A meta-analysis of case–control studies on the combined effect of hepatitis B and C virus infections in causing hepatocellular carcinoma in China

We investigated whether concurrent infection by hepatitis B virus (HBV) and hepatitis C virus (HCV) in China, a hyperepidemic area for these infections, was associated with a higher risk of causing hepatocellular carcinoma (HCC) than each infection alone in a meta-analysis in China, 32 case–control studies involving 3201 cases and 4005 controls, identified from a computer-based literature search from 1966 to 2004. The pooled odds ratio and 95% confidence interval (CI) for HBsAg positivity was 14.1 (95% CI: 10.6–18.8); for anti-HCV/HCV RNA positivity was 4.6 (95% CI: 3.6–5.9); for HBsAg positivity and anti-HCV/HCV RNA negativity were 15.6 (95% CI: 11.5–21.3); for HBsAg negativity and anti-HCV/HCV RNA positivity were 8.1 (95% CI: 5.0–13.0); and positivity for both HBsAg and anti-HCV/HCV RNA was 35.7 (95% CI: 26.2–48.5). We conclude that HBV and HCV infections are important independent risk factors for HCC in China, and that dual infection by HBV and HCV is associated with a higher risk of causing HCC than each infection alone, suggesting a synergism between HBV and HCV.

Hepatocellular carcinoma (HCC) is the fifth most common malignant neoplasm in the world, and the third most common cause of cancer-related death (Parkin et al, 2001). More than 500 000 new cases are currently diagnosed yearly, with an ageadjusted worldwide incidence of 5.5 -14.9 per 100 000 population (Llovet et al, 2003). About 110 000 persons die each year from HCC in China, which accounts for 45% of the deaths from HCC worldwide. China is also a hyperepidemic area for hepatitis B virus (HBV) infection (with an estimated carrier rate exceeding 10% in the general population) and hepatitis C virus (HCV) infection (with a prevalence of anti-HCV 3.1% of in the general population) (Xuezhong et al, 1999). HBV and HCV infections have each been shown to markedly increase the risk of developing HCC. In China, HBV has, for several decades, been the major cause of liver disease (Tang, 2001). Soon after the identification of HCV and the development of testing HCV infection, it has become increasingly evident that chronic HCV infection also plays an important role in chronic hepatic disease including HCC (Yano et al, 1993;Tsai et al, 1994). There have been several meta-analyses of the relation of HBV and HCV infections to HCC in China, but the relation of dual infection to HCC has not received similar attention. In fact, this is not rare in HCC patients and a higher morbidity in HCC patients with dual infection has been reported than in those with one infection (Benvegnu et al, 1994). The main difficulty in evaluating the relationship of dual infection to HCC is the rarity of concurrent infection in subjects without clinically evident liver disease. Only large case -control studies that include people unaffected by chronic liver diseases as controls can allow the interaction to be properly assessed. The aim of this meta-analysis is to assess the interaction between HBV and HCV infections among 32 casecontrol studies of HCC.

Data sources
We searched the published Chinese literature from 1979 to February 2004 and Medline database from 1966 to February 2004 to identify case -control studies of the combined effect of HBV and HCV in causing HCC using the search textwords 'hepatocellular carcinima' and 'HCV' or 'hepatitis non-A, non-B virus'. We also did a full manual search from bibliographies of selected papers. Studies were identified by two researchers independently and the two lists compared and discrepancies resolved. We also contacted the authors of studies containing relevant information who did not report their results suited to this analysis. Unpublished data were also accepted if an abstract was available and further information was obtained from the author. We contacted most of the journals on gastroenterology, hepatology and epidemiology in China for studies in press.

Study criteria
(i) Each case -control study was required to have published details of the number of participants and age-adjusted or age-matched odds ratio (OR) and 95% confidence interval (CI). (ii) We included only the studies that used the following serological markers of chronic viral infections: HBsAg for HBV infection, and anti-HCV (tested with enzyme-linked immunosorbent assay, ELISA) or HCV RNA (detected by reverse transfusion polymerase chain reaction (RT -PCR) for HCV infection. When data were available on both anti-HCV and HCV RNA tests, we used only the former for classifying subjects as positive or negative for HCV infection. (iii) Only studies were included that selected HCC patients as cases and subjects without chronic liver diseases as controls; blood donors and patients on usual blood transfusion, regular haemodialysis or peritoneal dialysis were not considered eligible as controls. (iv) The studies were ineligible if HBsAg positive subjects were excluded from controls. (v) Studies with less than 20 cases were excluded, since they could not detect any case positive for both HBsAg and anti-HCV/HCV RNA. (vi) When more than one publication from the same study was available, only the final one was used.

Data extraction
Data were independently extracted from each study using predefined forms, and disagreement was resolved by consensus. The following information was extracted: the area where studies were performed; number of cases and controls; mean age of caseand control-groups; type of controls; number of cases and controls for each category of HBV and HCV infections; the OR and 95% CI for each category of HBV and HCV infections. The most fully adjusted OR and 95% CI were extracted from each paper.

Statistical methods
The pooled OR and 95% CI were calculated from raw study data using the Mantel -Haenszel method (fixed effect model) as modified by Robbins et al (1986) or the DerSimonian -Laird method (random-effect model) (DerSimonian and Laird, 1986). When there was significant heterogeneity, the random-effect model was used; otherwise the fixed-effect model was used. Evidence of heterogeneity in the estimate of effect was checked using the Breslow-Day test. A linear regression of the reported ORs, or the ORs calculated by us, was also performed by a fixed effect weighted-mean/linear-regression analysis, using the OR variances as weights. Synergism between HBV and HCV infections was assessed by the method of Rothman (1986). To determine the source of heterogeneity, we analysed subgroups of case -control studies according to the following characteristics: type of controls (hospital vs community controls); geographical area (higher incidence area for HCC vs lower, the cutoff incidence being 30 per 100 000 population). (Studies in Jiangsu, Zhejiang, Guangdong, Guangxi, Fujian, Hainan were classified as higher incidence, and studies in other provinces as lower).

RESULTS
A total of 32 case -control studies satisfied the inclusion criteria (Xu et al, 1990;Gong et al, 1993;Zhang et al, 1993Zhang et al, , 1998Ye et al, 1994;Zhou et al, 1994;Fan et al, 1995a, b;Pan et al, 1995;Wang et al, 1996aWang et al, , b, 1999Guo and Yan, 1997;Lu, 1997;Lu and Hou, 1997;Ma et al, 1997;Shen et al, 1997;Yu et al, 1997;Zuo et al, 1997;Dai et al, 1998;Gao et al, 1998;Cui et al, 1999;Li et al, 1999Li et al, , 2002Yang et al, 1999;Zhao et al, 2000;Shi et al, 2001;Su et al, 2002;Wang, 2003;Wu et al, 2003;Zeng et al, 2003;Ding et al, 2004). Their basic characteristics are shown in Table 1. A total of 31 articles were published in Chinese and one in English. They included a total of 3201 cases and 4005 controls. Most studies took subjects, as controls, who were of the same sex, lived in the same area or saw a doctor at the same hospital as the cases. Mean age was not reported in 14 studies, but in most of these, the disparity in age between cases and controls was less than 3 or 5 years. Most studies showed no difference in mean age or in sex between cases and controls.
The summary OR and 95% CI for positivity for both HBsAg and anti-HCV/HCV RNA in the studies overall are presented in Table 2 by geographical area and type of controls. The test for heterogeneity showed significant differences (Po0.001) between studies whether considered overall or in subgroups. In the higher incidence areas, the summary ORs for HBsAg positivity and for anti-HCV/HCV RNA positivity were slightly higher than those found in lower incidence areas. The OR for HBsAg positivity using community controls was significantly higher than that with hospital controls. The OR for anti-HCV/HCV RNA positivity using community controls was rather similar to that using hospital controls.
The OR and 95% CI for each HBsAg and anti-HCV/HCV RNA combination in case and control groups are set out in Table 3 and 4. Large differences were observed, most studies showing a strong and statistically significant association between each marker and HCC. A very low proportion of cases unexposed to either marker was found in one study, which gave exceptionally high OR estimates for both HBsAg and anti-HCV/HCV RNA (Zhao et al, 2000). The proportion of HCC patients with anti-HCV/HCV RNA positivity was 19.13% among HBsAg positive patients and 28.01% among HBsAg negative patients in the studies overall (Po0.001). The corresponding proportions for reports using community controls were 15.46 and 21.70% (Po0.01), respectively, and using hospital controls, 22.84 and 32.01% (Po0.01).

DISCUSSION
This study represents a quantitative assessment of published data on the role of chronic HBV and HCV infections in causing HCC in China. The studies included differed widely in their OR estimates for each infective marker. Even after grouping the studies by geographical area and type of controls, we found significant differences within subgroups. However, in almost all studies, we found a strong association between HCC and positivity for both HBsAg and anti-HCV/HCV RNA, with summary OR 14.1 for HBsAg positivity and 4.6 for anti-HCV/HCV RNA positivity. In 1994, Zhao et al (1994) reported a summary OR for HBsAg positivity in HCC in China of 13.43, rather similar to our finding, indicating that the strength of the association in China has not altered much in the last decade. In 1998, Donato et al reported a summary OR for HBsAg positivity in HCC worldwide of 13.7, little different from that in western countries.
In 1997, Ge et al (1997) reported a summary OR for anti-HCV positivity in HCC in China of 6.7 and other workers reported a summary OR worldwide of 11.5 (Donato et al, 1998), while the summary OR in the present analysis of 4.6 was significantly lower than these. This is mainly due to the fact that the ORs for anti-HCV/HCV RNA positivity in HCC of most studies after 1997 were not high, especially in the larger of the studies (Yu et al, 1997;Gao et al, 1998;Li et al, 1999;Ding et al, 2004). The studies performed before 1997 mainly used the first or second generation anti-HCV test with low sensitivity and specificity, while since then third generation anti-HCV test and RT -PCR have been widely used with improved sensitivity and specificity. The OR for HBsAg positivity and for anti-HCV/HCV RNA positivity using hospital controls in this analysis were 10.5 and 4.4, both lower than the corresponding estimates using community controls (18.0 and 4.7), probably because chronic infections by HBV and HCV were more prevalent in hospital patients than those in the general population (Donato et al, 1998).
The above studies did not evaluate the dual infection by HBV and HCV as in our analysis. The OR for HBsAg positivity anti-HCV/HCV RNA negativity and for HBsAg negativity anti-HCV/ HCV RNA positivity in this analysis were 15.6 and 8.0, respectively. Although these indicate a strong association between each infection alone and HCC in China, the corresponding ORs worldwide of 22.5 and 17.3 were significantly higher, especially for HCV, showing that the risk is higher in western countries. In fact, the proportion of HCC patients for anti-HCV/HCV RNA positivity was 19.13% among HBsAg positive patients and 28.01% among HBsAg negative patients, and the corresponding proportions worldwide were 19.3 and 35.8% (Donato et al, 1998), which also supported the results. All the above results indicate that chronic infection by HBV and HCV alone were independent high risk factors for HCC.
In this analysis, the OR for positivity for both HBsAg and anti-HCV/HCV RNA was higher than the sum of the OR for HBsAg positivity anti-HCV/HCV RNA negativity and for HBsAg negativity anti-HCV/HCV RNA positivity, 35.7 vs 23.7 in total studies, 44.9 vs 17.8 in studies using hospital controls and 39.5 vs 31.0 in studies using community controls. All the results mentioned above indicate that the concurrent infection by HBV and HCV was associated with a much higher risk of HCC than each infection alone in China, pointing to a synergism between HBV and HCV in HCC. We did not find significant differences between the results in higher and lower incidence areas, probably owing to the diversity of risk factors for HCC in China, such as aflatoxin intake, drinking pond water, eating pickle, etc.
The different mechanisms that have been hypothesised as being associated with development of HBV-or HCV-related cancer suggest that both viruses could play an active role at different steps of the carcinogenic process when they are present together in hepatocytes. Most evidence suggests that HBV is capable of initiating the neoplastic process, while HCV could act as a promoter, and that they may be synergistic in causing HCC (Donato et al, 1998).
As meta-analyses are based on published studies, bias and confounding factors may be present (Egger and Smith, 1998). The interaction of other risk factors with HBV and HCV infections was not studied in this analysis. Bias and potential confounding factors may not be well controlled due to the limited details in the literature, and may therefore slightly affect the results. For more accurate results, further studies are required.