Introduction

Colorectal cancer (CRC) is one of the leading causes of cancer-related deaths worldwide. The morbidity and mortality increased quickly in both developing and developed countries [1–4]. In America, it was estimated that 148,610 new cases of colorectal cancer would be diagnosed and 55,170 patients would die from colorectal cancer in 2006 [3]. In China, from 1992/1993 to 2000/2001, CRC incidence had increased from 13.06 to 19.37 per 100,000, and the morbidity and mortality in China will be 26.12 and 17.43 per 100,000 respectively in 2006 [4]. Given the substantial injury for human quality of life, cancer prevention in its forms emerges as a very attractive approach [5]. Study results supported the colorectal cancer etiological hypothesis of “deficiency of dietary fiber and vegetables”, excessive intake of fatty food, cigarette smoking, alcohol consumption, and other environment factors [6–9]. The potent role of genetic instability in initiation and progression of colorectal cancers also has been well defined. Germline mutations of DNA mismatch repair genes are considered responsible for the development of colorectal cancer [10–12]. Recent publications reported higher levels of Helicobacter pylori antibodies in patients with colorectal cancer than that in control subjects. However, this is quite controversial for different studies hold various results [13–16].

H. pylori are Gram-negative bacterium. Its infection has been recognized as a major risk factor for gastric cancer by the International Agency for Research on Cancer in 1994 [17], and infected people are also at risk of developing a rare B cell tumor, so-called gastric mucosal-associated lymphoid tissue lymphoma [18–20]. Several studies suggested that H. pylori can increase plasma level of gastrin and stimulate mucosal cell proliferation. Persistent exposure to H. pylori for several years may result in gastric atrophy, cell mutation, and transformation of gastric mucosal cells into gastrin-producing cells, which also express gastrin receptors serving to stimulate cell proliferation and tumor growth. During this processes, the overexpression of cyclooxygenase-2 also stimulates the cells to release excessive amount of prostaglandin E2, leading to further proliferation [21, 22]. Whereas, how H. pylori predisposes gastric epithelial cells to become cancerous is not fully understood [23], the role of H. pylori in the development of colorectal cancer is still uncertain.

Therefore, we carried out a systematic review and meta-analysis of published studies to confirm this association between H. pylori infection and the risk of colorectal cancer.

Materials and methods

Literature search and standard of selection

Literature search was conducted in PubMed, EBSCO, HighWire Press, and other databases for relevant articles published up to Jan. 2007 about the association of H. pylori infection and colorectal cancer and/or colorectal adenoma (known as the precancer). We used the following medical subject heading (Mesh) terms and/or text words: Helicobacter pylori, colorectal carcinoma, or colorectal cancer to look for case–control studies. Only original articles in English were used and the cited references in these published articles were also reviewed. Since the quality of studies is important for meta-analysis, they should provide the description of how patients and control subjects were selected, information on data collection, sample size, matching procedures, prevalence of each group, statistical methods, and how to control the potential confounding. All the studies in our review are full articles.

Two reviewers independently applied eligibility criteria, assessed the studies for methodological quality, and finished data extraction simultaneously. The results were combined and cross-checked, and any differences were resolved by reviewing the original until conformity. Information abstracted included the author, year of publication, country of origin, type of design, number of patients and controls, matching condition, prevalence of H. pylori in each group, odds ratio (OR), adjustment of confounders, etc.

Assessment of homogeneity and treatment effect

OR with 95% confidence interval (CI) was computed for each study, and pooled estimates were generated by MIX (Meta-analysis with interactive explanations—version 1.51; http://www.mix-for-meta-analysis.info). Statistical homogeneity between studies was assessed using the Cochrane Q value. Since there was an obvious heterogeneity between studies (Q = 31.11, P = 0.0033, I 2 = 58.21%) [24], the random effect model was used, which has the effect of giving more weight to the smaller studies than the fixed effect model [25].

Graphical evaluation of publication bias and statistical analysis

One of the great problems in systematic review is that not all studies can be published. Those studies with statistically significant results are more likely to be submitted and published than studies without significant results. If smaller studies without significant results remain unpublished, publication bias may occur in meta-analysis [26]. The funnel plot is used as the main graphical method for identifying publication bias, which is a scatter plot of the effect estimates from individual studies (horizontal axis) against sample size or some other indicators of the estimates (vertical axis). The most precise estimates (those from larger studies) are at the top of the funnel and those from less precise or smaller studies are more likely spread at the base of the funnel because of their larger standard errors, but they should also be symmetrically around the average. In the absence of bias, the plot resembles a symmetrical inverted funnel plot. When the gap at the bottom of the graph and the asymmetry of funnel are detected, a causal relationship must be claimed with caution.

As the supplement of the funnel plot, another statistical approach “trim and fill” was used as a formal procedure [27] (first, the trimmed funnel plot is used to estimate the center of the funnel, and then the omitted studies and their missing counterparts around the center are replaced, and finally, received the true center and the adjusted effect, including the “filled” studies) to adjust the summary OR in our study. Furthermore, the possible presence of publication bias was assessed by rank correlation method, linear regression approach, and fail-safe method [28, 29].

Results

Study characteristics

From the primary electronic database (PubMed), a total of 68 studies have been identified. Fifty-nine of them were considered ineligible for our review. The reasons were as follows: 43 studies were irrelevant to association of H. pylori infection and colorectal cancer; there was one case study, one animal study, nine clinical case reports, four comments, and one meta-analysis. After excluding, nine case–control studies met the inclusion criteria. In addition, we conducted a manual search for the other four cited references in these articles in PubMed, EBSCO, HighWire Press and other databases, which met the inclusion criteria. Therefore, 13 studies published from Dec. 1991 to Jan. 2007 were included in the final analysis. Table 1 shows the main characteristics of these studies.

Table 1 Characteristics of 13 studies
Full size table

Of the 13 studies, two were nested case–control studies [30], and one only chose male smokers as subjects [16]. All the cases were matched by age, gender, or other factors (race, social class, education, etc.). But three studies did not offer any information for matching [14, 31, 32], and only convenience samples were chosen (convenience sample means the person that is easy to find). Since one study used two groups of control subjects [33] (both of the two groups of control subjects were enrolled in this paper), 14 groups of data from 13 studies were obtained.

Overall analysis

Data from 13 studies with a total of 1,709 patients were included in the first meta-analysis. Since heterogeneity was detected, the random effect model was used and summary OR for colorectal cancer related to H. pylori infection was 1.49 (95% CI 1.17–1.91). Figure 1 shows the ORs, 95% CIs of each study, and summary OR. The assessment result of heterogeneity was: Q = 31.11, P = 0.0033, I 2 = 58.21%.

Fig. 1
figure 1

Meta-analysis of 14 groups of data without discriminating the difference of H. pylori measures

Full size image

There are serials of measures to detect the H. pylori infection status, such as the enzyme-linked immunosorbent assay (ELISA), C-urea breath test (UBT), rapid urease test, and histological diagnosis of biopsy specimens. One or more of them were used in the 13 studies. The detected results were indicated by different indicators, which represent different infection status of H. pylori. Therefore, carefully restricted analyses for certain determined assays of H. pylori status seem necessary.

As we have known, immunoglobulin G (IgG) demonstrates the previous infection of H. pylori; the results of other methods demonstrate the present infection status of H. pylori. Three studies with the determination of H. pylori status not measured with IgG antibody were excluded in the second meta-analysis. Summary OR of 11 data from ten studies was 1.56 in the random effect model (95% CI 1.14–2.14; the assessment result of heterogeneity: Q = 28.54, P = 0.0015, I 2 = 64.96%). Comparing with 1.49 (95% CI 1.17–1.91) in Fig. 1, summary OR 1.56 (95% CI 1.14–2.14) is a little higher. It seems that there is a strong relationship between H. pylori infection and colorectal cancer while using IgG as indicator of H. pylori infection (Fig. 2).

Fig. 2
figure 2

Meta-analysis of 11 groups of data for using the IgG antibody

Full size image

Figure 3 shows the funnel plot of publication bias with OR value as the horizontal axis and SE of OR as the vertical axis. Judged by means of eyeballing, the graphical funnel plot of the 13 case–control studies appears asymmetry. We can assume the possibility of publication bias. After adjustment by trimming and filling, summary OR was still 1.49 (95% CI 1.17–1.91). Therefore, H. pylori infection may increase the risk of colorectal cancer.

Fig. 3
figure 3

Funnel plot for 13 case–control studies. The solid line represents the estimate of summary OR and the dashed lines show the confine of 95% CIs

Full size image

Additionally, in any case, eyeballing examination remains a qualitative and subjective matter and therefore prone to bias. We also used the rank correlation method proposed by Begg et al. and the linear regression approach proposed by Egger et al. to evaluate the publication bias. Both of the two approaches can be used to assess the relationship between treatment effects and SE, which are similar statistical methods with the funnel plot. The regression method is more sensitive than the rank correlation approach, but the sensitivity of both of the two methods is generally lower in meta-analysis based on less than 20 trials [29]. Neither the Begg’s rank correlation (P = 0.324, two-tailed) nor the Egger’s regression tests (P = 0.257, two-tailed) show any evidence of publication bias in our study. Based on the classical fail-safe method, there must be another 72 null studies in our analysis, which can make the summary OR nonsense.

In meta-analysis, any single study may influence the final result. We evaluated the stability of the result by calculating the summary OR and one-leave-out ORs. This method is used to exclude each study and see what happens to the combined value. In our analysis, ORs of 14 data ranged from 0.73 to 3.78. After excluding any one, ORs ranged from 1.36 (95% CI 1.10–1.68) to 1.58 (95% CI 1.23–2.02). No single study markedly influenced summary OR as shown in Fig. 4.

Fig. 4
figure 4

The vertical solid line represents the original combined value when all studies are included, and the interrupted vertical line represents its confidence interval limitation. The solid plots represent combined values of one-leave-out ORs, not values of individual studies (inverse variance graph of ORs and confidence interval limitations for combined studies). Numbers and weights of the leave-out studies are indicated on the figure

Full size image

Discussion

It is estimated that about 50% of the world’s population carry H. pylori in their stomach [34]. Hypergastrinemia is originating initially from G cells activated by H. pylori to release gastrin, leading to excessive production of ammonia and proinflammatory cytokines (IL-1, IL-8, TNF-α, etc.) and overexpression of growth factors such as TGF-α and EGF. During this process, cyclooxygenase-2 interacts with cytokines and growth factors and stimulates inflamed tissue or cancer tissue to release excessive amount of prostaglandin E2, which caused increased cell proliferation, angiogenesis, mutagenesis, and decreased apoptosis. Their joint action leads to the development of cancer [21].

It has been reported that H. pylori-infected people have a four- to sixfold increased risk of subsequently developing gastric adenocarcinoma [35, 36]. Several epidemiology studies have demonstrated the association between H. pylori seropositive and colorectal cancer risk. However, the results from these studies remain inconclusive [16, 30, 33]. Until now, there was no systematical review on the relationship between H. pylori infection and colorectal neoplasia concerning the difference of methods for testing H. pylori infection with satisfactory analysis. The purposes of this study were therefore to assess the association between H. pylori infection and the risk of colorectal carcinoma and to clarify any test-based differences.

A total of 13 studies with 14 data have been cited in this paper (due to one study used two groups of control subjects). In total, 72.3% (1,236/1,709) of patients and 60.8% (1,139/1,872) of control subjects were infected with H. pylori. As can be seen in Fig. 1, ORs of these 14 data ranged from 0.73 (95% CI 0.30–1.79) to 3.78 (95% CI 1.50–9.51) and summary OR was 1.49 (95% CI 1.17–1.91). It indicated that H. pylori infection may increase the risk of colorectal cancer. Although the graphical funnel plot appeared asymmetrical, neither rank correlation method nor linear regression approach supported statistical evidence of publication bias, and the result of fail-safe also suggested that the effect of publication bias was small.

Current diagnostic tests for H. pylori involve histological diagnosis of biopsy specimens, UBT, rapid urease test, serological tests, etc. The test for H. pylori serum IgG is a rapid, accurate, and reliable method. Moreover, the IgG antibody presents the previous infection of H. pylori to some extent, and other indicators illustrate that H. pylori exists while colorectal carcinoma was detected. If a person is infected with H. pylori ahead of neoplasia, it will be more reasonable to presume that H. pylori may be a risk factor of carcinoma. If we could not ensure the temporal relationship between H. pylori infection and neoplasia, the causal relationship cannot be assessed. In the meta-analysis of Zumkeller et al., different testing methods (ELISA assay, UBT, urease test, histological diagnosis of biopsied gastric specimens) were combined to determine the H. pylori infection status, which raised doubts about their conclusions [37]. In our meta-analysis, the IgG antibody was detected in ten of the 13 cited studies. Studies of detecting IgG antibody were further analyzed, the results of which showed a little more strength of the association between H. pylori infection and colorectal cancer (OR increased from 1.49 to 1.56).

Some case reports suggested the regression of colorectal cancer after eradication of H. pylori [38–40] and disappearance of rectal mucosa-associated lymphoid tissue lymphoma following antibiotic therapy [41]. van de Wouw et al. reported that after antibiotic therapy the number of H. pylori decreased and consequently led to a negative result of UBT and some other tests [42, 43]. Kosunen et al. pointed out that after 6 weeks of eradication therapy, the IgG titers had fallen by 20–30%, and 6 to 12 months after treatment the titer was 50% or less [44]. Little information is mentioned about whether the patients included in this meta-analysis used antibiotic therapy or not. Compared with control subjects, colorectal cancer patients were more possible to use antibiotic therapy. If the colorectal cancer patients who have been treated with H. pylori eradication therapy were also recruited in these case–control studies, and at the same time H. pylori infection status was detected without using IgG antibody, a false negative result of the relationship would be concluded.

Heterogeneity of these studies may be partially explained by one or more of the following design features of most reviewed studies: clinic- or hospital-based (rather than community-based) subject populations, small sample sizes, and inadequate consideration of potential confounding variables in the data analyses. Meta-analysis is a statistical procedure which is susceptible to various biases. First, the studies reviewed in this paper were carried out in different countries (three in America, two in Israel, two in Japan, two in western Europe, and one in eastern Europe). The race, environment, geography, or other conditions may influence the prevalence of H. pylori. Second, the selection of control subjects in some case–control studies may distort the results because hospital-based controls may not be as representative as population-based controls. Third, development of colorectal carcinoma is a long lasting process and is effected by multiple factors, such as body mass index, dairy foods, nutrients intake, and physical exercise [45–48]. It has been found that men were more susceptible to H. pylori infection compared to women, which may be caused by the propensity of smoking and drinking [49, 50], heavy physical labor in men, and different work environment between men and women. The estimate ORs were adjusted for age and gender only in the studies of Talley et al. [31] and Siddheshwar et al. [51]. The estimate ORs were adjusted for body mass index, nutritional level, etc. only in the studies of Breuer-Katschinski et al. [33] and Limburg et al. [16].

We excluded the studies published in non-English language and letters and clinical reports about the association of H. pylori infection and colorectal cancer. Since the data provided usually were not sufficient for assessing the quality of their studies, bias may result and influence the analysis result.

With these limitations in mind, all analyses in this paper supported the hypothesis of the relationship between H. pylori and colorectal cancer. Moreover, both of the two meta-analyses before and after concerning the difference of testing methods for H. pylori infection revealed that there are small risk increases of H. pylori infection in colorectal cancer (OR increased from 1.49 to 1.56). With the possible presence of publication bias, this result should be interpreted cautiously. We recommended that the important role of H. pylori in the development of colorectal cancer should not be disregarded. Larger and methodologically rigorous analytical studies with confounders being controlled using reliable outcomes and exposure measures are needed to further confirm this association.