Risk factors for the development of hepatocellular carcinoma (HCC) in chronic hepatitis B virus (HBV) infection: a systematic review and meta‐analysis

Abstract Hepatocellular carcinoma (HCC) is one of the leading contributors to cancer mortality worldwide and is a leading cause of death in individuals with chronic hepatitis B virus (HBV) infection. It is uncertain how the presence of other metabolic factors and comorbidities influences HCC risk in HBV. Therefore, we performed a systematic literature review and meta‐analysis to seek evidence for significant associations. MEDLINE, EMBASE and Web of Science databases were searched from 1 January 2000 to 24 June 2020 for studies investigating associations of metabolic factors and comorbidities with HCC risk in individuals with chronic HBV infection, written in English. We extracted data for meta‐analysis and generated pooled effect estimates from a fixed‐effects model. Pooled estimates from a random‐effects model were also generated if significant heterogeneity was present. We identified 40 observational studies reporting on associations of diabetes mellitus (DM), hypertension, dyslipidaemia and obesity with HCC risk. Only DM had a sufficient number of studies for meta‐analysis. DM was associated with >25% increase in hazards of HCC (fixed‐effects hazards ratio [HR] 1.26, 95% confidence interval (CI) 1.20–1.32, random‐effects HR 1.36, 95% CI 1.23–1.49). This association was attenuated towards the null in a sensitivity analysis restricted to studies adjusted for metformin use. In conclusion, in adults with chronic HBV infection, DM is a significant risk factor for HCC, but further investigation of the influence of antidiabetic drug use and glycaemic control on this association is needed. Enhanced screening of individuals with HBV and diabetes may be warranted.

CHB deathsdue to primary liver cancer and cirrhosis account for the majority of deaths attributable to viral hepatitis. A Global Burden of Disease study on the global hepatocellular carcinoma (HCC) burden reported a 42% increase in incident cases of HCC attributable to CHB infection between 1990 and 2015, 3 among which CHB infection was the largest contributor, responsible for more than 30% of incident cases in 2015. 3 Multiple risk factors for HCC in CHB-infected individuals have been established, including sex, age, cirrhosis and co-infection with human immunodeficiency virus (HIV) or other hepatitis viruses (including hepatitis C and D). Previous studies have investigated associations of comorbidities, such as diabetes mellitus (DM) 4-7 and hypertension, 8,9 with risk of HCC in the general population, and The European Association for the Study of the Liver (EASL) recognizes DM as a risk factor for HCC in CHB. 10 As the global prevalence of comorbidities such as DM, 11 renal disease, 12 hypertension 13 and coronary heart disease (CHD) 14 continues to rise, these conditions are increasingly relevant to the development of HCC.
Various risk scores have been developed to predict HCC risk: the PAGE-B risk score was developed to predict HCC risk in Caucasian patients on antiviral treatment, 15 while the REACH-B, 16,17 GAG-HCC 18,19 and CU-HCC 18,[20][21][22] risk scores apply to untreated Asian populations.
Existing risk scores use age, sex and HBV DNA viral load (VL) to predict risk. CU-HCC and GAG-HCC include parameters for cirrhosis; however, no score accounts for comorbidities such as DM or hypertension. It is possible that risk prediction could be improved by incorporating these comorbid conditions.
Despite rises in the global prevalence of relevant comorbidities, evidence concerning associations of comorbidities with HCC risk in CHB is poor and heterogeneous. Neither EASL, the American Association of the Study of Liver Disease (AASLD) 23 nor the Asian Pacific Association for the Study of the Liver (APASL) 24 guidelines for HBV management include recommendations for enhanced screening or DM management in CHB, despite recent clinical interest in the potential utility of metformin in preventing and treating various cancers. 25,26 Furthermore, there are few studies investigating associations of other potentially relevant comorbidities (such as hypertension, CHD and renal disease) and their metabolic risk factors (such as obesity and dyslipidaemia) with HCC risk. Therefore, we undertook a systematic review, aiming to summarize and critically appraise studies investigating associations of relevant comorbidities and metabolic factors with risk of HCC in CHB-infected individuals.

| Search strategy and selection criteria
In June 2020, we systematically searched three databases (Web of Science, EMBASE and MEDLINE) in accordance with PRISMA guidelines 27 ; search terms are listed in Table S1. We searched all databases from 1 January 2000 until 24 June 2020, without application of any restrictions for study design applied to search terms or results, but including only full-text human studies published in

English.
We combined and deduplicated search results from the three databases, prior to screening for eligibility. We excluded articles not investigating associations of comorbidities with risk of HCC and/ or not restricted to CHB-infected participants. We also searched reference lists of relevant systematic reviews/meta-analyses and studies identified for inclusion to identify additional studies for inclusion. Search terms were constructed and agreed on by three authors (PM, TW and CC) and articles were screened and selected by one author (CC).

| Data extraction and statistical analysis
One author (CC) extracted the following summary characteristics from included studies: country, publication year, study design, follow-up period, comorbidities investigated, number of participants, number of HCC cases, sex, age at baseline, risk ratio and covariates adjusted for.
We carried out meta-analysis in R (version 3.5.1) using the 'meta' package (version 4.12-0), 28 including only hazard ratios (HRs) minimally adjusting for age and sex reported in cohort or nested case-control studies. We calculated pooled summary effect estimates using the inverse-variance weighting of HRs on the natural logarithmic scale, and quantified between-study heterogeneity using the I 2 statistic; significance of heterogeneity was investigated using Cochran's Q test (p threshold = 0.05). Where I 2 was >0 and heterogeneity was significant, we present both fixed-and random-effects summary estimates. We undertook multiple sensitivity analyses whereby analyses were restricted to studies adjusting for various additional confounders and for DM treatment, and stratified by DM type, in order to investigate robustness of observed associations.

| Definitions
For DM, we considered diagnoses of type 1 and type 2 DM, as well as unspecified DM, for pooling the effect, followed by further stratification by subtypes of diabetes if enough studies were eligible. Hypertension (HT) was defined by either a diagnosis of HT recorded as part of the medical history or current health assessment, or a measurement with mean arterial pressure (MAP) above a specified threshold. Obesity was based on BMI values, by referring to the cut-off in the included studies, where 25, 27 and 30 kg/m 2 were the common threshold values used. Cardiovascular disease (CVD) was defined broadly as an umbrella term including any of the following disease subtypes: ischaemic heart disease (IHD)/coronary heart disease (CHD) and cerebrovascular disease. Dyslipidaemia was defined according to serum lipid concentrations above a certain threshold defined in the primary studies (thresholds may vary depending on healthcare setting).

| Quality appraisal
We used the Newcastle-Ottawa Scale (NOS) to assess the quality of nonrandomized studies, including cohort and case-control studies, 29 judging studies based on points awarded for selection of study groups, comparability of groups and exposure/outcome ascertainment. Studies with scores of <5, 5-7 and >7 points were considered to be of low, sufficient and high quality, respectively.

| Study characteristics
In total, our search identified 1,814 articles (899 from MEDLINE, 407 from EMBASE and 508 from Web of Science) ( Figure 1). After deduplication, 1136 individual articles were screened by title/abstract, from which 140 full texts were identified for full-text assessment.
After exclusion of ineligible articles and reference list searching of relevant articles, 40 articles were identified for inclusion in this review. Summary characteristics of included studies are reported in Table S2.
All studies were observational in design, with 33 cohort and seven case-control studies included (Table S2). Thirty-two studies were conducted in Asian countries. Four studies were restricted to male cohorts, and 36 were undertaken in mixed-sex cohorts. All studies recruited participants from health centres, health care or prescription databases, or pre-existing cohorts or cancer screening programmes. All studies were undertaken in adults, with mean/median ages of cohorts ranging between 40 and 65 years in 33 studies. Thirty-three studies investigated DM/insulin resistance/fasting serum glucose, 11 investigated blood hypertension/blood pressure, seven investigated dyslipidaemia, five investigated obesity and cardiovascular disease. Fewer than five studies investigated other factors including renal disease, statin use and use of antidiabetic drugs.
In the 40 studies including 536,456 adults, >30,500 HCC events occurred (we are unable to report an exact number, because one study did not report a precise number of HCC cases 30

| Quality assessment
Among 40 studies, 39 had quality scores ≥5 (Tables S3 and S4). All 33 cohort studies were of sufficient quality with 13 of these being scored as high quality. Six out of seven case-control studies were of sufficient quality and one of poor quality. Inclusion criteria varied widely, and therefore, study populations were heterogeneous.
In most studies, exposures and outcomes were ascertained using health assessment, imaging or record linkage. Twenty-three cohort studies and seven case-control studies accounted for age and sex.
HCC typically arises after long durations of infection, and therefore, prolonged follow-ups enable the detection of more HCC events; among 23 cohort studies identified, only five cohort studies had lengths of follow-up ≥10 years.

| Association of diabetes mellitus with HCC risk
Thirty-six studies investigated the association of DM with risk of CHB progression to HCC, comprising seven case-control studies (Table 1a) and 29 cohort studies (Table 1b). Four studies were restricted to males and the others included both sexes (Table S2) Effect sizes >1 ranged from 1.05 to 6.80, with 15 RRs being statistically significant. The single RR that was <1 was nonsignificant.
Minimal adjustment for confounders differed between case-control and cohort studies. Most case-control studies adjusted for age, sex, HCV coinfection, HIV coinfection and cirrhosis. Twenty cohort studies minimally adjusted for age and sex. Of these, 15 adjusted for HCV coinfection, 13 for cirrhosis, 12 for antiviral treatment, 10 for HIV coinfection, 9 for alcohol consumption, seven each for HBV viral DNA load and cigarette smoking and 6 for other liver disease (including alcoholic liver disease). Eight studies excluded participants who developed HCC within the first 3 to 12 months of follow-up in their main analyses. One study did so in a sensitivity analysis and found this did not modify associations observed.

| Meta-analysis of cohort studies
DM was associated with increased hazards of progression to HCC by metaanalysis restricted to HRs minimally adjusted for age and sex ( Figure 2).
As there was significant heterogeneity (I 2 = 52%, p < 0.01), results from both fixed-and random-effects analyses are presented. In random-effects analysis, the hazards of HCC were 36% higher (summary HR 1.36; 95% CI 1.23-1.49) in those with DM as compared to those without DM.

| Subgroup and sensitivity analyses
We performed sensitivity analyses in order to investigate the robustness of pooled estimates to additional adjustment for HCV or HIV coinfection, cirrhosis and DM treatment. After restricting metaanalysis to 16 studies adjusting for HCV coinfection in addition to age and sex ( Figure S1), pooled HRs did not change materially.
Considering eight studies adjusting for HIV and antiviral treatment ( Figure S2), pooled HR from the fixed-effects analysis was attenuated towards the null but remained significant. To investigate the robustness of the association of DM with HCC to adjustment for cirrhosis, a potential mediator, we restricted meta-analysis to studies adjusting for cirrhosis ( Figure S3). This did not change pooled HRs materially.
To investigate heterogeneity between type 2 DM and unspecified DM, sensitivity analysis was performed whereby studies were stratified by DM type. Among studies investigating type 2 DM, heterogeneity was 33% (p = 0.18) ( Figure S4). However, this HR did not differ substantially to that observed in the primary meta-analysis.
The association of DM with increased HCC risk was attenuated towards the null in studies that adjusted for metformin use, with risk TA B L E 1 A Effect estimates for case-control studies investigating the association of diabetes mellitus with hepatocellular carcinoma risk in adults with HBV infection  Figure S5). After restricting to studies adjusting to DM treatment, pooled HRs remained statistically significant.

| Association of hypertension with hepatocellular carcinoma risk
Eleven studies investigated the association of HT with risk of CHB progression to HCC, one case-control study and 10 cohort studies (

| Associations of other comorbidities with hepatocellular carcinoma risk
Seven studies investigated the association of dyslipidaemia with HCC risk in CHB patients (

| DISCUSS ION
Our meta-analysis suggests that DM is a risk factor for HCC in   in small studies. [31][32][33][34] Our findings are consistent with a previous meta-analysis 35  Although HRs did not significantly vary with length of follow-up in sensitivity analysis, it is possible that variable lengths of follow-up also contributed to between-study heterogeneity. Generally, cancer is a chronic disease with a slow development, and preclinical disease can be present for many years before clinical manifestation; follow-up periods <10 years may be insufficient to detect HCC outcomes. We were unable to provide effect estimates across most potential patient subgroups because the subgroups contained small numbers of studies, putting subgroup analyses at greater risk of chance findings as well as being subject to the influence of multiple testing.
The association of DM with HCC we report in this meta-analysis is weaker than those observed in patients with chronic HCV infection. In studies of individuals with chronic HCV infection, risk of HCC was elevated ~twofold in the presence of DM. 37,38 Previous studies also report increased risks of DM in HCV-infected individuals as compared to noninfected individuals. [39][40][41][42] However, this may be due to the various extra-hepatic manifestations of HCV which are not so clearly associated in HBV infection. 43,44 In sensitivity analysis restricted to studies adjusting for cirrhosis, the observed association of DM with HCC was attenuated towards the null. This may be explained by a confounding of the association by cirrhosis, accounted for by an independent association of cirrhosis  Three studies adjusted for metformin use, [54][55][56] and in sensitivity analysis restricted to these studies, the association between DM and We included all studies investigating the association of comorbidities with risk of CHB progression to HCC that minimally adjusted for age and sex in order to provide a comprehensive review of available evidence.
However, few studies investigated non-DM comorbidities, preventing meta-analysis for these comorbidities. Additionally, we were unable to restrict our meta-analysis of DM and HCC to studies adjusting for further confounders in addition to age and sex, as few studies minimally adjusted for all relevant factors. Publication bias may influence the outcome, as we restricted our search to the peer-reviewed English-language literature, and studies that do not report an association of DM with HCC may be less likely to be published. Our results may not be generalizable to the global CHB population, as there were a limited number of studies from non-Asian countries. The lack of studies from any African countries is of concern, given that the region carries both the highest HBV prevalence 67 and largest mortality burdens for cirrhosis and HCC. 68,69 Our finding that DM is a risk factor for HCC in CHB-infected individuals suggests that enhanced cancer surveillance may be justified in patients with CHB and DM to enable early detection and treatment. Improvements in guidelines could help to inform more consistent approaches to risk reduction. After adjustment for metformin use, this association remained significant but was attenuated, suggesting a potential benefit of metformin that warrants further study. Ongoing investigation is required in order to identify and characterize risk factors for HCC, to extend these analyses to diverse global populations and to elucidate disease mechanisms in order to inform prevention, screening and therapeutic intervention.

ACK N OWLED G EM ENTS
CC acknowledges doctoral funding from GlaxoSmithKline and thanks Iain A Gillespie (GlaxoSmithKline UK) for his contributions to this study.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data sharing not applicable to this article as no data sets were generated during the current study; all data were obtained from published articles.