Subclinical Atherosclerosis Is Associated With Incident Atrial Fibrillation

A Systematic Review and Meta-analysis

Kit Engedal Kristensen; Cille Cederholm Knage; Liv Havgaard Nyhegn; Bart A. Mulder; Michiel Rienstra; Isabelle C. Van Gelder; Axel Brandes


Europace. 2020;22(7):991-1000. 

In This Article


Study Selection and Data Extraction

This systematic review was conducted according to PRISMA standards.[8] The review protocol is registered at PROSPERO—the international prospective register of systematic reviews (; CRD42018114818). In October 2018, a systematic literature search was conducted with guidance from a medical librarian in the medical literature databases MEDLINE (OVID interface, 1990 onwards), EMBASE (OVID interface, 1990 onwards), and Cochrane Library (Wiley, 1999 onwards) using subject headings (MeSH and Emtree) and text words related to atherosclerosis, cIMT, CACS, and AF. We updated the literature search towards the end of the review in January 2019. The full literature search strategy is presented in Supplementary material online, Table S1 and Table S2.

Prospective and retrospective comparative cohort studies, investigating subclinical atherosclerosis as a predictor of AF, were eligible for inclusion. Furthermore, the articles had (i) to be available in an English full-text version; (ii) only to include adults over the age of 18 years; (iii) to include ≥100 participants; and (iv) to have a follow-up period ≥12 month. Authors were contacted if there were uncertainties concerning the articles.

Two reviewers (C.C.K. and L.H.N.) independently sorted the articles by screening titles and abstracts according to the predefined criteria. Hereafter, three reviewers (K.E.K., C.C.K., and L.H.N.) scrutinized the full text of the chosen articles for final selection of the primary literature. A chain search was conducted, in which eligible studies not found in previous searches were identified. Dispute was settled by mutual agreement between the three reviewers.

When multiple articles based on the same cohort and same data appeared in the search strategy, the study best fit was included.

Three reviewers independently extracted data from the seven articles regarding study design, population characteristics, baseline measurements (cIMT and CACS), and outcome measurements (AF). When additional study material was available, it was used to supplement the data extraction.

Validity and Study Quality Assessment

The Newcastle-Ottawa Scale (NOS) for assessing the quality of non-randomized studies in meta-analyses was used to evaluate the risk of bias for each study. It covers selection of study groups, comparability of groups, and ascertainment of exposure and outcomes ( The Cochrane Collaboration recommends the NOS for observational studies ( Three reviewers (K.E.K., C.C.K., and L.H.N.) independently assessed the risk of bias for each study. Dispute was solved by mutual agreement.

Statistical Analysis

Heterogeneity between studies was tested and visualized in forest plots. To quantify the heterogeneity, I 2 was calculated and the percentage evaluated according to the following percent ranges: might not be important (0–40%), may represent moderate heterogeneity (30–60%), may represent substantial heterogeneity (50–90%), and may represent considerable heterogeneity (75–100%). A statistical P-value <0.10 and/or I 2 >50% were considered representing significant heterogeneity.

Two meta-analyses were conducted on the most comprehensively adjusted hazard ratios (HRs) stated in each of the studies. Due to low heterogeneity, a fixed-effects model was used in the meta-analysis of the association between cIMT and AF. A random-effects model allowing for between-study variation was used in the meta-analysis of the association between CACS and AF, because of relatively high heterogeneity of the two included studies (I2 = 76.8%), e.g., due to the marked difference in follow-up time and inclusion criteria (Table 1).[9,10] Hazard ratios were calculated using Cox proportional hazards model. All data analyses were performed using statistical software. STATA (version 15.0; StataCorp LLC, College Station, TX, USA), Covidence (Melbourne, Australia), and EndNote (version X9; Clarivate Analytics, Philadelphia, PA, USA) were used.