC12722369.

31. Christopoulos G, Attia ZI, Van Houten HK, Yao X, Carter RE, Lopez-Jimenez F, Kapa S, Noseworthy PA, Friedman PA. Artificial intelligence-electrocardiography to detect atrial fibrillation: trend of probability before and after the first episode. Eur Heart J Digit Health. 2022 May 9;3(2):228-235. doi: 10.1093/ehjdh/ztac023. PMID: 36713006; PMCID: PMC9707931.

32. Tarabanis C, Koesmahargyo V, Tachmatzidis D, Sousonis V, Bakogiannis C, Ronan R, Bernstein SA, Barbhaiya C, Park DS, Holmes DS, Kushnir A, Yang F, Aizer A, Chinitz LA, Tzeis S, Vassilikos V, Jankelson L. Artificial intelligence-enabled sinus electrocardiograms for the detection of paroxysmal atrial fibrillation benchmarked against the CHARGE-AF score. Eur Heart J Digit Health. 2025 Aug 22;6(6):1134-1144. doi: 10.1093/ehjdh/ztaf100. PMID: 41267852; PMCID: PMC12629645.

33.Christopoulos G, Attia ZI, Van Houten HK, Yao X, Carter RE, Lopez-Jimenez F, Kapa S, Noseworthy PA, Friedman PA. Artificial intelligence-electrocardiography to detect atrial fibrillation: trend of probability before and after the first episode. Eur Heart J Digit Health. 2022 May 9;3(2):228-235. doi: 10.1093/ehjdh/ztac023. PMID: 36713006; PMCID: PMC9707931.

34. Khan S, Qayyum K, Qadeer A, Khalid M, Anthony S, Khan W, Ghulam M, Jamil Z, Anthony N. Efficacy of AI Models in Detecting Heart Failure Using ECG Data: A Systematic Review and Meta-Analysis. Cureus. 2025 Feb 7;17(2):e78683. doi: 10.7759/cureus.78683. PMID: 40065863; PMCID: PMC11891813.

35. Zhou Y, Zhang D, Chen Y, Geng S, Wei G, Tian Y, Shi L, Wang Y, Hong S, Liu X. Screening Tool for Paroxysmal Atrial Fibrillation Based on a Deep-Learning Algorithm Using Printed 12-Lead Electrocardiographic Records during Sinus Rhythm. Rev Cardiovasc Med. 2024 Jul 2;25(7):242. doi: 10.31083/j.rcm2507242. PMID: 39139435; PMCID: PMC11317324.

36. Noseworthy PA, Kaufman ES, Chen LY, Chung MK, Elkind MSV, Joglar JA, Leal MA, McCabe PJ, Pokorney SD, Yao X; American Heart Association Council on Clinical Cardiology Electrocardiography and Arrhythmias Committee; Council on Arteriosclerosis, Thrombosis and Vascular Biology; Council on Cardiovascular and Stroke Nursing; and Stroke Council. Subclinical and Device-Detected Atrial Fibrillation: Pondering the Knowledge Gap: A Scientific Statement From the American Heart Association. Circulation. 2019 Dec 17;140(25):e944-e963. doi: 10.1161/CIR.0000000000000740. Epub 2019 Nov 7. PMID: 31694402; PMCID: PMC8438789.

37. Choudhury A, Asan O. Role of Artificial Intelligence in Patient Safety Outcomes: Systematic Literature Review. JMIR Med Inform. 2020 Jul 24;8(7):e18599. doi: 10.2196/18599. PMID: 32706688; PMCID: PMC7414411.

38. Ozgul O, Habibi Z, Verhaert DVM, Hermans BJM, Philippens SAM, Chaldoupi SM, Maesen B, Isaacs A, Westra SW, Nijveldt R, Bonizzi P, Linz D, Vernooy K, Zeemering S, Schotten U. AI-based ECG analysis outperforms predefined ECG features to predict AF recurrences in patients undergoing AF ablation: Data from the ISOLATION registry. Europace. 2025 May 23;27(Suppl 1):euaf085.833. doi: 10.1093/europace/euaf085.833. PMCID: PMC12099964.

39. Awwad A, Parashar Y, Bagchi S, Siddiqui SA, Ajari O, deFilippi C. Preclinical screening for cardiovascular disease with high-sensitivity cardiac troponins: ready, set, go? Front Cardiovasc Med. 2024 Nov 14;11:1350573. doi: 10.3389/fcvm.2024.1350573. PMID: 39610975; PMCID: PMC11602307.