Effectiveness of Helicobacter pylori eradication in the prevention of primary gastric cancer in healthy asymptomatic people: A systematic review and meta-analysis comparing risk ratio with risk difference

Background Helicobacter pylori infection is strongly associated with gastric cancer occurrence. However, it is unclear whether eradication therapy reduces the risk of gastric cancer occurrence. We evaluated whether H. pylori eradication reduces the risk of primary gastric cancer by using both risk ratio (RR) and risk difference (RD). Methods Searches of PubMed, EMBASE, Google scholar, the Cochrane Library, and the Japan Medical Abstracts Society as well as those registered in databases of the Cochrane Central Register of Controlled Trials, metaRegister of Controlled Trials, ClinicalTrials.gov, controlled-trials.com, UMIN-CTR, JMACCT-CTR, and JAPIC-CTI between January 1965 and March 2017, supplemented with manual screening. Randomized controlled trials (RCTs) in which eradication therapy were implemented for the interventional group but not for the control group, and assessed the subsequent occurrence of primary gastric cancer as the main outcome. Two authors independently reviewed articles and extracted data. Integrated results for all data were presented as RR and RD. Results Seven studies met inclusion criteria. The reductions in risk of primary gastric cancer occurrence in terms of overall RR and RD were 0.67 (95% CI: 0.48 to 0.95) and -0.00 ([95% CI: -0.01 to 0.00]; number needed to treat: 125.5 [95% CI: 70.0 to 800.9]), respectively. Conclusions The effectiveness of H. pylori eradication therapy in suppressing the occurrence of primary gastric cancer was significant and comparable to that of previous studies in terms of the estimated RR. However, the estimated RD was slight and not statistically significant.


Introduction
Helicobacter pylori was designated a definite carcinogen of gastric cancer in 1994 by the International Agency for Research on Cancer (IARC) [1]. Infection of the gastric mucosa eventually leads to cancer via superficial gastritis, atrophic gastritis, intestinal metaplasia, and dysplasia [2]. Take et al. confirmed reductions in gastric cancer occurrence in follow-up studies of patients with peptic ulcers who had received eradication therapy [3,4]. Fukase et al. showed in a randomized controlled trial (RCT) that H. pylori eradication therapy following endoscopic treatment of early-stage gastric cancer reduced the occurrence of metachronous gastric cancer by approximately 30% [5].
Fuccio et al. conducted a meta-analysis of seven RCTs that had been published in 2009, and reported that H. pylori eradication significantly reduces primary gastric cancer by roughly 35% in patients with gastritis or precancerous lesions [6]. However, this analysis included an RCT, in which endoscopic resection of early-stage gastric cancer suppressed the occurrence of metachronous cancer [5], as well as a set of multiple articles for a single study [7]. Ford and Moayyedi re-examined these data when these studies were excluded, and found that although the effect size was comparable, the reduction was no longer significant [8]. Subsequently, Ford et al. conducted systematic reviews within Asia of gastritis or precancerous lesions triggered by H. pylori and demonstrated that gastric cancer occurrence may be prevented by eradication therapy, based on the results of six RCTs published in 2013 [9,10]. In 2016, Lee, et al. updated the systematic review and showed a similar result [11].
Risk ratio (RR), a synonym for relative risk, means the ratio of two risks, usually of exposed and not exposed [12]. An RR of <1 means the event is less likely to occur in the interventional group than in the control group. Risk difference (RD), a synonym for attributable risk, means the difference of two risk, usually risk in the exposed minus risk in the unexposed. When the probability of primary endpoint of intervention group and placebo group is low, it is sometimes replaced by a large number in terms of risk ratio despite being slightly expressed by risk difference. These four systematic reviews [6,[9][10][11] only used the risk ratio (RR) as an integrative index. In risk communication, the ratio index is concerned to exaggerate the effect size compared to the absolute value or difference index [13,14]. In based on the absolute value is inevitable for evaluating the size of the problem and is also useful for the deployment of policies. Although it has been repeatedly and strongly recommended to combine the risk difference and the risk ratio [15][16][17][18][19], it has been reported that the risk difference compared to the risk ratio is reported less [20][21][22]. Especially in the issue of primary prevention, in general, the absolute risk is small and the effect size of the intervention shown with the risk ratio will give an exaggerated impression.
This study aimed to re-examine the significance of H. pylori eradication therapy in suppressing gastric cancer by expanding the period and range of the literature search and by showing effect sizes with RR and risk difference (RD).

Methods
This systematic review and meta-analysis followed the PRISMA guidelines [23].

Literature search
We searched the literature for RCTs published between January 1965 and March 2017, using PubMed, EMBASE, Google scholar, and the Cochrane library as English literature databases. Considering the high prevalence of gastric cancer in Japan and the availability of Japanese research articles, we also searched the Japan Medical Abstracts Society database [24]. Search terms were as follows: Helicobacter pylori, gastric cancer, intestinal metaplasia, gastric atrophy, dysplasia, atrophic gastritis, and chronic gastritis (final search conducted in April 2017). Electronic search strategy for PubMed database was shown in Table 1.
A search was also conducted for clinical trials that had been registered by March 2017, using the following databases: the Cochrane Central Register of Controlled Trials (CENTRAL) [25] [30], and Japan Pharmaceutical Information Center Clinical Trials Information (JAPIC-CTI) [31] (final search conducted in April 2017). In addition, we searched for previous systematic reviews and meta-analyses on the same topic. After this, we read through all titles and abstracts of the following representative Japanese journals in gastroenterology: "I to Cho" (Stomach and Intestine, published by Igaku-Shoin Ltd., Tokyo), "Shokakinaishikyou" (Endoscopia Digestiva, published by Tokyo Igakusha Ltd., Tokyo), and "Rinsho Shokakinaika" (Clinical Gastroenterology, published by Nihon Medical Center Inc., Tokyo) (final search conducted in April 2017). We then read through the entire text of articles related to the present topic. In cases where the same study participants were observed for gastric cancer occurrence at different times, we used the research article with the longest follow-up period.
Finally, we checked for studies in which multiple articles resulted from a single study, wherein extended observations of the same research participants included in the initial phase of the study were reported after the observations from the initial study had been published. Specifically, we contacted corresponding authors via e-mail and asked whether a single study resulted in multiple articles [32]; we decided in advance that if there was no reply after one week, a reminder e-mail would be sent to the author, and that if there was still no reply, the follow-up would be discontinued.
We used the "PRISMA-2009-Checklist" to evaluate the quality of literature search and eligibility criteria for this systematic review and meta analysis in S1 Table. Eligibility criteria Inclusion criteria. Inclusion criteria were as follows: an RCT of H. pylori eradication therapy, or a study in which the occurrence of primary gastric cancer was subsequently followed up; an interventional group involving eradication therapy; a control group given "a placebo," "no treatment (observation)," "a supplement," or "an antacid containing a proton pump inhibitor (PPI)," none of which eradicated H. pylori; published in either English or Japanese; Search (((((randomized controlled trial) OR (randomized controlled trials))) AND ((chronic gastritis) OR (gastritis) OR (atrophic gastritis) OR (dysplasia) OR (gastric atrophy) OR (gastric metaplasia) OR (intestinal metaplasia))) AND helicobacter pylori) AND gastric cancer Filters: Humans; English; Japanese Exclusion criteria. Exclusion criteria were as follows: administration of eradication therapy in the control group; eradication therapy prior to group allocation; studies in which gastric cancer occurrence was not measured; studies that tracked whether metachronous cancer occurred after endoscopic therapy (endoscopic mucosal resection (EMR) or endoscopic submucosal dissection (ESD)) for primary cancer; animal studies; basic medical research; pathological research; reviews; and guidelines.

Main outcome measure
We designated primary gastric cancer occurrence as the outcome, regardless of whether the measure was the primary or secondary outcome in each study.

Data extraction
Based on eligibility criteria, two authors (T.S. and T.N.) independently checked the titles and summaries of all articles, searched and determined which were appropriate for inclusion. The other co-authors checked every process, and any issues that arose were resolved by discussion. The entire text of included articles was read, and the following information was extracted: year of publication; publication format; protocol; number of participating facilities in each study; subject characteristics; study country; language of publication; presence or absence of upper gastrointestinal symptoms at study initiation; method of eradication therapy; type of treatment in the control group; whether or not secondary eradication therapy was performed; treatment period; method of assessing H. pylori infection; number of times successful eradication was confirmed; primary and secondary outcomes; description of changes in gastric mucosa following eradication therapy; mean age; whether or not a difference was observed between the interventional and control groups; number of subjects; gastric cancer occurrence; histological type of gastric cancer; site of gastric cancer; degree of gastric cancer progression; gastric histological features at study initiation; histological assessment using updated Sydney System scores before and after eradication therapy [33]; proportion of subjects showing either intestinal metaplasia or dysplasia; clearly defined rationale for specifying the follow-up period; follow-up period; H. pylori eradication rate; drop-out rate; mean time to cancer occurrence; and time between endoscopies. Atrophic gastritis, intestinal metaplasia, and dysplasia were defined as precancerous lesions. We also noted whether there were multiple articles for a single study. If differences arose regarding details of the extracted data, all authors continued discussions until a consensus was achieved.

Quality assessment for primary trials
The quality of primary trials was assessed as described by Jadad et al [34]. This method assesses whether the trial is randomized, the appropriateness of randomization if present, whether the trial is double-blinded, the appropriateness of double-blinding if present, and withdrawals/ dropouts, using a score of 0 or 1 for each item. Total scores thus range from 0 to 5. A high quality trial in this meta-analysis was defined by a Jadad score of !3 points. The GRADE system was used to evaluate the risk of bias of each trial used in this meta-analysis [35]. An RCT was considered high quality if three or more of the six domains for assessing risk of bias were adequate. Ultimately, the quality of a study was determined by either the Jadad score or risk of bias.

Statistical analysis
In four previous meta-analyses [6,[9][10][11], the pooled RR was used as the main index. Because this index tends to overestimate the effect, we first used the pooled RD and then the pooled RR, after which the results were presented together. The RD was calculated as risk in the interventional group minus that in the control group. The RD described the absolute change in risk that was attributable to the intervention. In other words, if the intervention had an identical effect to the control, RD would be 0. If it reduced the risk, the risk difference would be less than 0; if it increased the risk, the RD would be bigger than 0. The RD ranged from -1 to 1. A "+" sign indicated that treatment was favored, while a "-" sign indicated that the control was favored. Afterwards, the weighted pooled estimates were calculated for binary data. A fixedeffect model weighted by the Mantel-Haenszel (M-H) method was used to pool RD [36], followed by a test of homogeneity. Homogeneity among trials was assessed using the I 2 test [37]. If the hypothesis of homogeneity was rejected, a random-effect model using the DerSimonian-Laird method was employed [38]. The potential for publication bias was examined by the funnel plot method [39], and the statistical significance of differences was evaluated in accordance with the methods of Begg or Egger [40,41]. Given the observed risk difference, the number of patients that need to be treated (NNT) to prevent one adverse effect was also used as a measure of treatment effect; computationally speaking, NNT = 1/RD. Furthermore, the impact of eradication therapy, compared with placebo or no treatment, was expressed as a relative risk of occurrence of gastric cancer with 95% confidence intervals (CIs). All statistical analyses were performed with STATA statistical software version 14 [42]. Results are expressed as means and 95% CIs, unless indicated otherwise. P values < 0.05 were considered statistically significant. In addition, in order to evaluate visually whether the suppressive effect of eradication therapy on gastric cancer tends to change with time, we plotted the RD on the y-axis against the mean follow-up period for the interventional and control groups of each study on the x-axis, and fit a simple linear regression line based on the least squares method.

Search results
We screened 3383 studies through database searches and by reading through publications. Of these, we excluded 3270 studies and evaluated 113 in detail, and were ultimately left with seven studies (Fig 1) [43][44][45][46][47][48][49]. An RCT by Wong et al. compared four groups including a placebo group [48]. Of these, the two groups that met the inclusion criteria and were thus extracted for this study included one for which treatment intervention was a typical eradication therapy, and placebo administration for the other. Of the 106 studies excluded, we confirmed three article sets that involved multiple articles for a single study, in which the subjects and study background were identical but the study period differed. The studies by Zhou et al. [49], Mera et al. [45], and Ma et al. [47] were the long-term versions of those by Leung et al. [7], Correa et al. [50], and You et al. [51], respectively. Prior to their final report [49], Zhou et al. had also published a summarized version [52]. In accordance with the methodology described above, we excluded short-term studies with the same background and ultimately included their longterm counterparts (each published after the short-term study). Although the study by Miehke et al. [44] was not included in previous four meta-analyses [6,[9][10][11], we included it here as it met our inclusion criteria. None of the RCTs in the Japan Medical Abstracts Society or the three gastrointestinal medical journals satisfied the inclusion criteria. The article by Saito et al. was an abstract published as a poster at an academic conference (Digestive Disease Week, DDW 2005) [46]. Because the subsequent clinical study was not presented as an article, only the conference abstract was used. The research results were first published in Japanese; we did not find any article that later reported the same results in English.
In terms of outcome evaluation, gastric cancer occurrence had been designated as the primary outcome measure in three studies [43,44,49], while a secondary outcome measure was specified in four [45][46][47][48]. In an upper gastrointestinal endoscopic examination following eradication therapy, one study [45] found significant improvement of the gastric mucosa whereas three found no significant improvement [46][47][48]; the other three did not assess this parameter [43,44,49]. Effectiveness of Helicobacter pylori eradication to prevent gastric cancer  The total numbers of subjects for the interventional and control groups were 3337 and 3270, respectively, with a mean age of 51.0 years for both groups. Mean follow-up periods for the interventional and control groups were 7.8 and 6.7 years, respectively; the rationale for the follow-up period was not indicated in any study. Two studies [44,48] reported the time to cancer onset following eradication therapy, and indicated mean times of 3.9 and 2.3 years for the interventional and control groups, respectively.
Histological assessment, using the updated Sydney System score before and after eradication therapy, was not possible for any of the studies. Endoscopy was performed annually in one study [43] or following a fixed non-annual schedule in four studies [44,45,47,48]; the schedule was unspecified in two studies [46,49].
We sent e-mail inquiries to six of the eight corresponding authors [43][44][45][47][48][49]] whose addresses were listed, asking if multiple articles with redundant data had been published. Emails sent to the listed addresses for two of these six authors did not reach their destinations [45,49]. None of the remaining four authors replied within the first week; therefore, we sent a reminder to all of these authors [43,44,47,49]. Because we received no replies even after another week, we conducted no further follow-ups.

Study quality
We first evaluated the quality of the seven RCTs in Table 3. The median Jadad score was 3.0 (range: 1-5). The primary outcome measure was described in the methodology of every study. Sample size had been specified in advance in three studies [44,47,49]. The funding source was public in four studies [44,45,47,49] and unspecified in three [43,46,48]. None of the studies mentioned whether there were any conflicts of interest. In addition, although one study indicated that drugs were actually provided, six lacked such description [43][44][45][46][47][48].
Regarding the risk of bias for each RCT, three studies mentioned allocation concealment [43,44,47], all noted adequate sequence generation, and three mentioned blinding [43,44,49]. Only two studies noted inadequate outcome data [43,44], and free of elective outcome reporting was mentioned in all studies. None of the studies commented on whether they were free of other biases. Consequently, the quality of each study was moderate, as was the quality of the body of evidence in general.

Suppressive effect of H. pylori eradication therapy on primary gastric cancer
Overall, eradication therapy of H. pylori infection significantly reduced the risk on primary gastric cancer (pooled risk ratio [RR], 0.67; 95% Confidence Interval [CI], 0.48 to 0.95 with low heterogeneity (I 2 = 0%) in Table 4 and Fig 2. A subgroup analysis, the pooled RR was significantly shown for RCTs of high quality, those within Asia, those in which the control group was given a placebo or no treatment, those that targeted subjects with a mean age between 41 and 50 years, and even those in which the mean study period exceeded 10 years.
Overall, eradication therapy of H. pylori infection did not significantly reduce the risk on primary gastric cancer (pooled difference [RD], -0.00; 95% Confidence Interval [CI], -0.01 to 0.00 with low heterogeneity (I 2 = 33%) and NNT = 125.5 [95% CI: 70.0 to 800.9]). When the analysis was restricted to the five high-quality RCTs, pooled RD was -0.01 [95% CI: -0.02 to 0.00] (I 2 = 50%, NNT = 101.7 [95% CI: 57.4 to 604.0]). When analyzing the relationship between RR and follow-up period for the three studies with a mean follow-up period of at least 10 years [45,48,49], linear regression yielded a line with a positive slope (y (RR) = 0.4932x (year)-4.6888) (Fig 3). Besides, when analyzing the relationship between RD and follow-up period for the three studies with a mean follow-up period of at least 10 years [45,48,49], linear     (Fig 4). This suggests that although the risk of cancer onset is higher in the control group than in the intervention group up to a follow-up period of 11.5 to 12 years, the relationship might be reversed beyond this point.

Publication bias
Funnel-plot analysis, Begg's test, and Egger's test were performed to evaluate the potential for publication bias in terms of overall RD (Fig 5). The funnel-plot did not show an asymmetric pattern. Neither of the statistical tests revealed significant publication bias (p = 0.348, p = 0.610, respectively).

Discussion
In this study, we conducted a systematic review of whether H. pylori eradication therapy suppresses the occurrence of primary gastric cancer in patients whose gastric cancer was not diagnosed endoscopically. Three RCTs [43,45,48] were added to the meta-analysis by Fuccio et al. [6] in 2009, and one [43] to the meta-analyses by Ford et al. [9,10] in 2014 and 2015. A metaanalysis of the seven articles ultimately included revealed that, whereas the gastric cancer suppressing effect of eradication therapy was comparable to that of previous research in terms of RR, the effect size in terms of RD, which was not assessed in previous studies, was minor and not statistically significant. Overall NNT was 125, and subgroup analyses were approximately from 100 to 300. Showing these values, we think the effectiveness of eradication of Helicobacter pylori to prevent primary gastric cancer was relatively low. This trend was observed in all subgroup analyses in Table 4. In this study, the overall pooled RR for the suppression of gastric cancer by H. pylori eradication was 0.67 (95% CI: 0.48 to 0.95), which was comparable to that of previous studies. However, when expressed in terms of RD, the overall pooled RD was -0.00 (95% CI: -0.01 to 0.00), i.e., the effect was slight and not statistically significant. It is common that "Relative Risk obtained from the same data looks larger than Risk Difference". Apart from whether it is one case symbolically seen in the present research question and whether it is deliberate or not intentional, in the previous study both RCT and meta-analysis were only indicators of risk ratio. The magnitude of the effect such as risk ratio 0.6-0.7 was emphasized, and the tone of the discussion was to justify the aggressive intervention. Although we used the same data to analyze pooled risk ratio and risk difference, pooled risk ratio was statistically significant but pooled risk difference was not. When the event-incidence was rare with meta-analysis, in another study, pooled risk ratio was statistically significant but pooled risk difference was not like our study [53]. We thought that the association between pooled risk ratio and risk Effectiveness of Helicobacter pylori eradication to prevent gastric cancer difference showed discrepancy because we used data of rare events. In fact, Lane, et al showed that this discrepancy occurred when the methods of statistical tests in the meta-analysis were the differences and outcome incidence of the original data used in randomized controlled trials was rare. Main purpose of our study is to facilitate careful discussion on the expected magnitude of the effect of sterilization by H. pylori eradication to prevent primary gastric cancer comparing two risk indices that have been rarely mentioned before. We think that this phenomenon can be called a new outcome-reporting bias in risk communication. Many epidemiological findings, including those from clinical trials, are expressed in terms of RR; however, it has been repeatedly pointed out that compared with absolute risk, this index overestimates the association. [21]. For example, in a clinical trial by Lipid Research Clinics reported in 1982 [54], a 19% reduction in the risk of ischemic heart disease due to a cholesterol-reducing drug was emphasized. However, this was expressed in terms of relative risk reduction (RRR); the absolute risk reduction (ARR) was 1.6%. A similar report also emphasized an RRR of 31% over an ARR of 2.3% [55]. The RRR used in these studies, although not identical to relative risk, is a ratio-based index; this expression was likely used with the intent to strengthen reader impressions. Given these findings, the FDA proposed the following in a 2011 report when Effectiveness of Helicobacter pylori eradication to prevent gastric cancer communicating risk: "Provide absolute risks, not just relative risks. Patients are unduly influenced when risk information is presented using a relative risk approach; this can result in suboptimal decisions" [13]. The RR results presented herein do not deny the effectiveness of H. pylori eradication in preventing cancer. Nonetheless, the results for absolute risk (AR) suggest that we should be cautious regarding the effect size and its level of certainty as evidence. Typically, the problem with selective reporting has been outcome reporting bias, in which numerous outcomes are measured and only those variables attaining statistical significance are published [56][57][58]. Previous studies of H. pylori eradication presented their results only as a ratio index that emphasizes its effect could be pointed out as a new selective reporting issue distinct from prevailing problems.
NNT is a treatment effect index based on ARR. Fuccio et al. [6] or Lee et al. [11] did not show NNT in their report. Although Ford et al. [9] showed NNT separately by country, their interpretation of this in their report was that the treatment would reduce the occurrence of gastric cancer within Asia. Obtaining the NNT for each population is appropriate because this index is influenced by the AR of a disease in a target population. However, a stable NNT cannot be obtained from findings in a single-population study that lacks power. To make evidence-based decisions, it will be necessary to alleviate the large impact that RR has on readers by obtaining NNT from an integrated RD and interpreting the results within the article. To whom can we recommend H. pylori eradication therapy for suppressing the occurrence of gastric cancer? Eradication therapy is necessary for patients who have just undergone earlystage gastric cancer surgery [59][60][61][62] and is well-established for patients who have undergone endoscopic therapy for this type of cancer [5,63]. This study, which targeted asymptomatic patients infected with H. pylori, predicted the following: first, that the reduction in the risk of gastric cancer by H. pylori eradication therapy would be at least 30% in terms of RR, similar to that of previous findings, but would be slight and statistically insignificant in terms of RD; and second, that after eradication, a suppressive effect on gastric cancer would likely occur in the short term but diminish in the long term due to increased occurrence of cancer from aging (Fig 3). Therefore, those with severe gastric mucosal atrophy or precancerous lesions, which are relatively elderly high-risk groups, tend not to benefit from the treatment effects. In contrast, young individuals without atrophy might be expected to benefit from the effects of eradication therapy in preventing gastric cancer. In the long-term, endoscopy will likely be necessary because cancer begins to occur more frequently with age, even in the intervention group.
This study has several limitations. First, the mean study periods of the RCTs used were short (6-7 years). In a review which cited an article by Graham et al. and the fact that the occurrence of cancer had also been demonstrated in RCTs with long-term follow-up, Tan et al. [64] stated that cohort studies on the same topic should also be referred to in evaluating the gastric cancer suppressing effects of H. pylori eradication therapy. Considering that many of the RCTs used here were completed within a few years, a re-examination including a cohort study with a longer observation period would be meaningful. For observational studies, preregistration has not become as widespread as it has for RCTs; such studies should also be interpreted cautiously because publication bias is more severe than it is for interventional studies, and because cohort studies, given their high bias risk, may show results that diverge from true results. Second, the eradication effect could vary depending on histopathology, i.e., gastritis, atrophy, intestinal metaplasia, or dysplasia. However, these could not be clearly distinguished in the present study.
There is ABC classification for assessing gastric cancer risk and evaluating the degree of gastric mucosal atrophy objectively and quantitatively [65]. In our study, we couldn't compare the degree of gastric mucosal atrophy quantitatively. Among youths, a H. pylori carrier without gastric mucosal atrophy could possibly lower the risk of primary gastric cancer [66]. This problem could be addressed by future studies with the development of a standardized method for tissue evaluation.
From the present study, we conclude that the suppressive effect of H. pylori eradication therapy on the occurrence of primary gastric cancer was statistically significant and comparable to that of previous studies in terms of estimated RR. However, in terms of RD, the effect size was minor and not statistically significant. At this point, caution must be exercised when promoting evidence-based eradication measures.
Supporting information S1