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.

Background Atrial fibrillation (AF) presents a considerable burden on our health care systems. Early detection of AF may prevent AF-associated complications, such as stroke and heart failure. Given our aging populations, the number of new AF cases is expected to double over the next decades. As such, there is renewed interest to screen for AF in the community. To optimise screening efforts, risk prediction models may help us identify at-risk patients. Purpose To identify and evaluate the performance of prediction models for AF that may be applicable for screening in community settings. Methods We searched PubMed, Embase, and CINAHL databases for studies that derived and/or validated AF risk models from population-based cohorts. Three investigators independently assessed risk of bias (CHARMS checklist), and performed data extraction and evidence synthesis. The primary expression of associations in meta-analysis was the C-statistic for discrimination between AF and non-AF cases during follow-up, using a random effects model. We calculated 95% prediction intervals (PI) due to high heterogeneity (I2 >30%) in all analyses. Results We identified 23 studies that presented data on 8 risk models derived from 18 cohorts with a total of 1,4 million participants from across the globe. Average age in these cohorts ranged from 43–76 years and follow-up ranged from 3 to 20 years. Two of the 8 risk models had a sufficient number of validation studies to be included in the meta-analysis. The CHARGE-AF (Cohorts for Heart and Aging Research in Genomic Epidemiology) score had a summary C-statistic of 0.72 (95%-PI: 0.67–0.77; n=7 cohorts, n=53.040 patients). The FHS (Framingham Heart Study) score for AF had a summary C-statistic of 0.71 (95%-PI: 0.59–0.83; n=4 cohorts, n=19.300 patients). Both models include age, height and weight, blood pressure, prevalent heart failure, and antihypertensive medication use as variables. CHARGE-AF additionally includes race, current smoking, and history of diabetes and myocardial infarction. FHS additionally includes sex, PR interval, and significant murmur. Conclusions Currently two risk scores, CHARGE-AF and FHS, have been rigorously tested for predicting atrial fibrillation in general populations. The CHARGE-AF score may present the more promising, user-friendly score for future community screening efforts, as it solely relies on readily available clinical parameters. Acknowledgement/Funding This work was supported by the Netherlands Organisation for Health Research and Development (ZonMw) [80-83910-98-13046]