Association of non-alcoholic fatty liver disease with major adverse cardiovascular events: A systematic review and meta-analysis

Increasing evidence connects non-alcoholic fatty liver disease (NAFLD) to cardiovascular disease (CVD). The aim of this study is to assess whether and to what extent the excess risk of CVD is conferred by NAFLD in a meta-analysis. We systematically searched PubMed, EmBase, Web of Science, and Cochrane Library for reports published between 1965 and July 3, 2015. Studies that reported data on association between NAFLD and adverse cardiovascular events or mortality were included. Thirty-four studies (164,494 participants, 21 cross-sectional studies, and 13 cohort studies) were included. NAFLD was not associated with overall mortality (HR = 1.14, 95% CI: 0.99–1.32) and CVD mortality (HR = 1.10, 95% CI: 0.86–1.41). However, NAFLD was associated with an increased risk of prevalent (OR = 1.81, 95% CI: 1.23–2.66) and incident (HR = 1.37, 95% CI: 1.10–1.72) CVD. For some specific CVDs, NAFLD was associated with an increased risk of prevalent (OR = 1.87, 95% CI: 1.47–2.37) and incident (HR = 2.31, 95% CI: 1.46–3.65) coronary artery disease (CAD), prevalent (OR = 1.24, 95% CI: 1.14–1.36) and incident (HR = 1.16, 95% CI: 1.06–1.27) hypertension, and prevalent (OR = 1.32, 95% CI: 1.07–1.62) atherosclerosis. In conclusion, the presence of NAFLD is associated with an increased risk of major adverse cardiovascular events, although it is not related to mortality from all causes or CVD.


NAFLD and prevalent/incident CVD. Cross-sectional studies.
In cross-sectional studies, pooled OR for the presence of CVD of NAFLD versus non-NAFLD was 1.81 (95% CI: 1.23-2.66) ( Fig. 2A). There was potential heterogeneity among the individual studies (I 2 = 79.8%, P < 0.001). Subgroup analyses indicated that the significantly higher risk was not seen participants with mean age over 50 years, in studies including exclusively Asian participants, with relatively low study quality, and not adjusting for age or BMI/obesity or smoking (Table S5). Sensitivity analysis indicated that the significantly higher risk was not materially changed in the leave-one-out analyses by omitting one study in turn, with pooled ORs range from 1.48 (95% CI: 1.09-1.99) to 2.15 (95% CI: 1.34-3.46) ( Figure S2C).
Cohort studies. In cohort studies, pooled HR for incident CVD of NAFLD versus non-NAFLD was 1.37 (95% CI: 1.10-1.72) (Fig. 2B). Potential heterogeneity among the studies was observed (I 2 = 55.1%, P = 0.038). The pooled HRs did not differ significantly in the subgroup analyses that had more than one comparison (Table S5). The leave-one-out analyses indicated that the significantly higher risk was not materially changed by omitting one study in turn, with pooled HRs range from 1.26 (95% CI: 1.07-1.49) to 1.62 (95% CI: 1.11-2.35) ( Figure S2D).

NAFLD and prevalent/incident CAD. Cross-sectional studies.
In cross-sectional studies, pooled OR for prevalent CAD of NAFLD versus non-NAFLD was 1.87 (95% CI: 1.47-2.37) (Fig. 3A). There was potential heterogeneity among the individual studies (I 2 = 80.2%, P < 0.001). In the subgroup analyses, the higher risk was not significant in studies using population-based design and including exclusively diabetic participants (Table S5). In the univariate meta-regression analysis, the regression coefficients of age (P = 0.035), male percent (P = 0.061), study design (P = 0.092), ethnicity (P = 0.078), and study quality (P = 0.013) were significant at the level of 0.1, and these five covariates were entered into the multivariate meta-regression analysis. After including these five covariates in the model, the τ 2 changed from 0.06891 to 0.01922, which means that 72.10% of heterogeneity between the studies can be explained by these covariates. Sensitivity analysis indicated that the significantly higher risk was not materially changed in the leave-one-out analyses by omitting one study in turn, with pooled ORs range from 1.49 (95% CI: 1.28-1.75) to 2.02 (95% CI: 1.52-2.69) ( Figure S2E).
Cohort studies. Only one cohort study was available for assessment of association between NAFLD and incident CAD. The HR reported by the study was 2.31 (95% CI: 1.46-3.65) (Fig. 3B).
There was no potential heterogeneity among the individual studies (I 2 = 0.0%, P = 0.525). The magnitude and direction of the associations were unaltered across studies in the subgroup analyses (Table S5). The leave-one-out analyses indicated that the significantly higher risk was not materially changed by omitting one study in turn, with pooled ORs range from 1.23 (95% CI: 1.12-1.35) to 1.26 (95% CI: 1.10-1.45) ( Figure S2F).
In the subgroup analyses, the higher risk was not significant in studies not adjusting for age or BMI/obesity or smoking (Table S5). Sensitivity analysis indicated that the significantly higher risk was not materially changed in the leave-one-out analyses by omitting one study in turn, with pooled HRs range from 1.11 (95% CI: 1.06-1.17) to 1.20 (95% CI: 1.09-1.33) ( Figure S2G).
NAFLD and prevalent atherosclerosis. Data for association between NAFLD and atherosclerosis were only available in cross-sectional studies. Pooled OR for prevalent atherosclerosis of NAFLD versus non-NAFLD was 1.32 (95% CI: 1.07-1.62) (Fig. 5). There was no potential heterogeneity among the individual studies (I 2 = 34.0%, P = 0.218). In the subgroup analyses, the higher risk was not significant in studies including exclusively non-Asian participants and not adjusting for age or BMI/obesity or smoking (Table S5). Sensitivity analysis indicated that the higher risk was not materially changed in the leave-one-out analyses by omitting one study in turn except for one comparison in the study of Huang and colleagues 18 , with pooled ORs range from 1.27 (95% CI: 1.14-1.43) to 1.40 (95% CI: 1.13-1.75) ( Figure S2H).

NASH, mortality and CVD incidence. Data for association between NASH and adverse cardiovascular
events were only available for assessment of overall mortality, CVD mortality, and CVD incidence in cohort studies. Meta-analyses indicated that NASH was not associated with overall mortality (HR = 1.37, 95% CI: 0.86-2.19, I 2 = 86.4%, P < 0.001) and CVD mortality (HR = 1.18, 95% CI: 0.57-2.48, I 2 = 83.3%, P < 0.001) but significantly increased the incident CVD risk (HR = 2.97, 95% CI: 1.03-8.52, I 2 = 92.0%, P < 0.001) (Fig. 6). Subgroup analyses and sensitivity analyses were only conducted for overall mortality and CVD mortality, because there were only two comparisons for CVD incidence. In the subgroup analyses, the pooled HRs did not differ significantly (Table S5). Sensitivity analysis indicated that the non-significant risks were not materially changed in the leave-one-out analyses by omitting one study in turn ( Figure S2I,J).

Publication bias.
There was no potential publication bias in most of our analyses as assessed by funnel plots, Egger's regression test and Begg-Mazumdar test ( Figure S3). Egger's regression test (P = 0.020) indicated there was potential publication bias when assessing NAFLD and CVD incidence in cohort studies ( Figure S3D). After using the trim and fill approach, one study was filled and the pooled result did not reverse (HR = 1.36, 95% CI: 1.06-1.74) ( Figure S3E). Both Egger's regression test (P = 0.013) and Begg-Mazumdar test (P = 0.012) indicated there was potential publication bias when assessing NAFLD and CAD prevalence in cross-sectional studies ( Figure S3F). After using the trim and fill approach, six studies were filled and the pooled result did not reverse (OR = 1.36, 95% CI: 1.04-1.77) ( Figure S3G).

Discussion
The main results of our meta-analysis are the following: (1) NAFLD was not associated with overall mortality and CVD mortality; (2) NAFLD was associated with an increased prevalence and/or incidence of other adverse cardiovascular events, including CVD, CAD, hypertension, and atherosclerosis; (3) NASH was not associated with overall mortality and CVD mortality but was associated with an increased incidence of CVD. These results are disease; CT, computed tomography; CVD, cardiovascular disease; DBP, diastolic blood pressure; DM, diabetes mellitus; eGFR, estimated glomerular filtration rate; FG, fasting glucose; FRS, Framingham risk score; GGT, γ -glutamyltranspeptidase; HDL-C, high-density lipoprotein cholesterol; HOMA-IR, Homeostatic model assessment of insulin resistance; hsCRP, high sensitivity C-reactive protein; HTN, hypertension; LDL-C, lowdensity lipoprotein cholesterol; Met Sy, metabolic syndrome; NA, not available; NAFLD, non-alcoholic fatty liver disease; NASH, nonalcoholic steatohepatitis; SBP, systolic blood pressure; SUA, serum uric acid; TC, total cholesterol; TG, triglycerides; WC, waist circumference.
important given the high prevalence of NAFLD in the general population and the concerns raised by the adverse metabolic profile associated with this disease and NASH. Based on the current evidence from the literature, the association between NAFLD and adverse cardiovascular events is mixed. Several researchers reported that NAFLD is associated with CVD in diabetic patients 24,33,42,43 , but others did not confirm this finding 8,44 . We believe the explanation for this contrasting finding lies in important differences in the study populations, sample size, study design, study duration and disease ascertainment methods. The value of the current meta-analysis compensates for the individual lack of precision in most of the studies, a problem that was alleviated by pooling the data of all the studies.
In our study, we found no association between NAFLD or NASH and deaths from all causes or CVD. This result was similar with a cohort study from Sweden. This study followed up a cohort of 144 patients with NAFLD for over 13.7 years, and found patients with NAFLD had similar survival to the general Swedish population (matched for age and gender). However, their study found the risk of death was increased in patients with NASH 45 . Similarly, two other studies also found no evidence of an increased risk of death among patients with NAFLD compared with the general population of either the United Kingdom or Denmark 46,47 . However, there was no adjustment for potential confounders in these studies. In contrast, one study reported an increased risk of death from all causes among 1804 patients with fatty liver, with standardized mortality ratio compared with general Danish population of 2.6; but this study was not able to adjust for confounders either 48 . The potential confounders in these studies make it difficult to draw firm conclusions about the impact of NAFLD and NASH on mortality.
In addition, our study indicated that NAFLD was an independent risk factor in determining cardiovascular events (including CVD, CAD, hypertension, and atherosclerosis) by pooling the multiple-adjusted data together, and NASH was an independent risk factor in determining CVD. Hamaguchi and colleagues 49 showed NAFLD is strongly related to metabolic syndrome and therefore shares many risk factors with cardiovascular disease, suggesting a close relationship between NAFLD and adverse cardiovascular events. Some other mechanisms may explain the higher risks of cardiovascular events in patients with NAFLD. The disease is associated with a proatherogenic lipid profile 50 and increased production of pro-inflammatory cytokines including IL-6 51 . Experimental researches also showed that the degree of liver injury and chronic inflammation play an important role in the pathogenesis of atherosclerosis 52 . In addition, it has been reported that patients with NAFLD had larger carotid intima-media thickness 53,54 , increased prevalence of endothelial dysfunction 55 and calcified and noncalcified Our study showed that NASH, a more severe stage of NAFLD, was associated with higher risks in overall mortality, CVD mortality, and incident CVD with larger HRs, compared with NAFLD, although the risks were not significant in overall mortality and CVD mortality. A previous study showed that NASH is associated with a more severe inflammatory and insulin-resistant state that promotes atherosclerosis 58 . Singh et al. 59 demonstrated that the annual fibrosis progression rate in patients with NASH (0.14) was higher than that in patients with non-alcoholic fatty liver (0.07). On the other hand, long-standing NASH can result advanced fibrosis and may indirectly reflect exposure to risk factors for cardiovascular disease. Histology-based studies have also suggested that cardiovascular disease is mainly associated with more severe forms of NAFLD 45,60 . Subgroup analysis by mean age of the participants showed that in the analyses which had more than one comparison, elderly participants (mean age ≥ 50 years) had higher cardiovascular risks with larger point estimates, suggesting further increased risk involving an age-related mechanisms. Frith and colleagues 61 indicated that elderly patients with NAFLD had greater fibrosis in biopsy and significantly more cardiovascular risk factors, including obesity, diabetes, hypertension, and hyperlipidemia. Another study also reported that fat may become dysfunctional and redistribute from subcutaneous to intra-abdominal visceral depots in old-age patients 62 , and the fat redistribution in the elderly has been shown to be related to increased traditional cardiovascular risk factors such as hypertension, diabetes, central obesity, atherosclerosis and dyslipidemia. Recently, Targher et al. 63 also conducted a meta-analysis on NAFLD and risk of incident CVD. The differences between their study and ours are that they pooled prospective and retrospective studies together to calculate the risk of incident CVD, while our study included cross-sectional and prospective cohort study and pooled them separately to calculate the risk of prevalent and incident CVD. In addition, their study only reported the risk of overall CVD, while our study not only reported the risk of overall CVD but also reported the risks of some specific CVDs.
This meta-analysis provides the most definitive and convincing evidence so far of NAFLD-related risk of cardiovascular events. The findings were robust and applicable across a broad range of populations. However, the study has several limitations. First, the most pooled analyses revealed heterogeneity among studies. Although subgroup analysis and meta-regression analysis gave some clues to explain the heterogeneity, these are unlikely to have fully accounted for heterogeneity. Therefore, the results of the meta-analysis must be interpreted with caution. Because of potential additional heterogeneity in the populations, designs, and analyses of the various studies, we assumed that the true effect being estimated would vary between studies, in addition to the usual sampling variation in the estimates. To account for the heterogeneity, we used the random-effect model to combine the results of the primary studies. The random-effect approach provides some allowance for heterogeneity in studies beyond sampling error. This does not necessarily rule out the effect of heterogeneity between the studies, but one can expect a very limited influence because of it. Second, the meta-analysis is based on observational studies, which leaves the possibility that residual confounding factors, including measurement errors, affect the relation between NAFLD and adverse cardiovascular events. Third, data were only available for analysis for risks of overall mortality, CVD mortality, prevalent/incident CVD, prevalent/incident CAD, prevalent/incident hypertension, and prevalent atherosclerosis, and other adverse cardiovascular events are not analyzed because studies on other adverse cardiovascular events are quite few for solid meta-analysis. More studies are needed to confirm the relationship between NAFLD and risk of other specific cardiovascular diseases. Fourth, there were limited studies in several subgroup analyses, which may lead to low statistical power in these analyses. Another important limitation of these data, which warrants further investigation by future studies, was that a standard definition of diabetes was not used across studies and ascertainment of diabetes probably varied between studies. Additionally, although we did subgroup analysis by the presence of diabetes, data on exclusively diabetic participants and non-diabetic participants were quite few. Further studies are needed to examine the potential contribution that the presence of diabetes might have on the excess risk of adverse cardiovascular events.
In conclusion, our analysis shows that the presence of NAFLD is not associated with overall mortality and CVD mortality but is associated with an increased risk of other adverse cardiovascular events, including CVD, CAD, hypertension, and atherosclerosis. Future studies should evaluate strategies and interventions to prevent cardiovascular disease progression in individuals with NAFLD.

Methods
Search strategy and eligibility criteria. We followed the PRISMA guidelines 64 to complete the meta-analysis. Two investigators (SW and YD) conducted a systematic literature search of PubMed, EmBase, Web of Science, and Cochrane Library for identification of articles published between 1965 and July 3, 2015, using a combined text and MeSH heading search strategy with the terms: "non-alcoholic fatty liver disease", "nonalcoholic fatty liver disease", "NAFLD", "non-alcoholic steatohepatitis", "NASH", "fatty liver", "liver fat", "steatosis", "cardiovascular diseases", "atherosclerosis", "stroke", "atrial fibrillation", "overall mortality", "coronary artery disease", "coronary heart disease", "hypertension", and "mortality". The search was restricted to studies in human beings and no language restriction was imposed. We also checked the reference lists of identified reports for other potentially relevant studies. We contacted the authors of the included studies to ask them for additional information and unpublished data.
We included studies that met the following criteria: participants aged 18 years or older; cross-sectional design, prospective design, or retrospective design; the association between NAFLD and adverse cardiovascular events or mortality was assessed; and reported data on odds ratios (ORs) or hazard ratios (HRs) with confidence intervals (CIs) or sufficient information to calculate these, for the association between NAFLD and adverse cardiovascular events or mortality. Studies were excluded if they did not provide information to calculate the point estimate, did not make comparison between NAFLD and adverse cardiovascular events or mortality, or were review studies. Articles that clearly did not meet inclusion criteria were rejected on initial review. If uncertainty existed, the full text of the article was reviewed. Two reviewers (FW and JH) independently assessed all potentially relevant studies for inclusion. Disagreements were resolved by consensus. Data extraction and study quality evaluation. Study characteristics were extracted independently by two researchers (PM and YH). We extracted risk estimates (95% CI) for different genders and disease severities separately when possible. If a study reported more than one measure of adverse cardiovascular events, each adverse cardiovascular event was extracted separately. The most adjusted estimate was included when a study Figure 6. Forest plot of comparison. NASH versus non-NASH, outcome: overall mortality (A), cardiovascular disease mortality (B) and incident cardiovascular disease (C) based on cohort studies. Studies assessing NASH by ultrasound or liver biopsy were considered separately. reported more than one risk estimate. The quality of each study was assessed by two researchers (FW and PM), using the Newcastle-Ottawa Scale (NOS) recommended by Wells and colleagues 65 . The quality of each included studies ranges from 1 to 9 stars for cohort and case-control studies and 1 to 5 stars for cross-sectional studies.
Statistical analysis. Associations with continuous outcome variables were expressed as weighted mean differences (WMDs) with 95% CIs. For evaluation of the relative risk of adverse cardiovascular events, the effect size was estimated as ORs or HRs with 95% CIs, according to NAFLD patients vs non-NAFLD patients reported in each study, using non-NAFLD patients as the reference group. The impact of NAFLD histological subtypes (NASH) on adverse cardiovascular events was also examined.
We used the random-effect model in this meta-analysis to take into account heterogeneity among studies, because the study design and measuring time were different across studies 66 . The I-squared (I 2 ) statistic and Q-statistic were used to explore the heterogeneity among studies. Large I 2 (> 50%) or P < 0.1 for Q-statistic suggests substantial heterogeneity among studies. We separately analyzed cross-sectional and longitudinal studies. Moreover, the results of studies defining NAFLD by ultrasound, liver biopsy, CT images, or liver enzyme elevation are presented separately. We did several subgroup analyses: study design (population-based or hospital-based), mean age of the participants (≥50 years or <50 years), ethnicity (non-Asian or Asian), presence of diabetes (diabetic participants or non-diabetic participants or combined), study quality (high vs relatively low), and adjustment of some major risk factors (yes vs no). When eight or more comparisons were available, the effect of several variables including age, publication year, male percent, study design, follow-up duration (for cohort studies), body mass index (BMI), ethnicity, smoke, presence of diabetes, and study quality was assessed by meta-regression analysis to further evaluate the source and strength of heterogeneity. We also performed sensitivity analyses by removing each individual study from the meta-analysis 67 . Funnel plots were used to examine the presence of publication bias (ie, by plotting the natural log of the odds ratio against its standard error). We used Egger's regression test and Begg-Mazumdar test to further assess publication bias. All statistical analyses were done with Stata Version 12.0 software (Stata Corp, College Station, TX).