Non-traditional risk factors and artificial intelligence in the management of atrial fibrillation

For the podcast associated with this article, please visit https://academic.oup.com/eurheartj/pages/Podcasts. This Focus Issue on arrhythmias, digital health, and innovation contains the State of the Art Review entitled ‘Non-traditional risk factors for atrial fibrillation: epidemiology, mechanisms, and strategies’ by Yingli Lu from the Shanghai JiaoTong University School of Medicine in China, and colleagues.1 Lu et al. point out that atrial fibrillation (AF) has become the predominant arrhythmia worldwide and is associated with high morbidity and mortality.2–8 Its pathogenesis is intricately linked to the deleterious impact of cardiovascular risk factors, emphasizing the pivotal imperative for early detection and mitigation strategies targeting these factors for the prevention of primary AF. While traditional risk factors are well recognized, an increasing number of novel risk factors have been identified in recent decades. This review explores the emerging non-traditional risk factors for the primary prevention of AF, including unhealthy lifestyle factors in current society (sleep, night shift work, and diet), biomarkers (gut microbiota, hyperuricaemia, and homocysteine), adverse conditions or diseases (depression, epilepsy, clonal haematopoiesis of indeterminate potential, infections, and asthma), and environmental factors (acoustic pollution and other environmental factors). Unlike traditional risk factors, individuals have limited control over many of these non-traditional risk factors, posing challenges to conventional prevention strategies. The purpose of this review is to outline the current evidence on the associations of non-traditional risk factors with new-onset AF and the potential mechanisms related to these risk factors. Furthermore, this review aims to explore potential interventions targeting these risk factors at both the individual and societal levels to mitigate the growing burden of AF, suggesting guideline updates for primary AF prevention. Convincing evidence for the efficacy of ablation as first-line therapy in paroxysmal AF (PAF) and its clear superiority to medical therapy for rhythm control in both PAF and persistent AF (PsAF) has generated considerable interest in the optimal timing of ablation.9–13 In a State of the Art Review article entitled ‘Atrial fibrillation ablation timing: where is the sweet spot?’, Jonathan Kalman from the University of Melbourne in Australia, and colleagues indicate that based on these data, there is a widespread view that the principle of ‘the earlier the better’ should be generally applied.14 However, the natural history of AF is highly variable and non-linear, and for this reason it is difficult to be emphatic that all patients are best served by ablation early after their initial AF episodes. Sufficient evidence exists to indicate that a conservative approach is reasonable in patients with infrequent and non-progressive episodes (i.e. absence of a progressive increase in burden culminating in PsAF) in whom symptoms remain mild and well controlled. A conservative management phase should be marked by assiduous attention to risk factor modification, changes in frequency and duration of AF episodes, and patient preferences. If and when AF does begin to progress, accumulating evidence indicates that early ablation accompanied by ongoing attention to risk factors provides the best outcomes. Recent advances have given rise to a spectrum of digital health technologies that are transforming medicine.15,16 These technologies have the potential to revolutionize the design and conduct of cardiovascular clinical trials. In a State of the Art Review article entitled ‘Artificial intelligence and digital tools for design and execution of cardiovascular clinical trials’, Jiun-Ruey Hu from Yale University in New Haven, CT, USA, and colleagues indicate that advances in domain tasks such as automated diagnosis and classification, synthesis of high-volume data and latent data from adjacent modalities, patient discovery, telemedicine, remote monitoring, augmented reality, and in silico modelling have the potential to enhance the efficiency, accuracy, and cost-effectiveness of cardiovascular clinical trials.17 However, early experience with these tools has also exposed important issues, including regulatory barriers, clinical validation and acceptance, technological literacy, integration with care models, and health equity concerns. This narrative review summarizes the landscape of digital tools at each stage of clinical trial planning and execution, and outlines roadblocks and opportunities for successful implementation of digital tools in cardiovascular clinical trials. Identifying the haemodynamic mechanism of autonomic syncope is the essential prerequisite for effective and personalized therapy aimed at preventing recurrences.18 In a Clinical Research article entitled ‘Mechanism of syncope: role of ambulatory blood pressure monitoring and cardiovascular autonomic function assessment’, Antonella Groppelli from the IRCCS Istituto Auxologico Italiano in Brescia, Italy, and colleagues assessed the diagnostic efficacy of a two-step assessment.19 In this multicentre, prospective, cross-sectional, observational study, patients affected by severe autonomic syncope underwent a two-step assessment including 24 h ambulatory blood pressure monitoring (ABPM) and short cardiovascular autonomic function assessment (SCAFA). SCAFA consisted of carotid sinus massage (CSM), performed in patients ≥40 years old, a passive standing test, and a ‘fast’ head-up tilt test scheduled sequentially during one session on a tilt table. The study population consisted of 333 patients. Any positive response was observed in 89% of patients, with hypotensive phenotype accounting for 68%, bradycardic phenotype for 6%, and mixed phenotype for 15% of cases. The mean duration of the SCAFA procedure was 25 (interquartile range [IQR] 20–32) min. ABPM, CSM, passive standing, and head-up tilt test were positive in 60, 15, 3, and 71% of patients, respectively. More than one test was positive in 51% and 49% of patients ≤40 and >40 years, respectively. Large intercentre variability of CSM positivity rate, which remained significant after adjustment for demographic and clinical variables, was observed (P = 0.003) (Figure 1). Detection of the haemodynamic mechanism (phenotype) of autonomic syncope by the two-step standardized assessment. ABPM, ambulatory blood pressure monitoring; CSM, carotid sinus massage; CV, cardiovascular; HUT, head-up tilt test; NTG, nitroglycerine; SCAFA, short cardiovascular autonomic function assessment.19 The authors conclude that the standardized 2STEPS protocol offers an easy-to-perform and time-saving diagnostic work-up allowing identification of the haemodynamic mechanism of loss of consciousness in most patients with autonomic syncope. This protocol provides the necessary background for a personalized mechanism-specific therapy. The manuscript is accompanied by an Editorial by Peter Mitro from the University of Kosice in Slovakia.20 The author notes that ABPM is an important tool not only for the diagnosis but also for the treatment of reflex syncope. From a therapeutical perspective, it appears to be meaningful to also perform ABPM in patients with the head-up tilt test-diagnosed hypotensive phenotype, as it sets the starting point for the interventions aimed at increasing blood pressure. Artificial intelligence (AI) algorithms applied to the 12-lead electrocardiogram (ECG) is a promising age prediction method. In a Clinical Research article entitled ‘Artificial intelligence-derived electrocardiographic ageing and risk of atrial fibrillation: a multi-national study’, Seunghoon Cho from the Yonsei University College of Medicine Seoul, Korea, and colleagues investigated whether the discrepancy between ECG-derived AI-predicted age (AI-ECG age) and chronological age, termed electrocardiographic ageing (ECG ageing), is associated with AF risk.21 An AI-ECG age prediction model was developed using a large-scale dataset (>1 500 000 ECGs from almost 700 000 participants) and validated with six independent and multinational datasets. The AI-ECG age gap was calculated across two South Korean cohorts, one US cohort, and one UK cohort. Participants were classified into two groups: normal group (age gap <7 years) and ECG-aged group (age gap ≥7 years). The predictive capability of ECG ageing for new- and early-onset AF risk was assessed. In the ECG-aged group, increased risks of new-onset AF were observed, with hazard ratios of 2.50, 1.89, 1.90, and 1.76 in the two South Korean, UK, and US cohorts, respectively. For early-onset AF, odds ratios were 2.89, 1.94, 1.58, and 1.79 in these cohorts compared with the normal group (Figure 2). This study investigated whether the discrepancy between artificial intelligence-predicted electrocardiography-derived age (AI-ECG age) and actual age could predict the risk of atrial fibrillation (AF) and further serve as an indicator of cardiac ageing. The AI-ECG age prediction model was developed using >1.5 million ECGs and validated across large-scale, multinational datasets. The results showed that individuals with significant electrocardiographic ageing (ECG ageing) have a higher risk of both new- and early-onset AF across diverse populations, suggesting its potential as a novel biomarker for AF risk. AF, atrial fibrillation; AI, artificial intelligence; CI, confidence interval; ECG, electrocardiogram; ECG ageing, electrocardiographic ageing; F/U, follow-up.21 The authors conclude that AI-derived ECG ageing is associated with the risk of new- and early-onset AF, suggesting its potential utility to identify individuals for AF prevention across diverse populations. The contribution is accompanied by an Editorial by Antonio Ribeiro and Antonio Luiz Ribeiro from the Uppsala University in Sweden.22 The authors highlight that while the AI-ECG presents an exciting new frontier in cardiovascular care, its journey from research to routine clinical practice is still in the early stages. The work by Cho et al. provides a strong foundation, demonstrating the potential of AI-ECG age as a biomarker for AF risk, particularly in diverse populations. However, before it can become a clinically useful tool, further studies are required to confirm its reproducibility, validate its incremental value over existing risk scores, and demonstrate its impact on patient outcomes. As AI revolutionizes cardiology, rigorous validation will be key to ensuring that such innovative tools can be safely and effectively integrated into everyday practice. Robust and convenient risk stratification of patients with paediatric and adult congenital heart disease (CHD) is lacking. In a Clinical Research article entitled ‘Electrocardiogram-based deep learning to predict mortality in paediatric and adult congenital heart disease’, Joshua Mayourian from the Harvard Medical School in Boston, MA, USA and colleagues aims to address this gap with an AI-enhanced ECG tool across the life span of a large, diverse cohort with CHD.23 A convolutional neural network was trained (50%) and tested (50%) on ECGs obtained in the cardiology clinic at the Boston Children's Hospital to detect 5 year mortality. Temporal validation on a contemporary cohort was performed. Model performance was evaluated using the area under the receiver operating characteristic and precision–recall curves. The training and test cohorts composed of 112 804 ECGs (39 784 patients; ECG age range 0–85 years; 4.9% 5 year mortality) and 112 575 ECGs (39 784 patients; ECG age range 0–92 years; 4.6% 5 year mortality from ECG), respectively. Model performance (area under the receiver operating characteristic curve 0.79, 95% confidence interval 0.77–0.81; area under the precision–recall curve 0.17, 95% confidence interval 0.15–0.19) outperformed age at ECG, QRS duration, and left ventricular ejection fraction, and was similar during temporal validation. In subgroup analysis, AI-enhanced ECG outperformed left ventricular ejection fraction across a wide range of CHD lesions. Kaplan–Meier analysis demonstrates predictive value for longer term mortality in the overall cohort and for lesion subgroups. In the overall cohort, pre-cordial lead QRS complexes were most salient, with high-risk features including wide and low-amplitude QRS complexes. Lesion-specific high-risk features such as QRS fragmentation in tetralogy of Fallot were identified. The authors conclude that this temporally validated model shows promise to inexpensively risk-stratify individuals with CHD across their life span, which may inform the timing of imaging/interventions and facilitate improved access to care. The manuscript is accompanied by an Editorial by Evangelos Oikonomou and Rohan Khera from the Yale School of Medicine in New Haven, CT, USA.24 The authors note that conceptually, there is room for further development based on other advances in the field. This could be through the development of larger case repositories, particularly ones that include ECGs from many different vendors, locations, and countries to train algorithms that will be generalizable across heterogeneous populations and geographical cohorts. The next frontier may also include the expansion of AI to multimodal data inputs, such as integrating ECGs with other elements in the Electronic Health Record, including clinical notes through natural language processing and medical imaging through computer vision applications to define other patterns that may provide richer risk stratification. Such approaches have already been used in adult ECGs and echocardiograms. Given the expanding availability of dedicated CHD centres, the growing number of patients with CHD surviving into adulthood, and the extent of phenotypic heterogeneity that is uncovered through the multimodal phenotyping of these patients, CHD is well positioned to make the most of the AI revolution in medicine. The editors hope that this issue of the European Heart Journal will be of interest to its readers. Dr. Crea reports speaker fees from Abbott, Amgen, Astra Zeneca, BMS, Chiesi, Daiichi Sankyo, Menarini outside the submitted work. With thanks to Amelia Meier-Batschelet, Johanna Huggler, and Martin Meyer for help with compilation of this article.