Development of colorectal cancer predicts increased risk of subsequent hepatocellular carcinoma in patients with alcoholic liver disease: case-control and cohort study

Alcohol increases the risk of both hepatocellular carcinoma (HCC) and colorectal neoplasia. In this hospital-based case-control and retrospective cohort study, we sought to determine whether development of colorectal neoplasia increases the risk of HCC in patients with alcoholic liver disease (ALD). In the phase I case-control analysis, the association between history of colorectal cancer (CRC) and HCC development was assessed in patients with ALD by logistic regression modeling (n = 1,659). In the phase II retrospective cohort analysis, the relative risk of HCC development was compared in ALD patients with respect to the history of CRC by a Cox model (n = 1,184). The history of CRC was significantly associated with HCC in the case-control analysis (adjusted odds ratio, 1.82; 95% CI, 1.06–3.15; P < 0.05). ALD patients with CRC had higher risk of developing HCC compared to those without CRC (adjusted hazards ratio [HR], 5.48; 95% CI, 1.63–18.36; P = 0.006) in the cohort analysis. Presence of CRC, liver cirrhosis, elevated baseline alpha-fetoprotein level, and low platelet counts were independent predictors of HCC development in ALD patients. Patients with history of CRC had an increased risk of HCC in both cirrhotic (HR, 3.76; 95% CI, 1.05–13.34, P = 0.041) and non-cirrhotic (HR, 23.46; 95% CI, 2.81–195.83, P = 0.004) ALD patients. In conclusion, ALD patients with CRC are at increased risk of developing HCC.

www.nature.com/scientificreports www.nature.com/scientificreports/ Alcohol is a risk factor not only for HCC but also for colorectal neoplasm: excessive alcohol consumption increases the likelihood of developing adenomas and colorectal cancer (CRC) [14][15][16][17] . Chronic ethanol consumption usually promotes carcinogenesis by production of toxic acetaldehyde, induction of cytochrome P450 2E1 that causes oxidative stress and procarcinogen activation, and provocation of global DNA hypomethylation 18 . Since these carcinogenic pathways are known to contribute to the development of both colorectal neoplasm and HCC 10,19,20 , we hypothesized that ALD patients who develop colorectal neoplasm are at risk of further developing HCC. The aim of this study was to elucidate the relationship between the presence of colorectal neoplasm and the subsequent risk of HCC in patients with ALD.

Methods
Data sources. We recruited ALD patients from two university-affiliated, tertiary referral centers in South Korea. Patients with ALD were identified from the electronic health record of the centers 21,22 by the World Health Organization International Classification of Diseases, tenth revision (ICD-10) code of K70. The diagnosis of alcoholic liver disease was made by four hepatologists (WK, ESJ, SHJ and JWK) by documentation of alcoholic excess and evidence of liver disease 23 . Alcohol consumption amount was assessed via the hepatologists' interview with patients and their family members. Diagnosis of fatty liver was made radiologically 24 , and liver cirrhosis was diagnosed by liver biopsy, radiological findings or presence of esophageal varix 25 . Clinical investigations were conducted according to the principles expressed in the Declaration of Helsinki 26 , and the requirement for informed consent was waived due to the retrospective nature of the current study and anonymous analysis of the data. The institutional review board approved this study (Seoul National University Bundang Hospital IRB; IRB no. B-1703/388-108).

Study design.
This study consists of two-phase investigations: phase I case-control analysis and phase II retrospective cohort analysis (Fig. 1). The case-control analysis aimed to investigate the possible association between history of CRC and HCC risk in ALD patients, whereas the cohort analysis sought to determine the relative risk of HCC in ALD patients with the history of colorectal adenoma (CRA) or CRC.
Phase I case-control analysis. In the phase I single-center case-control analysis, we selected 1,659 ALD patients who received imaging work-up for HCC, i.e., liver ultrasound (US), computed tomography (CT) or magnetic resonance imaging (MRI) at Seoul National University Bundang Hospital between April, 2003 and April, 2018. Patients with evidence of other chronic liver disease such as hepatitis B or C virus infection, hemochromatosis, Wilson disease, primary biliary cholangitis, autoimmune hepatitis, and primary sclerosing cholangitis were excluded. Decompensated patients with Child-Pugh class C were also excluded from analysis.
ALD patients diagnosed with HCC were selected as cases, and those without any evidence of HCC were selected as controls. The index date was defined as the date of HCC diagnosis for cases, and the date of initial imaging work-up for controls. The history of CRC was coded positive if a patient had received surgical treatment for CRC or if colonoscopic examination(s), performed as described below, revealed colorectal adenocarcinoma before the index date. www.nature.com/scientificreports www.nature.com/scientificreports/ Phase II cohort analysis. In the phase II retrospective cohort analysis, an ALD cohort was electronically formed by assembling ALD patients who received imaging work-up for HCC and colonoscopic examination(s) from the two centers. The cohort was composed of a subgroup of the phase I controls and a separate cohort formed at Boramae Medical Center. Patients with evidence of other chronic liver disease were excluded in the same way as the phase I cohort, and the following additional exclusion criteria were applied: patients were excluded if HCC was detected at initial screening or within 6 months after initial screening or if follow-up data was incomplete. The observation starting point for the longitudinal cohort analysis was the time of colonoscopic examination(s). During follow-up, patients underwent additional liver US with or without alpha-fetoprotein (AFP) levels. Follow-up US at 6 months of interval was recommended for patients with alcoholic liver cirrhosis, and follow-up US was performed at the discretion of attending hepatologists for patients without evidence of liver cirrhosis. Dynamic CT or MRI was performed if serum AFP levels or surveillance US indicated a need for further evaluation according to the recall policies 27,28 . Liver biopsy was performed if CT or MRI showed new hepatic nodule(s) which did not show typical enhancement pattern diagnostic of HCC. The primary outcome of this study was the incidence of HCC, which was diagnosed based on histological or radiological findings 29 .
CRC screening in the study population. CRC is the third most common cancer in Korea 30 , and Korean National Cancer Screening Programme has implemented programmatic CRC screening by means of fecal occult blood tests for all Koreans aged ≥ 50 years since 2004 31 . If the fecal occult blood test is positive, colonoscopy or double-contrast barium enema has been provided. Colonoscopy has also been recommended for CRC screening regardless of the occult blood test result in average-risk asymptomatic adults according to clinicians' discretion and recipients' preference 32 . In our cohorts, colonoscopic CRC screening was performed either on the programmatic basis following the Korean guideline or on the opportunistic basis. If colonoscopic examinations were performed more than once, the highest grade of colonic pathology was determined and the earliest date of examination was chosen as the ultimate grade of pathology.

Statistical analysis. Statistical analysis was performed using STATA version 14 (Stata Corp, College
Station, TX) and R version 3.4.2 (http://www.cran.r-project.org/). Wilcoxon rank sum test or one-way ANOVA test were used for continuous variables, and chi-square test was used to analyze categorical variables. In phase I case-control analysis, the odds ratios (ORs) for HCC were calculated by logistic regression analysis. In phase II cohort analysis, the cumulative incidence of HCC was calculated using the Kaplan-Meier method and the hazard ratios (HRs) for HCC were derived by a Cox proportional hazards model.

Results
Phase I case-control analysis: association between the history of CRC and the risk of HCC in patients with ALD. The case-control analysis compared 445 ALD-related HCC cases and 1,214 ALD controls without HCC with respect to the history of CRC. Of the 445 HCC patients, 24 had history of CRC, among whom 7 HCC was confirmed pathologically. HCC patients showed older age, higher proportion of female sex, higher smoking rate and drinking amount, higher probability of the history of CRC (5.4% vs. 2.7%; p = 0.008) ( Table 1). Logistic regression analysis showed that past history of CRC was significantly associated with future development of HCC (unadjusted OR, 2.04; 95% CI, 1.19-3.49; P < 0.001), along with sex, drinking amount and smoking as additional significant variables associated with HCC development ( Table 2). The association of CRC history with HCC remained significant after adjustment for confounders (adjusted OR, 1.82; 95% CI, 1.06-3.15; P < 0.05), indicating possible association between development of CRC and HCC in ALD.
Phase II cohort analysis: presence of CRC as an independent predictor of future HCC development in ALD. A total of 1,263 patients were assembled for the longitudinal cohort. After excluding 79 patients with incomplete follow-up information, 1,184 patients were finally selected for the cohort analysis ( Fig. 1). CRA and CRC were present in 46% and 2% of ALD cohort, respectively, at the baseline screening colonoscopic examination (Table 3). One-third of CRC patients were diagnosed at the stage of carcinoma in situ (Tis, n = 8/24). Patients with CRC showed higher age, higher proportion of cirrhosis with worse hepatic reserve, higher frequency of smoking and greater amount of alcohol consumption.
During the median follow-up of 55 months (interquartile range [IQR] = 60, 72,117 person-years), 24 patients developed HCC with the incidence of 33.3/100,000 person-years (95% CI, 22.3-49.7). Among the 24 patients, 4 had history of CRC and HCC was pathologically confirmed in all 4 patients. When the cumulative HCC incidences were compared according to colonic pathologic findings, patients with CRC showed significantly higher probability of further developing HCC compared to patients without colonic neoplasia or patients with CRA (Fig. 2). The median (IQR) size of HCC was not significantly different between patients with and without preceding CRC: 2.7 (5.6) cm vs. 1.9 (3.8) cm for patients with vs. without preceding CRC, respectively (P = 0.312 by Wilcox rank sum test).

Discussion
Our study found that the risk of subsequent HCC development increased when CRCs were detected in patients with ALD. The initial case-control analysis revealed higher proportion of the prior history of CRC in HCC, suggesting association between the development of CRC and HCC in ALD. The Phase I case-control design cannot confirm the causal relationship, but Phase II longitudinal cohort analysis also demonstrated increased OR of HCC by the history of CRC. Taken together, we believe that the Phase I and II analyses indicate a significant association between the history of CRC and the risk of HCC. Several risk factors for HCC have been established in ALD 8,10 , but CRC has not been recognized as one yet. The association between CRC and subsequent HCC remained significant after adjusting for known covariates for HCC, indicating that presence of CRC has additional predictive power.
We did not include presence/absence of cirrhosis in Phase I logistic model because not all patients underwent histologic examinations or endoscopic varix study. Since thrombocytopenia is one of the clinical diagnostic criteria of cirrhosis, we believe that the significant difference in platelet counts in Phase I analysis indicate significant association of liver cirrhosis with development of HCC in ALD. Similarly, low albumin levels and platelet counts in the CRC group may be explained by the higher prevalence of liver cirrhosis in Phase II analysis (Table 3). However, detection of CRC remained significant in the multivariate Cox model ( Table 4), indicating that CRC history may be an independent risk factor for further HCC irrespective of liver cirrhosis.
Our study did not provide mechanistic explanation of the association between HCC and preceding CRC. We speculate that presence of CRC may indicate either exposure to alcohol exceeding a certain threshold level, or increased host susceptibility to alcohol for malignancies for the following findings. First, the carcinogenic effect of alcohol and/or aldehyde metabolites on colonic mucosa might also contribute to the carcinogenesis of HCC 10,19,20 . Second, alcohol-induced changes in gut microbiota increase acetaldehyde production 33,34 , which is a common inducer of both cancers. Moreover, alcohol-induced dysbiosis increases the level of deoxycholic acid (DCA), a gut bacterial metabolite known as a hepatocarcinogen by inducing a senescence-associated secretory phenotype in hepatic stellate cells 35 . Colonic neoplasia as well as alcohol intake may enhance intestinal permeability and enterohepatic circulation of DCA, which in turn facilitate the development of HCC. Epigenetic 36 or immunologic 37 alterations by alcohol may also contribute to the pathogenesis of both CRC and HCC. www.nature.com/scientificreports www.nature.com/scientificreports/ If alcohol have already caused significant genetic and epigenetic alterations in the liver of ALD patients with CRC, then the diagnosis of CRC may be regarded as a surrogate marker for the risk of developing subsequent HCC in ALD. It can be argued that ALD patients who developed CRC might receive more frequent imaging studies, resulting in an increased chance of HCC detection (lead-time bias). If this is the case, then HCC stages may be less advanced in ALD patients with CRC compared to patients who developed HCC without preceding CRC, which is not the case in our data: the size of HCC was not significantly different between patients with and without CRC. Further studies will be needed, however, to elucidate the mechanisms underlying the association between CRC and HCC in ALD.
Liver cirrhosis 38 , elevated baseline AFP level 39 and low platelet counts 25,40 have been reported as significant predictors of HCC in ALD, and our data are line with previous reports. Alcohol intake amount was associated with the risk of HCC in phase I case-control analysis and univariate cohort study, but it was not an independent predictor of HCC in multivariate cohort analysis, probability due to multicollinearity with CRC (Table 3).
HCC surveillance has not been routinely recommended for patients with ALD, due to the widely variable range of its risk and relatively lower contribution of HCC to the mortality of alcoholic hepatitis and/or cirrhosis 13,41 . Our cohorts also showed relatively low incidence of HCC (33/100,000 person-years), which is consistent with the previous reports 38 . In contrast, diagnosis of CRC increased the risk of developing HCC by more than 12 folds in univariate analysis. Although CRC develops infrequently in patients with ALD, we suppose that HCC surveillance should be considered when CRC, even with early-stage, is detected in those with ALD. Since patients with no risk factor may have very low risk of developing HCC (Fig. 3), regular HCC surveillance may not be warranted 6 , whereas those with CRC, especially with additional risk factor(s), may have significantly higher risk of HCC and regular HCC surveillance may be justified. However, further external validation needs to be conducted  www.nature.com/scientificreports www.nature.com/scientificreports/ to evaluate the clinical utility of our risk stratification and the pragmatic strategy of individualized HCC surveillance in ALD.
We excluded patients with Child-Pugh class C (n = 6; 1 with HCC and 5 without HCC) in Phase I analysis because the history of CRC might be less accurate in this group (limited colonoscopic examinations due to grave hepatic prognosis and/or presence of ascites). However, inclusion of these 6 patients for logistic regression analysis has not changed the results of Table 2 (shown in Supplementary Table 1 and Table 2).
Smoking history had the greatest odds ratio among the predictors in Phase I analysis, but showed insignificant hazard ratio in Phase II Cox analysis. We believe that the number of HCC was larger in Phase I analysis (24 vs. 445), so that type 2 error (beta error) occurred for smoking in Phase II analysis, considering the recent reports supporting the carcinogenic potential of smoking in HCC 42,43 . Since most CRC patients were smokers (23/24, Table 3), further data would be needed to determine whether non-smokers with CRC pose higher risk for HCC.
Our study has several limitations. First, due to the case-control and retrospective cohort design, biases cannot be completely avoided. We do not suppose that the probability of recall bias may be high, because CRC is not a casual history. We also employed an electronic health record system in order to include every eligible patient without omission 21,22,44 . Despite these efforts, our findings still need to be externally validated by prospective cohort studies. Second, relatively small number of patients had CRC compared to CRA in our cohort. However, since the CRC history is a significant predictor of HCC regardless of cirrhosis, these patients may need to receive enhanced surveillance for HCC. Again, validation in larger-scale studies is still needed with more CRC cases. Third, our multivariate analyses did not include the stage of fibrosis, which may be a major determinant of HCC