Non-alcoholic fatty liver disease as a risk factor for cholangiocarcinoma: a systematic review and meta-analysis

Background Non-alcoholic fatty liver disease (NAFLD) has been recently identified as a risk factor of gastrointestinal tract cancers, especially hepatocellular carcinoma, and colorectal cancer. Whether NAFLD is a risk factor for cholangiocarcinoma (CCA) remains inconclusive. The aim of this study is to determine a potential association between NAFLD and CCA, stratifying by its subtypes; intrahepatic CCA (iCCA), and extrahepatic CCA (eCCA). Methods A search was conducted for relevant studies published up to April 2017 using MEDLINE, EMBASE, Scopus and Cochrane databases. Odds ratio (OR) and adjusted OR with 95% confidence interval (CI) were estimated using a random-effects model. Subgroup analyses were conducted with study characteristics. Results Seven case-control studies were included in the analysis, with a total of 9,102 CCA patients (5,067 iCCA and 4,035 eCCA) and 129,111 controls. Overall, NAFLD was associated with an increased risk for CCA, with pooled OR of 1.95 (95%CI: 1.36–2.79, I 2=76%). When classified by subtypes, NAFLD was associated with both iCCA and eCCA, with ORs of 2.22 (95%CI: 1.52–3.24, I 2=67%) and 1.55 (95%CI: 1.03–2.33, I 2=69%), respectively. The overall pooled adjusted ORs were 1.97 (95%CI: 1.41–2.75, I 2=71%), 2.09 (95%CI, 1.49–2.91, I 2=42%) and 2.05 (95%CI, 1.59–2.64, I 2=0%) for all CCAs, iCCA, and eCCA, respectively. Conclusions This meta-analysis suggests that NAFLD may potentially increase the risk of CCA development. The magnitude of NAFLD on CCA risk is greater for iCCA than eCCA subtype, suggestive of a common pathogenesis of iCCA and hepatocellular carcinoma. Further studies to confirm this association are warranted. Trial registration The protocol for this study was registered with PROSPERO (International Prospective Register of Systematic Reviews; no. CRD42016046573). Electronic supplementary material The online version of this article (10.1186/s12876-017-0696-4) contains supplementary material, which is available to authorized users.


Background
Cholangiocarcinoma (CCA) is a cancer arising from bile duct epithelium. The incidence of CCA varies by geographic regions, ranging from 0.4-3.4 per 100,000 persons-year in North America and Europe to 1-85 per 100,000 persons-year in East Asia [1]. Although CCA is a relatively uncommon cancer, its incidence has been rising worldwide over the past decade [1]. It remains unclear why the incidence of CCA has been on the rise. This might be due to an increase in prevalence of some host or environmental factors potentially related to CCA development.
CCA is broadly categorized into 2 subtypes by anatomic locations as intrahepatic (iCCA) and extrahepatic (eCCA) subtypes. The two subtypes hold different genetics, presentations, management, and outcomes [2]. Since diagnosis in early stages of disease is difficult, most patients have poor prognosis. Thus, recognition of CCA risk factors would potentially identify individuals at risk and may consequently improve patient outcomes.
Non-alcoholic fatty liver disease (NAFLD) is a spectrum of liver disease, ranging from fatty liver to non-alcoholic steatohepatitis (NASH) and cirrhosis. It is estimated that one-third of the general population has NAFLD [3][4][5]. Accumulating evidence suggested that NAFLD was associated with an increased risk of various cancers, including hepatocellular carcinoma (HCC), esophageal, gastric, colorectal, breast, and prostate cancer [6,7]. Recent data suggested that NAFLD might increase the risk of CCA, particularly the iCCA subtype [8][9][10]. Whether NAFLD is associated with the other subtype of CCA remains uncertain.
To determine an association between NAFLD and CCA including its subtypes, we conducted a meta-analysis of case-control and cohort studies.

Data sources and searches
The protocol for this study was registered with PROS-PERO (International Prospective Register of Systematic Reviews; no. CRD42016046573). We performed a systematic review in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-analyses) guidelines (Additional file 1: Data S1) [11]. We searched several databases including Ovid MEDLINE, Epub Ahead of Print, Ovid Medline In-Process & Other Non-Indexed Citations, Ovid MED-LINE, Ovid Cochrane Central Register of Controlled Trials, Ovid EMBASE and Scopus from the inception of the databases through April 5, 2017. The search strategy was designed and conducted by an experienced librarian with input from the study's principal investigator. Searching terms included cholangiocarcinoma, NAFLD, NASH, risk factors, and their related terms (detail of the full search strategy is provided in Additional file 2: Data S2A, S2B, and S2C). Search strategies were confined to human studies and case-control, cohort, or trial studies. Title and abstract of included studies were screened. References of the included studies and all articles that cited the included studies were reviewed for other potential relevant studies.

Study selection
Primarily, two authors (NW and BA) independently screened titles and abstracts of all studies that determined any risk factors for CCA and reached for the full text. From full text, studies were included if they that met all of the following criteria: (i) case-control, cohort or trial study, (ii) NAFLD or NASH, defined by either histopathological examination, imaging study or International Classification of Diseases, Ninth Revision (ICD-9) or ICD-10 codes, as one of the exposure of interests, (iii) CCA either iCCA, eCCA, or both as outcome of interest, (iv) study that provided adequate information for calculation of odds ratio (OR) or relative risk for case-control study and cohort study, respectively. Studies of patient cohorts with recurrent CCA or combined hepatocellular-cholangiocarcinoma were excluded. If the same patient cohort was included in more than one study, the study with a larger sample size and a higher quality score assessed by Newcastle-Ottawa scale (NOS) scale was included.

Data extraction and quality assessment
Data were independently extracted from full-text articles by two authors (NW and BA). Disagreements were identified and discussed with the third author (RC). The information extracted included diagnostic criteria of NAFLD/NASH, a crude number of individuals with and without NAFLD/NASH in the case and control groups, the source of controls (population-based or hospitalbased), number of participants, country where a study was conducted, and publication year. Adjusted odds ratio (AOR) and covariates in the adjusted analysis model were also extracted.

Data synthesis and analysis
Pooled OR along with 95% confidence interval (CI) of CCA, iCCA, and eCCA were estimated from the crude number of study patients using a random-effects model. Heterogeneity among studies was assessed using both the I 2 statistics and P value. An I 2 value of >50% indicates substantial heterogeneity. Univariate randomeffects model meta-regressions of the adjusted OR (AOR) from the included studies were performed. Publication bias was assessed by the Egger test. All analyses were performed using Comprehensive Meta-Analysis Software version 2.0, and the Metan and Metareg Commands of Stata 11 (College Station, TX).

Literature search
After excluding duplicates, 1843 articles that met the search criteria were identified. After title and abstract screening, 213 studies that met the selective criteria were further assessed (Fig. 1). Two hundred and three studies were excluded due to the following reasons: NAFLD was not assessed in the study (n = 149); the case group was not diagnosed with CCA (n = 25); no control group in the study (n = 13); and irrelevant study (n = 16). There were ten potential relevant studies, of which studies from Chaiteerakij et al. [14] and Haung et al. [15] were excluded because their patient cohort was a part of the cohort in studies of Choi et al. [16] and Chang et al. [17], respectively (detail in Additional file 2: Data S3). Finally, eight studies met the criteria: seven case-control studies and one cohort study. All studies were published in full-text article.
Given the difference in study design, 7 case-control studies were included in this meta-analysis (Table 1) [9,10,[16][17][18][19][20]. The single cohort study was analyzed separately because this study provided limited information, i.e. CCA was not classified as iCCA or eCCA; and performed restricted analysis, i.e. the analysis was not adjusted for potential confounders (Additional file 2: Data S4) [21].

Publication bias
No publication bias was detected by the Egger's regression asymmetry test, with P = 0.82 and 0.86 for unadjusted and adjusted OR of NAFLD, respectively. However, due to the limited number of the included

Discussion
The present meta-analysis of seven case-control studies found that NAFLD was significantly associated with an increased risk of CCA for both iCCA and eCCA subtypes. This finding might in part explain the recent trend of rising incidence of CCA. Although the mechanism by which NAFLD causes CCA development has not yet been well studied, it is biologically plausible that NAFLD promotes cholangiocarcinogenesis a b Fig. 2 Forest plots of seven studies showing unadjusted odds ratio (a) and adjusted odds ratio (b) of the association between NAFLD and the risk of cholangiocarcinoma directly through an induction of hepatic inflammation, or indirectly via cirrhosis. Cirrhosis, regardless of etiology, has recently been recognized as a risk factor for CCA [10,16,17]. Alteration of the microenvironment in presence of cirrhosis is a hallmark of carcinogenesis [22]. Up to 5% of patients with NAFLD developed cirrhosis during 8 years of follow-up [23]. In vivo and in vitro studies found that a number of signaling molecules crucial for carcinogenesis were aberrantly expressed in NAFLD [24,25]. These include an increased expression of pro-inflammatory cytokines, particularly interleukin-6 (IL-6) [24], which plays a pivotal role in induction of cholangiocyte proliferation [26]. Tumor necrosis factor-alpha (TNF-α), another proinflammatory cytokine, activated inducible nitric oxide synthase (iNOS) leading to nitric oxide production. This change consequently promotes DNA damage and inhibits DNA repair mechanisms [27]. iNOS activation also upregulated COX2 expression, which promotes cholangiocyte growth [28].
It is important to note that the prevalence of NAFLD in this meta-analysis varied from 0.8 to 44.1% and 0.3 to 18.8% in CCA cases and controls, respectively. These numbers were relatively low as compared to previous studies reporting prevalence of NAFLD in the general population of 11-45% and 8-42% in North America and Asia, respectively [3,4]. Substantial variation among studies can be explained by differences in diagnostic methods of NAFLD, with the highest prevalence in studies that NAFLD was diagnosed by histopathological exam (15.6-69.3%), followed by imaging (20.0-30.9%), and elevation of liver enzyme (7.9-14.9%) [5]. The study from Stepien et al. [19] which used hepatic steatosis index to diagnosed NAFLD demonstrated the highest prevalence of NAFLD of 61.8% and 37.3% in the case and control groups, respectively. Moreover, the included studies that used histopathological exam as part of NAFLD diagnosis [9,16,18] had significantly higher prevalence than the studies that used only imaging or ICD coding [10,17,20] (4.7-44.1% vs. 0.8-4.5%). This could reflect an underestimated number of patients with NAFLD in the studies that used only imaging as diagnostic criteria.
Our finding showed the effect of NAFLD was more pronounced for the development of iCCA than eCCA (pooled OR of 2.22 vs. 1.55). This finding is consistent with the previous study showing that iCCA was more associated with chronic liver diseases than eCCA [1]. Moreover, a recent in vivo study showed that iCCA was originated from hepatic cell lineage rather than bile duct cell lineage [29]. On the other hand, eCCA, including perihilar and distal CCA, tended to be associated with biliary tract diseases, as supported by a strong association between primary sclerosing cholangitis and perihilar CCA [1].
As expected, the estimated pooled OR of NAFLD obtained from community-based studies was greater than that obtained from hospital-based studies (2.46 vs. 1.47) (Data not shown). This can be explained by a relatively lesser proportion of controls diagnosed with NAFLD in community-based studies compared to hospital-based studies. Because individuals with NAFLD are mostly asymptomatic, thus controls in community-based studies might have less likelihood to undergo radiologic images, which could potentially increase the possibility of under-detection of NAFLD in controls in community-based studies. This is suggested by the finding that the proportion of NAFLD in controls in community-based studies were lower than the previous reports [3][4][5]. To minimize the impact of underdetection rate of NAFLD in the control groups, we performed subgroup analysis of the studies that diagnostic procedures, i.e. laboratory test, radiologic imaging, or histopathological assessment, were performed in all cases and controls and found that the estimated pooled OR was 2.0. We believe that this pooled OR was the estimate most approximate to the true value.
The present meta-analysis has some limitations. Diagnostic methods for NAFLD were different among studies. This bias was accounted by the aforementioned analysis of subgroup by the same diagnostic methods for both case and control groups. The results of the two analyses were consistent, implying that our finding was not influenced by NAFLD diagnosis method. In addition, subgroup analyses of some underlying diseases predisposing to CCA, e.g. primary sclerosing cholangitis, cholelithiasis, biliary tract infection, and other chronic liver diseases, cannot be performed due to the limited number of participants with these conditions. Because CCA is considered to be a relatively rare cancer, the number of publications in this field is quite small compared to other cancers. Consequently, the number of studies qualified for this meta-analysis was limited. Accordingly, the probability of type I errors in the process of hypothesis testing existed, particularly with multiple subgroup analyses. However, the subgroup analyses are necessary to explore the finding of the main analysis. We limited the subgroup analyses to the three most important subgroups, including CCA subtypes, study region, and study quality. Another limitation was that NAFLD is a spectrum of disease, ranging from simple steatosis to cirrhosis; in this study, we were not able to estimate the magnitude of impact of steatosis, steatohepatitis or NASH-related cirrhosis on CCA risk individually because they used different diagnostic methods and most papers grouped steatosis and steatohepatitis together. Lastly, cirrhosis itself has recently been recognized as a risk factor for CCA, the possibility that the observed association between NAFLD and CCA in this study was confounded by cirrhosis cannot be excluded. A study to determine whether the increased risk of CCA in patients with NAFLD is independent to cirrhosis status is needed.

Conclusions
This meta-analysis suggests that NAFLD may potentially increase the risk of CCA development. The magnitude of NAFLD on CCA risk is greater for iCCA than eCCA subtype, suggestive of a common pathogenesis of iCCA and hepatocellular carcinoma. Further studies to elucidate both the strength of the association between NAFLD and CCA, as well as the mechanisms that underlie this relationship are warranted.

Additional files
Additional file 1: Data S1. PRISMA 2009 checklist for systematic review.