Prediction models for atrial fibrillation applicable in the community: a systematic review and meta-analysis

• Published in: Europace (London, England) Vol. 22; no. 5; pp. 684 - 694
• Main Authors: Himmelreich, Jelle C L, Veelers, Lieke, Lucassen, Wim A M, Schnabel, Renate B, Rienstra, Michiel, van Weert, Henk C P M, Harskamp, Ralf E
• Format: Journal Article
• Language: English
• Published: England Oxford University Press 01.05.2020
• Subjects: Clinical Research; Risk model; Screening; Systematic review; Atrial fibrillation; Community; Meta-analysis
• ISSN: 1099-5129; 1532-2092
• DOI: 10.1093/europace/euaa005

Abstract Aims Atrial fibrillation (AF) is a common arrhythmia associated with an increased stroke risk. The use of multivariable prediction models could result in more efficient primary AF screening by selecting at-risk individuals. We aimed to determine which model may be best suitable for increasing efficiency of future primary AF screening efforts. Methods and results We performed a systematic review on multivariable models derived, validated, and/or augmented for AF prediction in community cohorts using Pubmed, Embase, and CINAHL (Cumulative Index to Nursing and Allied Health Literature) through 1 August 2019. We performed meta-analysis of model discrimination with the summary C-statistic as the primary expression of associations using a random effects model. In case of high heterogeneity, we calculated a 95% prediction interval. We used the CHARMS (Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies) checklist for risk of bias assessment. We included 27 studies with a total of 2 978 659 unique participants among 20 cohorts with mean age ranging from 42 to 76 years. We identified 21 risk models used for incident AF risk in community cohorts. Three models showed significant summary discrimination despite high heterogeneity: CHARGE-AF (Cohorts for Heart and Aging Research in Genomic Epidemiology) [summary C-statistic 0.71; 95% confidence interval (95% CI) 0.66–0.76], FHS-AF (Framingham Heart Study risk score for AF) (summary C-statistic 0.70; 95% CI 0.64–0.76), and CHA2DS2-VASc (summary C-statistic 0.69; 95% CI 0.64–0.74). Of these, CHARGE-AF and FHS-AF had originally been derived for AF incidence prediction. Only CHARGE-AF, which comprises easily obtainable measurements and medical history elements, showed significant summary discrimination among cohorts that had applied a uniform (5-year) risk prediction window. Conclusion CHARGE-AF appeared most suitable for primary screening purposes in terms of performance and applicability in older community cohorts of predominantly European descent.