In recent years, increasing evidence has suggested a strong association between diabetes mellitus (DM) and hepatocellular carcinoma (HCC). To provide a quantitative assessment of this association, we performed a systematic review and meta-analysis of cohort studies. We collected studies through a literature search of Medline from January 1, 1966 and EMBASE from January 1, 1974, through July 31, 2010. Summary relative risks (SRRs) with their corresponding 95% confidence intervals (CIs) were calculated using a random-effects model. A total of 25 cohort studies that met our inclusion and exclusion criteria were included in our analysis. Among these, 18 studies showed that DM was associated with an increased incidence of HCC (SRRs = 2.01, 95% CI: 1.61–2.51), compared with individuals without DM. There was a statistically significant heterogeneity among these studies (Q = 136.68, p < 0.001, I2 = 87.6%). Analyses subgrouped by controlling confounders revealed that the increased incidence of HCC was independent of geographic location, alcohol consumption, history of cirrhosis, or infections with hepatitis B (HBV) or hepatitis C virus (HCV). In addition, DM was also positively associated with HCC mortality (SRR = 1.56; 95% CI: 1.30–1.87), with no significant evidence of heterogeneity among studies (Q = 1.16, p = 0.56, I2 =0%). There were no significant publication bias (p = 0.79 for Egger's regression asymmetry test). These findings strongly support a positive association between DM and increased risk of HCC in both males and females.
Hepatocellular carcinoma (HCC) is one of the most prevalent tumor types, with increased incidence and mortality in recent years. Annually, approximately 500,000 new cases are diagnosed worldwide and a nearly equivalent number die of this disease. Chronic infection with hepatitis B (HBV) or hepatitis C virus (HCV) has been considered the most important etiologic risk factors for human HCC.1 In addition, excessive alcohol consumption, cigarette smoking, and environmental exposure to aflatoxin are also risk factors for HCC development.2 However, many individuals who have been exposed to these risk factors never develop HCC, while 15–50% of cases occur among those without exposure.3
Coinciding with increased HCC incidence, the prevalence of DM has also grown markedly over the past two decades in most countries.4, 5 In recent years, DM has become highly suspect as a risk factor for several malignancies, including cancers of the breast,6 endometrium,7 pancreas,8 and liver.9 A systematic review and meta-analysis of 26 studies (13 case–control and 13 cohort) conducted in 2005 showed that individuals with DM had a 2.5-fold greater risk of HCC [pooled odds ratio in case−control studies = 2.5; 95% confidence interval (CI): 1.8–3.5; pooled relative risk (RR) in cohort studies = 2.5; 95% CI: 1.9–3.2] compared to controls. This significant association between diabetes and HCC was independent of alcohol consumption or hepatitis viral infections.9 Since then, other relevant studies on the association between DM and HCC have been published.10–16 In this study, we aimed to further analyze this relationship by conducting an updated detailed meta-analysis of cohort studies. We also evaluated whether this association differs to various study characteristics.
A comprehensive, computerized literature search was conducted in Medline (from January 1, 1966) and EMBASE (from January 1, 1974) through July 31, 2010 by two independent investigators. Research papers were selected using the following keywords or Medical Subject Heading (MeSH) terms: diabetes, or diabetes mellitus; liver cancer or hepatocellular carcinoma or HCC; and, epidemiologic studies. We also reviewed the reference lists of selected research papers to identify additional relevant studies. No language restrictions were imposed.
Inclusion and exclusion criteria
Studies were included in the meta-analysis if (1) they had a cohort design; (2) the exposure of interest was DM; (3) the outcome of interest was primary liver cancer or HCC; and (4) RR, hazard ratio (HR), or standardized incidence or standardized mortality ratios with their 95% CIs (or data to calculate them) were reported. If more than one study used the same cohort and objectives, the one with the most comprehensive population or most adjusted estimate of risk associated with history of DM were included. Using these criteria, two articles were excluded from our systematic review.17, 18 Articles or reports from non-peer-reviewed sources were not considered. We also excluded two articles that reported young-onset DM and HCC risk because most individuals with young-onset DM are type 1 DM.19, 20
We extracted the following data from each publication: the first author's last name, year of publication, country where the study was performed, sample size, how exposure was ascertained, estimates of exposure effects with their 95% CIs, and covariates adjusted for in the analysis. For each study, we extracted the risk estimates that reflected the greatest degree of control for potential confounders. When studies provided more than one RR according to the duration of diabetes, we combined the RRs extracted from individuals who were diagnosed with diabetes more than 1 year prior to the diagnosis of HCC. Two researchers independently performed the data extraction.
We divided epidemiologic studies of the relationship between diabetes and HCC risk into two general types according to the measurement of RRs: cohort studies (rate ratio or HR), and cohort studies of patients with diabetes using external population comparisons (standardized incidence or standardized mortality ratios). The latter cohort studies were analyzed separately. SRR estimates with their corresponding 95% CIs were derived with the method of DerSimonian and Laird using the assumptions of a random effects model, which incorporates between-study variability.21
We performed a sensitivity analysis in which one study at a time was removed, and the rest were analyzed to evaluate whether the results could have been affected markedly by a single study. We also conducted analyses stratified by (a) geographic area, (b) infection with HBV or HCV, (c) alcohol consumption, (d) cirrhosis, and (e) duration of follow-ups. Statistical heterogeneity between studies was evaluated with Cochran's Q test and the inconsistency index I2, which is the proportion of total variation that is due to between-study variation. A probability p-value close to 1 suggests a high probability that the observed heterogeneity was due to sampling error. Publication bias was evaluated using funnel plot and the Egger's test, which is based on a regression model to determine funnel plot asymmetry. For Egger's test, a p-value less than 0.10 was considered to indicate statistically significant publication bias. All statistical analyses were performed using STATA, version 11.0 (STATA, College Station, TX).
Characteristics in selected studies
A total of 25 cohort studies that met the inclusion and exclusion criteria were used in this systematic review (Tables 1 and 2). Of these, 20 studies used incidence and/or mortality rates as the measurement of RR4, 9, 11-17, 22-32 (Table 1), and five cohort studies used standardized incidence and/or mortality ratios as the measurement of RR33–37 (Table 2). The countries or continents in which the studies were conducted were: Japan (n = 9)11, 13, 22-25, 30-32; Taiwan (n = 4)4, 10, 12, 16; the United States (n = 4)26, 27, 29, 35, Europe (n = 6),15, 28, 33, 34, 36, 37 Europe and Canada (n = 1),14 and Korea (n = 1).38
Table 1. Characteristics of 20 cohort studies of DM and HCC incidence and mortality
Table 2. Characteristics of cohort studies of DM and HCC based on standardized incidence/mortality ratio
The cohorts were comprised of between 47 and 1,283,112 persons with a mean follow-up period of 8.8 years (range: 2.8–20 years). Most of the studies included both men and women, and only two studies consisted entirely of men,13, 25 and one study was 98% men.27 In six studies, DM was determined on the basis of a positive history only4, 10, 11, 15, 22, 26 and in three studies, the criteria for DM were not indicated clearly.25, 29, 31 In all other studies, DM was determined by blood glucose levels. In 12 studies, the outcomes were incidence or mortality of primary liver cancer.10, 11, 15, 22, 26, 32-38 In 13 studies, the outcomes were incidence or mortality of HCC4, 12-14, 16, 23-25, 27-31. HCC or liver cancer diagnosis was made by histological evaluation in five studies,13, 14, 24, 28, 30 a combination of increased alpha-fetoprotein and positive images in four studies,12, 23, 25, 31 and cancer registry alone in 12 studies.4, 10, 11, 15-17, 22, 26, 27, 29, 32, 37
Potential confounders (at least for age) were controlled in all of the studies, except for two studies.15, 39 In four studies, all subjects were reported to be infected with HCV.14, 23, 24, 30 Five studies reported that all subjects included were afflicted with cirrhosis from various causes,13, 14, 28, 29, 31 and two studies reported that all subjects were addicted to alcohol.13, 25
DM and HCC incidence
We identified 18 cohort studies that presented results on diabetes and HCC or liver cancer incidence. Of these, only five studies did not demonstrate a significantly increased risk of HCC or liver cancer in patients with diabetes,14, 24, 25, 29, 30 and the remaining 13 studies reported a significantly increased risk of HCC in diabetic individuals. As shown in Figure 1a, the SRRs with 95% CIs were 2.01 (95% CI: 1.61–2.51) in a random-effects model for individuals with diabetes, compared with individuals without diabetes. However, there was significant heterogeneity among these studies (Q = 136.68, p < 0.001, I2 = 87.6%).
In sensitivity analysis, the overall homogeneity and effect size were calculated by removing one study at a time. This analysis confirmed the stability of the positive association between DM and HCC incidence. For example, when we excluded the study of Chen et al.10 from the analysis (this was the study that clearly carried the most weight), the estimated pooled RR was somewhat similar (RR = 2.05, 95% CI: 1.72–2.43), and the heterogeneity remained the same.
Next, we conducted subgroup meta-analyses by various study characteristics (Table 3). The SRRs were similar for studies conducted in Asia (SRR = 2.08; 95% CI: 1.17–2.52; test for heterogeneity Q = 32.79, p = 0.001, I2 = 63.4%), and outside Asia (America and Europe; SRR = 1.93; 95% CI: 1.53–2.43; test for heterogeneity Q = 7.15, p = 0.125, I2 = 44.1%).
Table 3. Summary relative risks for the association between diabetes and liver cancer by geographical region, gender, duration of follow-up, HCV-infection patients, cirrhosis and adjustment for HBV, HCV, alcohol assumption
Four studies provided results on cancer incidence specific for gender10, 11, 32, 38 and three studies consisted entirely of men, or 98% men.13, 25, 27 In stratified analyses by gender, diabetes was associated with an increased risk of HCC in both males and females (SRRs =1.96, 95% CI: 1.71–2.24 in males, and SRRs = 1.66, 95% CI: 1.14–2.41 in females). The difference in SRRs across gender strata was not significant (z = 0.82, p = 0.41), although the SRRs in males was somewhat higher than those in females (Table 3).
In four studies that consisted of entirely HCV-positive patients,14, 23, 24, 30 DM was also found to be associated with an increased risk of HCC (SRRs = 2.31, 95% CI: 1.05–5.07), with evidence of heterogeneity (Q = 14.81, p = 0.002, I2 = 79.7). Five research studies consisted entirely of patients with cirrhosis due to heavy alcohol drinking or HCV infection.13, 14, 28, 29, 31 These cirrhotic individuals with a history of diabetes had a pooled RR of 1.98 and a 95% CI of 1.27–3.09, compared with cirrhotic patients without DM. There was significant heterogeneity among studies (Q = 8.03, p = 0.091, I2 = 50.2%; Table 3).
We also investigated the impact of confounding factors on the estimates of RR. Ten studies were controlled for infection with HBV or HCV, and the SRR with 95% CI was somewhat stronger in these studies than those that did not control for these infections (SRR = 2.22, 95% CI: 1.77–2.78 versus SRR = 1.86, 95% CI: 1.48–2.33).
When we restricted the meta-analysis to those studies controlled for alcohol consumption, the positive association between diabetes and risk of HCC remained (SRR = 2.09, 95% CI: 1.73–2.54), with statistically significant heterogeneity among studies (Q = 35.32; p < 0.001, I2 = 66.0%). The association between DM and HCC incidence was stronger in studies with a follow-up period longer than 6 years (SRR = 2.00; 95% CI: 1.72–2.32) than in studies in which the follow-up period was less than 6 years (SRR = 1.74; 95% CI: 1.42–2.14), but not significantly so.
When only those studies using verified diabetic cases (i.e., excluding self-reported cases and registry) were included in the meta-analysis, the positive association between diabetes and the risk of HCC did not change (SRR= 1.97, 95% CI: 1.53–2.54; p heterogeneity < 0.001). In addition, when subgroup analysis was stratified by outcome, the associations between diabetes and the risk of HCC and primary liver cancer were similar (SRR = 2.06, 95% CI: 1.64–2.60 for HCC, and SRR = 1.75, 95% CI: 1.25–2.47 for primary liver cancer; Table 3).
Three cohort studies reported standardized incidence rate,33, 34, 37 and a positive association between DM and HCC was found (SRR = 3.77; 95% CI: 2.96–4.81; test for heterogeneity Q = 2.53, p = 0.283, I2 = 20.8%). Two of these three studies reported categories for the duration of diabetes and were similar across the studies: 1–4 years, 5–10 years, and >10 years. Combining these two studies,33, 34 we found that individuals with diabetes for 1–4 years had a higher risk of developing liver cancer than those with diabetes for 5–10 years, although the difference was not significant (SRR = 3.76; 95% CI: 2.90–4.87 vs. SRR = 3.17; 95% CI: 2.40–4.17; z = 0.88, p = 0.38). However, the risk of liver cancer in patients with diabetes 1–4 years was lower than individuals who had had diabetes for more than 10 years (SRR = 4.15; 95% CI: 3.20–5.40; z = –0.53, p = 0.60).
DM and HCC mortality
We identified three cohort studies that reported results on diabetes and the mortality due to HCC or liver cancer22, 26, 38 (Table 1 and Fig. 1b). Combining the results from these studies yielded an SRR of 1.56 and 95% CI of 1.30–1.87 for patients with diabetes, compared to those without. There was no significant heterogeneity among the studies (Q = 1.16, p = 0.56, I2 = 0%). Two studies reported results specific for gender, and when the two studies were pooled specific for males or females, a positive association between DM and HCC mortality was found for both males (SRR = 1.84; 95% CI: 1.34–2.51; test for heterogeneity Q = 7.07, p = 0.008, I2 = 85.9%) and females (SRR = 1.31; 95% CI: 1.06–1.61; test for heterogeneity Q = 0.09, p = 0.77, I2 = 0%).
Four cohort studies reported standardized mortality rate,33, 35–37 and when these were pooled, a positive association between DM and HCC was also observed (SRR = 1.70; 95% CI: 1.11–2.59; test for heterogeneity Q = 12.66, p = 0.005, I2 = 76.3%).
There was no funnel plot asymmetry for the association between DM and HCC risk. The p-value for Egger's regression asymmetry test was 0.79, suggesting a low probability of publication bias (Fig. 2).
Our meta-analysis revealed that individuals with diabetes have a 2.0-fold increased risk of HCC, compared with nondiabetics. This significant association between diabetes and HCC was independent of alcohol consumption or viral hepatitis status. The link was observed in women and men, as well as individuals with or without a history of liver disease.
However, the study has several limitations. First, great heterogeneity exists in terms of population demographics, how diabetes was ascertained, duration of follow-ups and adjustment for confounders. We are unable to account for these differences, despite the use of appropriate meta-analytic techniques with random-effect models. Nevertheless, multiple subgroup analyses found that the risk estimate was robust across various quality components. Second, cohort studies, which are not subject to recall and selection biases, might be affected by detection bias because patients with diabetes are under increased medical surveillance and thus might be more likely to be diagnosed with HCC at an early stage. Third, most of the studies did not distinguish between type 1 and type 2 diabetes (although we excluded two studies which included only patients with young-onset diabetes), which might attenuate any true relation between diabetes and HCC risk, as type-1 diabetes may not be related to liver cancer risk.19 However, it is likely that the majority of individuals with diabetes included in these studies were type-2 diabetes because it is the most common form of diabetes particularly in older individuals. Fourth, confounding is also likely to be present since diabetes and HCC share several risk factors, such as HCV infection, heavy alcohol consumption and obesity. Indeed, several studies have suggested that HCV has a direct role in promoting DM risk: HCV core protein suppresses insulin signaling through a PA28gamma-dependent pathway,40 and pancreatic islets in HCV-infected humans exhibit morphologic changes as well as derangement in glucose-stimulated insulin release.41 However, adjustment for a wide range of potential confounders only marginally attenuated the relationship between diabetes and HCC risk. Finally, as in any meta-analysis, the possibility of publication bias is a concern because small studies with null results tend not to be published. Publication bias may have resulted in an overestimate of the relationship between DM and the risk of HCC. However, the results of funnel plot analysis and formal statistical tests did not produce evidence for such a bias.
Although the mechanisms underlying the association between DM and the risk of HCC are not clear, several possibilities have been hypothesized. In early stage type 2 diabetes, hyperinsulinemia may upregulate the production of insulin-like growth factor-1 (IGF-1). IGF-1 stimulates cellular proliferation and inhibits apoptosis within the liver.42 The involvement of insulin and IGF-1 in carcinogenesis of liver has been supported by in vitro studies, animal models, and epidemiologic studies.43 In addition, diabetes is also an important risk factor for development of nonalcoholic fatty liver disease (NAFLD), the most common form of chronic liver disease in developed countries.44 About 30% of individuals with NAFLD based on ultrasound were identified as having its severe form, nonalcoholic steatohepatitis (NASH), and up to 9% of NASH may progress to cirrhosis.45 Of patients with NASH-related cirrhosis 40–62% developed complications due to cirrhosis, including HCC, after 5–7 years of follow-up.46 Although the exact mechanism behind the development of HCC in NASH remains unclear, insulin resistance and the subsequent inflammatory cascade likely contribute to NASH's carcinogenic potential. Insulin resistance leads to increased release of multiple pro-inflammatory cytokines, including tumor necrosis factor-alpha (TNF-α), interleukin-6 (IL-6), and leptin, which favors the development of hepatic steatosis and inflammation and subsequent cancer within the liver.47 Jee et al.38 reported a graded dose-responsive association between fasting glucose and liver cancer risk, which also supported a causal relationship between the two variables.
Interestingly, the RR of liver cancer among individuals with a short history of diabetes (<5 years) was somewhat higher, though not significantly so, than for individuals with diabetes for 5–10 years (RR = 3.76, vs. 3.17). At least in some cases, this may be because diabetes could be an early indicator of the tumor, making reverse causation an explanation. Otherwise, the RR would be expected to increase, rather than decrease, with duration of diabetes.
This study demonstrated a slightly lower RR compared to a previous systematic review conducted by EL-Serag et al.,9 in which diabetic individuals had a 2.5-fold greater risk of HCC compared with individuals without DM. However, their study included both case–control and cohort studies. Case–control studies are subject to several biases, recall and selection bias among them, which tend to overestimate positive links even when there are none. In this study, we included 25 cohort studies and excluded those that were case-controlled, which will at least partially increase the validity of the association between DM and HCC risk. Although most studies found a significant positive association between diabetes and HCC incidence, there are some studies that found an increased, but not significant risk between these two variables.14, 24, 25, 29, 30 Such results may be due to a small number of samples or follow-ups of short duration.
In addition, we also included sufficient cohorts with known liver disease: four cohorts with HCV-infection14, 23, 24, 30 and five cohorts with cirrhosis.13, 25, 28, 29, 31 Among these studies, individuals with DM and known liver disease had an increased risk of HCC compared with those without diabetes. These findings suggested that DM might increase the risk of liver cancer even after the acquisition of chronic liver disease. Chen et al.4 found a synergistic association between DM and HCC risk among patients with both diabetes and hepatitis (B or C) infections, compared to those with the viral infections but without diabetes.
In summary, our meta-analysis results strongly support a positive association between diabetes and the risk of HCC. The mechanisms involved need to be elucidated in future studies.