Hepatitis C Virus and Risk of Lymphoma and Other Lymphoid Neoplasms: A Meta-analysis of Epidemiologic Studies

Introduction

Lymphoid neoplasms are malignant diseases arising from the lymphoid tissues (1). Several causative factors have been shown (e.g., immunodeficiency, Epstein Barr Virus (EBV), human T-cell leukemia virus, Helicobacter pylori) or suspected (e.g., pesticides, smoking habits, and some dietary aspects; refs. 2, 3); however, partly because of their heterogeneity, in terms of histologic characteristics and anatomic location, the identification of the causes of lymphomas has turned out to be difficult.

Hepatitis C Virus (HCV) is an RNA virus belonging to the family of flaviviruses. HCV is hepatotrophic and causes hepatitis, liver cirrhosis, and hepatocellular carcinoma (4). HCV is involved in the etiology of essential mixed cryoglobulinemia, a lymphoproliferative disease that can evolve into B-cell non–Hodgkin lymphoma (NHL; B-NHL; ref. 5). Approximately 170 million people are infected with HCV worldwide, making HCV a major public health problem.

Several studies found a high prevalence of HCV seropositivity in patients with B-cell lymphoproliferative disorders, particularly B-cell NHL, including cases where essential mixed cryoglobulinemia was absent (6-8). The association between HCV and B-NHL also seems to have a strong regional connotation (e.g., Italy and Japan; ref. 8).

Three meta-analyses on HCV and NHL were published between 2003 and 2004 (6-8), and eight case-control studies and one cohort study were published thereafter (9-18). The aim of the present study was to systematically review epidemiologic data on the association between NHL and other lymphoid neoplasms with HCV while adequately adjusting for at least sex and age and assessing major sources of heterogeneity in the available data. Moreover, pooled results for subtypes of B-NHL and other lymphoid neoplasms are presented for the first time.

Materials and Methods

Search Strategy for Identification of Studies

To identify the relevant literature, searches of MEDLINE and EMBASE for articles on HCV and lymphomas were conducted up to July 2006. The search strategy was based on the following words (all fields): [“hepatitis C” (MeSH) or “hepatitis C virus” or HCV] and [lymphoma (MeSH) or “lymphoma, non-Hodgkin” (MeSH) or myeloma (MeSH) or “lymphomproliferative disease” or “lymphoproliferative disorder”]. Only articles published in English were included. Additional relevant articles were retrieved from the reference lists of the identified papers and of large recent reviews (2, 3, 6-8).

Inclusion Criteria

Eligible papers had to include the prevalence of HCV infection in NHL patients and in a control group. The definition of control groups varied greatly between studies, e.g., patients with lymphoproliferative diseases other than NHL, solid cancers, patients undergoing a given hospital procedure such as biopsy or colonoscopy, blood donors, or population-based samples. Most recent studies were designed as formal case-control or cohort studies, with clearly defined inclusion criteria for the control group, and estimates of odds ratios (OR) or relative risks (RR) adjusted for age, sex, and other potential confounding factors. The term RR was also used in tables for case-control studies, instead of OR. Patients with other lymphoproliferative diseases were never used as a control group in the present meta-analysis, because these conditions may also be associated with HCV seropositivity. Moreover, because the prevalence of HCV seropositivity varied widely according to age and sex, only studies that fulfilled at least one of the following conditions were considered eligible:

• Sex- and age-adjusted RRs were presented.

• Matching of cases and controls by age and sex was reported.

• Some measure of age (mean, median, or distribution) and sex ratio in cases and controls, and evidence of good comparability of the two groups, were provided.

To avoid multiple inclusions of the same data, whenever the same author appeared in more than one article, the place and time of recruitment of study subjects were checked, and, in case of overlapping, only the most recent publication was included. Only studies with ≥100 cases were included in the overall analyses on NHL and hepatitis viruses. Findings for specific NHL subtypes and other lymphoid neoplasms were shown only for studies where ≥20 cases had been collected. All prospective studies (including a case-control study nested in a cohort) were included regardless of the number of lymphoid neoplasms reported.

Assessment of Study Quality

The definition of lymphoma adopted by the various authors differed and the definition given by the authors was used whenever possible. Chronic lymphocytic leukemia (CLL), for instance, is similar to small lymphocytic lymphoma (SLL; ref. 1) and was included among NHLs by most, but not all, authors. Consequently, when some studies separately published results for CLL/SLL and other NHL, they were combined in the present pooled analysis. Patients with Human Immunodeficiency Virus (HIV) infection were explicitly excluded in most studies, and studies on HCV and NHL among HIV patients were excluded from the present meta-analysis.

The method of testing HCV infection also differed among studies; some relied only on Enzyme-Linked Immunosorbent Assay (ELISA), whereas others also used recombinant immunoblot assay confirmation and/or looked for HCV RNA. First-generation ELISAs had unsatisfactory sensitivity and specificity, therefore our review considered only studies where second- or third-generation ELISAs were used. The presence of HCV RNA is the best marker for hepatocellular carcinoma risk, but whether detection of HCV RNA, in addition to anti-HCV antibodies, is a requirement in the association between HCV and NHL is still unclear. The availability of HCV-RNA findings was not, therefore, a prerequisite for inclusion in the present study.

With respect to NHL classification, most studies used, or could be recoded into, the WHO/Revised European-American classification of lymphoid neoplasms (1) classification. To reduce differences and errors in the classification of NHL subcategories, only a few major NHL subtypes (e.g., diffuse large B-cell, follicular B cell, all types of marginal zone, T-cell, and CLL/SLL) were evaluated separately in the present study.

To investigate the specificity of the association of HCV with NHL, the prevalence of HCV in patients with Hodgkin's lymphoma and multiple myeloma was also assessed. Few studies showed results for leukemia other than CLL/SLL (17, 18).

Statistical Methods

To provide a quantitative estimate of the risk of lymphoid neoplasms among HCV-positive individuals from all studies combined, the RRs and the corresponding 95% confidence intervals (95% CI) were abstracted from published articles. Whenever available, we used the estimates adjusted for relevant confounding factors included in the original studies. The variance of the log RR was derived from the log CIs, for case-control studies, and estimated from the number of cases observed/expected for cohort studies. When RRs were not given, they were computed from tabular data using Wald estimates (19).

We then calculated the summary RR and corresponding 95% CI using random-effects models of DerSimonian and Laird (20), which incorporates both within- and between-study variabilities, as a weighted average of the estimated RRs, by giving each study a weight proportional to its precision (i.e., to the inverse of the variance of the log RR). Although presented in figures, studies without events (HCV-positive cases) in any subgroup were excluded from pooled estimates, as RR variance could not be estimated (20, 21).

Statistical heterogeneity among studies was evaluated using the Q statistics (20). We conducted subgroup analyses by study design (cohort versus case-control), prevalence of HCV among NHL-free study subjects (below 5% versus above, as a proxy of population HCV-prevalence), geographic area of the study (Southern Europe, United States, Asia, other), the publication year (before 2002 versus 2003 or thereafter), and type of control subjects (hospital-based versus population-based) in case-control studies. Publication bias was assessed with Begg's and Egger's asymmetry tests (20, 22) and Begg's funnel plot was reported herein.

The results of the meta-analysis were presented graphically, plotting the RR as a black square, whose size was inversely proportional to the variance of the log RR. For several studies, the 95% CIs slightly differed from those published in the original papers because they were derived from estimated study variances (19, 20). Diamonds were used to plot the summary RR for all studies and for different study designs; their center represents the RR and their extremes the 95% CI.

Results

Table 1 summarizes the main characteristics of studies on NHL included in the present review (9-16, 18, 23-32). The highest prevalence of HCV in a control group (over 20%) was reported in Egypt (9). High prevalences (5-10%) were also found in Italy (15, 18, 25, 27, 29, 31). Conversely, relatively low HCV prevalence (<5%) was found in South Korea (24), Spain (10), France (14), United States (12, 13), and Australia (16). The majority of the studies (13 of 18) included different histologic NHL subtypes.

The pooled RR from the 18 studies on HCV and NHL risk was 2.5 (95% CI, 2.1-3.0), and significantly elevated RRs were found in most of them (12 of 18; Fig. 1 ). The corresponding RRs from 15 case-control studies was 2.5 (95% CI, 2.1-3.1), based on 4,678 cases (509 HCV-positive), and the pooled RR from cohort studies was 2.0 (95% CI, 1.8-2.2). Significant heterogeneity (P < 0.01) was found between case-control studies and by study design. In the analyses stratified by selected study characteristics (Fig. 2 ), the greatest source of heterogeneity seemed to be HCV prevalence among NHL-free study subjects. Studies conducted where HCV prevalence among NHL-free subjects was above 5% (9, 15, 18, 23, 25-29, 31, 32) showed more elevated RR (3.0; 95% CI, 2.4-3.8) than studies where prevalence was lower (1.9; 95% CI 1.8-2.1; refs. 10-14, 16, 24, 30). Significant heterogeneity also emerged between geographic areas (RR, 2.7 in Southern European studies, 2.6 in USA studies, and 2.1 in Japanese/Korean ones), period (RR was 2.9 for studies published up to 2003 compared with 2.2 for studies published thereafter), and type of controls (RR of 2.7 for hospital-based and 1.9 for population-based case-control studies; Fig. 2).

Publication bias was explored graphically in Fig. 3 . A good symmetrical distribution of the studies emerged, and neither Begg's nor Egger's tests support the presence of publication bias.

The five case-control studies that showed RRs separately for nodal and extranodal NHL are reported in Fig. 4 (12, 15, 26, 27, 32). Pooled RRs were 2.5 (95% CI, 2.1-2.9) for nodal NHL, based on 887 cases, and 3.7 (95% CI, 2.4-5.9) for extranodal NHL, based on 471 cases. For extranodal NHL, but not for nodal NHL, heterogeneity emerged between studies. The exclusion of the study by Zuckerman et al. (32), which showed the highest RRs, notably for extranodal NHL, made the RR for nodal and extranodal NHL very similar (RR, 2.4; 95% CI, 2.1-2.8 and RR, 3.0; 95% CI, 2.1-4.4, respectively).

Studies including at least 20 cases of major specific B-NHL subtypes [i.e., diffuse large B-cell, follicular B-cell, marginal zone (including Mucosa-Associated Lymphoid Tissue (MALT), splenic, extranodal, and nodal marginal zone), and CLL/SLL ref. 1] are shown in Fig. 5 (11-16, 18, 25, 26). Similarly, positive associations with HCV positivity were seen for all NHL histologies examined. In particular, RR was 2.7 (95% CI, 1.9-3.7) for diffuse large B cell (1,020 cases), 2.7 (95% CI, 2.2-3.4) for follicular NHL (542 cases), 3.4 (95% CI, 2.4-4.9) for marginal zone (220 cases), and, to a lesser extent, for CLL/SLL (RR, 1.7; 95% CI, 1.4-2.0). Heterogeneity between studies was present only for diffuse large B-cell. In two of seven studies (14, 15), no cases with follicular NHL were HCV positive.

Figure 6 shows the association of HCV with T-NHL, Hodgkin lymphoma, and multiple myeloma (plasma cell myeloma according to WHO classification; ref. 1). Six studies included data on T-NHL (10, 12, 13, 18, 23, 26) and the corresponding pooled RR was 1.5 (95% CI, 1.1-2.1), based on 122 T-NHL cases. No significantly increased RR emerged in any study of Hodgkin lymphoma and the pooled RR was 1.5 (95% CI, 1.0-2.1), based on 337 Hodgkin lymphoma cases (Fig. 6; refs. 10, 18, 27, 33).

With respect to multiple myeloma, two studies, one case-control in Italy (27) and one cohort in Sweden (11), showed a direct association with HCV seropositivity, with RRs of 4.5 (95% CI, 1.9-10.7) and 2.4 (95% CI, 1.0-5.0), respectively. However, two multicenter case-control studies, one from Italy (18) and one from Spain (10), showed a nonsignificant decreased risk in patients with multiple myeloma (Fig. 6). The pooled RR was 1.6 (95% CI, 0.7-3.6), but the heterogeneity between studies was highly significant.

With regard to HCV seropositivity and leukemia risk, Bianco et al. (18) reported on 49 cases of chronic myeloid leukemia (RR, 2.3; 95% CI, 0.8-6.4), 140 cases of acute myeloid leukemia (RR, 1.3; 95% CI, 0.6-2.9), and 54 cases of acute lymphocytic leukemia (RR, 2.4; 95% CI, 0.8-7.1). Gentile et al. (17), in a study on leukemia and HCV seropositivity, found no association for acute lymphocytic leukemia (67 cases; RR, 1.3; 95% CI, 0.5-3.5), for acute myeloid leukemia (172 cases; RR, 1.5; 95% CI, 0.8-2.9), or for chronic myeloid leukemia (125 cases; RR, 0.3; 95% CI, 0.1-1.1).

Discussion

The present meta-analysis provides a sex- and age-adjusted estimates of the association (RR, 2.5) between HCV seropositivity and NHL. In addition, for the first time, pooled RRs could be computed separately for nodal and extranodal NHL and for several histologic subtypes, showing that the association with HCV is present and of similar magnitude in all NHL subgroups (Figs. 4-6).

Some early studies found that HCV seropositivity was more strongly related to extranodal NHL (27, 32), or to specific NHL sites of involvement (e.g., liver refs. 34), than to nodal NHL. Present results show slightly higher RRs for extranodal compared with nodal NHL, but this difference was largely due to early studies. Some reports also suggested that the relationship between HCV and NHL might be confined to certain histologic subtypes of NHL (35). In the present meta-analysis, no clear difference emerged in the association between HCV and major B-NHL subtypes (diffuse large B cell, follicular, marginal zone, and CLL/SLL).

A few meta-analyses have previously reported higher RRs for HCV and NHL than our study. Gisbert et al. (6), including only B-cell NHL, showed a pooled RR of 10.8 from nine studies that compared NHL cases with blood donors or healthy subjects, and a pooled RR of 4.2 from 16 studies where other hematologic malignancies were used as a comparison group. Matsuo et al. (7), which included only case-control studies that used noncancer patients as controls, estimated a pooled RR of 5.7 from 23 studies, but they did not take into account possible differences in age distribution or other confounding factors between NHL cases and the control group. HCV prevalence is strongly related to age, although the age groups where a peak in HCV prevalence is seen varied somewhat from one country to another. The good comparability of NHL cases with control subjects by age group is thus essential to provide an unbiased RR (19). Consequently, the lower RRs reported in the present study compared with previous meta-analyses (6, 7) are largely explained by the different inclusion criteria we used, mainly with respect to the possibility of adjusting for age and sex.

Only a small proportion of studies included in the two previous meta-analyses [6 of 23 (6) and 8 of 23 (7)] fulfilled the rather strict inclusion criteria we adopted in the present meta-analysis in relation to comparability or adjustment for age and sex. Moreover, half (9 of 18) of the original studies included in the present overview were published after previous meta-analyses, including >70% of cases and 60% of controls.

It should be remembered that the overall associations estimated by random-effect models should be interpreted as the average of the RRs across studies (20) and the main sources of heterogeneity of RR estimates between available studies must be considered before drawing final conclusions on the real strength of the association between HCV and lymphomas. The most relevant source of heterogeneity was prevalence of HCV among NHL-free study subjects with a more than 3-fold excess risk of NHL in areas with ≥5% prevalence of HCV, compared with less than 2-fold excess in lower prevalence areas. This result suggests the possibility that, in areas where HCV prevalence among NHL-free subjects is low, the spread of the virus might be recent (36) and not have yet produced full consequences on NHL development.

Other significant sources of heterogeneity were study design and geographic area. Cohort studies reported consistently lower RRs than case-control ones and, among them, hospital-based case-control studies showed the highest RR estimates. However, there is a clear correlation between high HCV seropositivity among NHL-free study subjects and the study's case-control design and location in South Europe.

Another possible source of heterogeneity between studies could be the definition of HCV infection that was not consistent across studies or different ELISA generations used. Some studies defined HCV positivity as positivity for HCV antibodies only (24), others as positivity for both HCV antibodies and HCV RNA (9), whereas the majority defined it as positivity for either marker (10, 12, 13, 25). However, the pooled RRs were substantially the same using the more restrictive definition (ELISA3-positive and RNA-positive). Thus, when only studies using third-generation ELISA were included, the pooled RR was 2.5 (95% CI, 2.1-3.0), suggesting that HCV definition and ELISA generation neither explained the heterogeneity between studies nor induced substantial bias.

Publication bias could have somewhat affected the present findings inflating the prevalence of HCV positivity and the RR estimates in early case-control studies (8). Thus, publication bias was evaluated by formal statistical tests (20, 22) in previous meta-analyses providing inconsistent results, absent in one case (7) but strong in the other (6). In the present study, no relevant publication bias emerged.

A direct oncogenic role of HCV through infection and deregulation of B cells has been hypothesized, as the virus is lymphotropic (37). However, only a small subset of the neoplastic cells in HCV-associated NHL harbor the viral genome or viral proteins, whereas these are frequently found in the stromal cells surrounding neoplastic cells (38). It is, therefore, likelier that neoplastic transformation is causally linked to the chronic antigen stimulation (indirect effect) of B cells by HCV rather than to direct oncogenic effect (5). Immunoglobulin-variable region genes expressed by B-NHL cells from HCV-positive patients have been shown to exhibit somatic mutations indicative of an antigen-selection process (39). Moreover, the broad spectrum of histologic NHL subtypes we found to be associated with HCV infection suggested an involvement of the majority of germinal and postgerminal center B-NHLs, as expected. Indeed, HCV-related lymphomagenesis occurs when B-cells proliferate in response to a virus-associated antigen (25, 38). Notably, the sequences of B-cell receptors in HCV-associated lymphoproliferations showed a certain similarity with anti-HCV antibodies (39).

Six distinct but related HCV genotypes and multiple subtypes have been identified (4). In the United States and Western Europe, genotypes 1a and 1b are the most common, followed by genotypes 2 and 3. Other genotypes are rare in the United States and Western Europe, but common in other areas, such as Egypt (genotype 4), South Africa (genotype 5), and Southeast Asia (genotype 6). Knowledge of the genotype is important because it has been shown to have a predictive value in terms of response to antiviral therapy in HCV-positive patients with chronic hepatitis (4). However, no clear difference in excess risk for NHL emerged between different genotypes in most of the studies providing relevant information (10, 15, 23, 25, 32).

The reactivation of HCV infection due to lymphoma-related immunosuppression may account, at least in part, for the high HCV positivity found among lymphoma patients. However, reactivation of HCV infection has been reported after cytotoxic treatment (24, 25), but not in untreated HIV-negative NHL patients, who were the only eligible cases in many recent case-control studies (10, 12, 13, 15, 18, 24, 25). The most important potential problem is, however, the presence of unknown confounding factor(s). Several viruses are involved in the etiopathogenesis of lymphomas and others, not yet identified, may be implicated. It is, therefore, impossible to rule out the possibility that HCV may be a marker of some other unknown virus that has the same transmission route as HCV and may constitute a real cause of NHL.

In terms of attributable risks (40), assuming an RR of 2.5, the fraction of NHL attributable to HCV infection would be upward of 10% in countries such as Italy, where ∼20% of NHL cases were found to be HCV positive (15). In countries where the prevalence of HCV seropositivity in the general population is very low (<1%), the proportion of NHL attributable to HCV infection would be less than 1%; thus, it could have easily been missed in studies including a few hundred NHL cases.