Prognostic models for predicting incident or recurrent atrial fibrillation: protocol for a systematic review

Atrial fibrillation (AF) is the most common arrhythmia diagnosed in clinical practice, and the worldwide incidence and prevalence are increasing [1]. AF is predicted to affect around 18 million people in Europe by 2060, and 6–12 million in the USA by 2050 [1, 2]. Drivers for the increase in prevalence include an ageing population, better survival from conditions such as ischaemic heart disease, but also increasing multimorbidity [3, 4]. AF is associated with increased morbidity, such as stroke or heart failure, and increased mortality, particularly cardiovascular related [3, 4]. Currently available treatments can reduce the mortality and morbidity associated with AF, particularly via anticoagulation for prevention of strokes [4], but patients with AF remain at high risk of cardiovascular complications even on optimal therapy, often manifesting as heart failure or sudden death [5]. AF also commonly occurs after cardiac surgery, with patients subsequently at increased risk of stroke, congestive heart failure, and haemodynamic instability [6]. Post-operative AF rates have been estimated at up to 60% for cardiothoracic surgery and 5–10% in non-cardiothoracic surgery [7].

Given the growing burden of AF, there has been a recent shift in attention to primary and secondary prevention strategies [8, 9]. It has long been known that approximately 70% of patients experience recurrent AF after, e.g. a cardioversion [10], while a small population of AF patients does not show progression of AF, with rare recurrences over decades [11]. Being able to identify those patients who are most likely to develop recurrent AF, or progress from paroxysmal to sustained AF, would be beneficial for tailoring treatment strategies and implementing targeted preventative measures. There is a range of risk factors associated with the development of AF, the best-established ones being age, hypertension, diabetes mellitus, and heart failure, as well as environmental factors such as smoking and alcohol consumption [12, 13]. Less validated risk factors include subclinical hyperthyroidism, obesity and sleep apnoea syndrome [12]. There may also be a role for biomarkers in assessing AF risk, including serum biomarkers such as brain natriuretic peptide (BNP) [14, 15] or fibroblast growth factor 23 (FGF-23) [14], imaging of atrial function, ECG-based parameters, and genetic factors [12].

A number of prognostic models, or risk scores, which attempt to combine several predictors to generate an individualised estimate of risk in different populations have been developed for AF prediction. These include the Cohorts for Ageing and Research in Genomic Epidemiology (CHARGE)-AF score [16], the HAVOC score [17], the ATLAS score [18], the HATCH score [19], and the APPLE score [20].

Scoping searches for existing and ongoing systematic reviews of prognostic models for AF were carried out in MEDLINE, Embase, the Cochrane Library, and PROSPERO (October 2018). There were several systematic reviews relating to either the association of individual risk factors with AF or focussing on a single AF risk prediction model; only one identified systematic review gave an overview of available risk scores; this had a focus on recurrence following catheter ablation [21]. The review had a limited search strategy and did not include formal quality appraisal of the identified models. More recent primary studies were identified since the publication of this review. There is therefore a need for an up-to-date systematic review, which includes all relevant populations and uses methodologically robust methods for study identification, risk of bias assessment, and synthesis.

The increasing burden from AF means that a focus on prevention is becoming more important, and being able to identify patients most at risk of incident or recurrent AF is vital for tailoring management strategies. To the best of our knowledge, this will be the first systematic review to collate evidence from all studies reporting on validated prognostic models, or on the impact of such models, in any population at risk of incident or recurrent AF.

We will use recognised systematic review methods for identifying, appraising, and synthesising the available evidence on existing prognostic models, which will strengthen the robustness of any findings. Publication bias, poor reporting, and extensive heterogeneity are recognised issues in prognostic research [23‐25]. Careful consideration of heterogeneity before analysis will ensure that studies are not inappropriately grouped. The potential impact of publication bias on any findings will be discussed, and recommendations made for future reporting of prognostic models if appropriate.

Should one or more well-validated models with high predictive accuracy be identified, then this will be useful for planning future comparative studies on the impact of using such models in clinical practice. The review will also identify gaps in the evidence, i.e. where model validity is poor or where models are lacking for a particular population. Review findings will therefore inform research recommendations relating to model development, validation, or impact assessment. Findings will be considered in the context of any models already used in clinical practice, and the extent to which these have been validated.