Nonalcoholic Fatty Liver Disease Is Associated With Increased Risk of Atrial Fibrillation

Xiaoyan Cai; Sulin Zheng; Ying Liu; Yan Zhang; Jianhua Lu; Yuli Huang


Liver International. 2020;40(7):1594-1600. 

In This Article

Materials and Methods

Search Strategy and Selection Criteria

We performed the study in accordance with the recommendations of the Meta-analysis of Observational Studies in Epidemiology (MOOSE) Group.[12] Electronic databases (PubMed and Embase) were searched for studies until January 31, 2020, using a combined MeSH heading and text search strategy with multiple terms associated with "non-alcoholic fatty liver disease" and "atrial fibrillation". The detailed search strategy for PubMed is presented in Online Supplementary File S1. Reference lists were also manually reviewed to identify other potential studies.

Studies were included for analysis if they met the following criteria: (a) cohort study involving adult individuals (aged ≥18 years); (b) NAFLD and other cardiometabolic risk factors were evaluated at baseline and (c) multivariable-adjusted relative risks (RRs), odds ratios (ORs) or hazard ratios (HRs) and their 95% confidence intervals (CIs) were reported for AF associated with NAFLD versus those without NAFLD. Studies were excluded if a) they were case-control or cross-sectional studies without follow-up evaluation; b) data were derived from the same cohort and c) only unadjusted risk, or age- and/or sex-adjusted risk were reported.

Data Extraction and Quality Assessment

According to the predefined criteria, two investigators (XC and SZ) independently conducted literature searches and reviewed the full copies of potentially suitable studies. Quality of the included studies was evaluated based on the Newcastle-Ottawa Quality Assessment Scale for cohort studies.[13] In this analysis, included studies were graded in quality as good (≥7 points), fair (4-6 points) or poor (<4 points).[14,15] We also evaluated whether the studies had adequately adjusted for potential confounders of at least five of the following seven factors: gender, age, smoking, body mass index or overweight/obesity, hypertension or blood pressure or antihypertensive treatment, fasting plasma glucose or haemoglobin A1c or diabetes mellitus, and serum cholesterol levels or hypercholesterolemia.[16,17]

Data Analysis

Outcomes data adjusted for the maximal number of confounders were extracted for the main analysis. We also compared the pooled outcome data adjusted for the maximal number of variables with those were minimally adjusted (unadjusted or age/gender adjusted), if these data were available, to evaluate the strength of confounders on the estimated RRs. We combined the log RRs and corresponding standard errors by the inverse variance approach. ORs were converted into RRs by the formula (RR = OR/([1 − pRef] + [pRef × OR]), where pRef is the incidence of the outcome (AF) in the reference group (non-NAFLD).[18] HRs were regarded as approximate to RRs.[19] The I2 statistics were used to test heterogeneity. A fixed-effects model was used for meta-analysis if there was no significant heterogeneity among the included studies (I2 < 50%). Otherwise, a random-effects model was used.

We calculated the absolute risk difference for the incidence of AF associated with NAFLD by multiplying the assumed comparator risk by the pooled RR-1.[20] The median risk of AF in non-NAFLD across studies was considered as the assumed comparator risk. Absolute risk difference was expressed in events per 1000 person years. If the pooled RR of AF was increased in NAFLD, we further determined the population-attributable risk (PAR) for NAFLD. PAR% was the proportion of disease in the study population that is attributable to the exposure (NAFLD), and could be eliminated if the exposure was eliminated. The PAR% was calculated as PAR% = (Pe)(RRadjusted − 1)/((Pe)(RRadjusted − 1) + 1)) × 100, where Pe is the proportion of the exposure and RRadjusted indicates the pooled multivariable-adjusted RR.[21]

Subgroup analyses were conducted according to gender (men vs women), ethnicity (Asian vs non-Asian), study design (prospective vs retrospective), age (average <55 years vs ≥ 55 years), methods to define NAFLD (ultrasonography vs computed tomography vs fatty liver index), enrolment population (diabetes vs general population), follow-up duration (<10 years vs ≥ 10 years), sample size (<10 000 vs ≥ 10 000) and adequate adjustment of confounders (yes vs no). We combined studies in each subgroup using the statistical model consistent with the main analysis. Publication bias was evaluated by inspecting funnel plots, and further tested by using Begg's test and Egger's test. Sensitivity analyses were conducted using a fixed-effects model for the meta-analysis, or where the RRs were recalculated by omitting one study at a time to evaluate the impact of individual studies on the estimated risk.

Analyses were performed using RevMan 5.3 (The Cochrane Collaboration, Copenhagen, Denmark). All P values were two tailed, and statistical significance was set at <0.05.