Abstract

Background

Several observational studies have shown that statins may modify the risk of gastric cancer (GC). We carried out a systematic review and meta-analysis of studies evaluating the effect of statins on GC risk.

Patients and methods

We conducted a systematic search of multiple databases up to December 2012. Studies that evaluated exposure to statins, reported GC outcomes and odds ratio (OR) or provided data for their estimation were included in the meta-analysis. Pooled OR estimates with 95% confidence intervals (CIs) were calculated using the random-effects model.

Results

Eleven studies (eight observational, three post-hoc analyses of 26 clinical trials) reporting 5581 cases of GC were included. Meta-analysis showed a significant 32% reduction in GC risk with statin use (adjusted OR, 0.68; 95% CI, 0.51–0.91). After exclusion of one study which was contributing to considerable heterogeneity, a significant 16% reduction in GC risk was a more conservative, consistent estimate (adjusted OR, 0.84; 95% CI, 0.78–0.90). This chemopreventive association was present in both Asian (adjusted OR, 0.68; 95% CI, 0.53–0.87) and Western population (adjusted OR, 0.86; 95% CI, 0.79–0.93).

Conclusions

Meta-analysis of studies supports a protective association between statin use and GC risk, in both Asian and Western population, in a dose-dependent manner.

introduction

With an annual incidence of nearly 1 million, gastric cancer (GC) is the fourth most common cancer world wide and the second most frequent cause of cancer death with >0.7 million deaths annually [1]. Over 70% of new cases and deaths occur in developing countries, predominantly in Eastern Asia and Eastern Europe, which has been attributed to chronic Helicobacter pylori infection and dietary habits. Unfortunately, almost half of the patients present with advanced, inoperable disease with a 5-year survival rate of <5% [2]. Even in patients with resectable disease, prognosis is dismal with 5-year survival rates in the order of 25–35% [3, 4]. Early endoscopic detection of GC is feasible and could potentially improve outcomes, but is cost prohibitive [5]. Current strategies for improving outcomes in GC are aimed at reducing chronic H. pylori infection, screening patients at highest risk and identifying chemopreventive agents.

Statins, or 3-hydroxy-3-methylglutaryl coenzyme A (HMG-CoA) reductase inhibitors, used for primary and secondary prevention of cardiovascular diseases, decrease the risk of certain cancers [6, 7] and reduce cancer-related mortality [8]. In vitro and animal studies have shown that in addition to cholesterol reduction, statins have anti-proliferative, pro-apoptotic, anti-angiogenic and immunomodulatory effects, which prevent cancer development and growth [9]. This effect also been shown in human GC-derived cell lines [10]. Some recent observational studies have shown that statins may be associated with a lower risk of GC [11, 12], whereas others have shown no beneficial effect [13, 14].

To better understand this issue, we carried out a systematic review with meta-analysis of existing randomized, controlled trials (RCT) and observational studies that investigated the association between statins and the risk of developing GC.

methods

This systematic review was conducted following guidance provided by the Cochrane Handbook [15] and is reported according to the Meta-analysis of Observational Studies in Epidemiology guidelines [16]. The process followed a priori established protocol.

data sources and search strategy

First, a systematic literature search of PubMed (1966 through 1 December 2012), Embase (1988 through 1 December 2012) and Web of Science (1993 through 1 December 2012) databases was conducted for all relevant articles on the effect of statins use on the risk of GC. Search terms used included ‘HMG-CoA reductase inhibitor(s)’, ‘statin(s)’, ‘atorvastatin’, ‘fluvastatin’, ‘lovastatin’, ‘pravastatin’, ‘rosuvastatin’, ‘pitavastatin’ or ‘simvastatin’ combined with ‘cancer’ or ‘neoplasm(s)’. The results were exported to a common EndNote (reference manager) file. After this, duplicates were removed, and a total of 2370 unique studies were identified. Then, according to the protocol-defined study inclusion and exclusion criteria (see below) both authors independently reviewed the studies to see whether the inclusion criteria were met. Based on the review of the title and abstract, 2281 studies were excluded since they provided insufficient information on the risk of GC and statins. For the remaining 89 articles which were related to this topic, both authors reviewed the full text of the articles, independently, and reported the studies that each felt should be included or excluded. The coefficient of agreement between the two reviewers (κ = 0.86, 95% CI, 0.73–1.00) was excellent. Second, we searched the bibliographies of these selected articles as well as relevant narrative and systematic review articles to identify any additional studies. Third, we also searched conference proceedings of major oncology (American Society of Clinical Oncology annual meeting as well as the Gastrointestinal Research Forum; European Society of Medical Oncology annual meeting and World Congress on GI Cancer) and gastroenterology conferences (Digestive Diseases Week, American College of Gastroenterology annual meeting) from 2005 to 2012 for studies that had been published only in the abstract form. Figure 1 summarizes the study identification and selection process.

Table 1.

Characteristics of included studies assessing the risk of gastric cancer (GC) with statin use

Study Study design Location/setting Time Period Exposure ascertainment Outcome assessment All subjects
 
On statins
 
Not on Statins
 
Confounding variables adjusted fora 
GC Total GC Total GC Total 
Observational studies 
 Chiu et al. [11C–C Taiwan; Population-based 2005–2008 National Pharmacy Database Medical diagnostic codes 337 1685 56 354 281 1331 1,2,6,7,10, 13 
 Huakka et al. [28Nested C–C Finland; Population-based 1996–2005 National Pharmacy Database Record linkage, with Finnish Cancer Registry 1667 944 962 770 472 481 897 472 481 1,2,14 
 Kaye et al. [13C–C UK; Population-based 1990–2002 National Pharmacy Database Medical diagnostic codes 39 18 088 3244 35 14 844 1,2,4,5,10, 14 
 Graaf et al. [27Nested C–C Netherlands; Population-based 1985–2008 PHARMO Record Linkage Hospital discharge records 104 20 105 NR 1444 NR 18 661 1,2,8,9,10, 12,14 
 Vinogradova et al.[14Nested C–C UK; Population-based 1998–2008 National Pharmacy Database Medical diagnostic codes 1992 10 271 322 1685 1670 8586 1,2,4,5,8,1112,13,14 
 Lee et al. [12C–C South Korea; Hospital-based 1999–2008 Pharmacy dispensing records Medical diagnostic codes 983 1966 99 466 884 1500 1,2,8 
 Friedman et al.[26Cohort USA; Population-based 1994–2003 Pharmacy dispensing records Kaiser Permanente Cancer Registry 137 4 222 660 NR 361 849 NR 3 860 811 8,13 
 Marelli et al. [29Nested C–C USA; Population-based 1991–2009 Pharmacy dispensing records Electronic Medical Record review 31 91 714 13 45 857 18 45 857 1,2,3,4,5 
Randomized, controlled trials (RCTs) 
  Cholesterol Treatment Trialists' (CTT) [2522 RCTs (post-hocEurope/USA /Australia; Hospital-based – Individual drug dispension Adverse event reporting by investigators 192 134 537 92 67 258 100 67 279 Variable 
 Matshushita et al. [30Three clinical trials (individual. patient data) Japan; Hospital-based – Individual drug dispension Adverse event reporting by investigators 95 13 724 43 7375 52 6349 Variable 
 Sato et al. [31RCT (post-hocJapan; Hospital-based 1991–1995 Individual drug dispension – 263 179 84 1,2,5 
Study Study design Location/setting Time Period Exposure ascertainment Outcome assessment All subjects
 
On statins
 
Not on Statins
 
Confounding variables adjusted fora 
GC Total GC Total GC Total 
Observational studies 
 Chiu et al. [11C–C Taiwan; Population-based 2005–2008 National Pharmacy Database Medical diagnostic codes 337 1685 56 354 281 1331 1,2,6,7,10, 13 
 Huakka et al. [28Nested C–C Finland; Population-based 1996–2005 National Pharmacy Database Record linkage, with Finnish Cancer Registry 1667 944 962 770 472 481 897 472 481 1,2,14 
 Kaye et al. [13C–C UK; Population-based 1990–2002 National Pharmacy Database Medical diagnostic codes 39 18 088 3244 35 14 844 1,2,4,5,10, 14 
 Graaf et al. [27Nested C–C Netherlands; Population-based 1985–2008 PHARMO Record Linkage Hospital discharge records 104 20 105 NR 1444 NR 18 661 1,2,8,9,10, 12,14 
 Vinogradova et al.[14Nested C–C UK; Population-based 1998–2008 National Pharmacy Database Medical diagnostic codes 1992 10 271 322 1685 1670 8586 1,2,4,5,8,1112,13,14 
 Lee et al. [12C–C South Korea; Hospital-based 1999–2008 Pharmacy dispensing records Medical diagnostic codes 983 1966 99 466 884 1500 1,2,8 
 Friedman et al.[26Cohort USA; Population-based 1994–2003 Pharmacy dispensing records Kaiser Permanente Cancer Registry 137 4 222 660 NR 361 849 NR 3 860 811 8,13 
 Marelli et al. [29Nested C–C USA; Population-based 1991–2009 Pharmacy dispensing records Electronic Medical Record review 31 91 714 13 45 857 18 45 857 1,2,3,4,5 
Randomized, controlled trials (RCTs) 
  Cholesterol Treatment Trialists' (CTT) [2522 RCTs (post-hocEurope/USA /Australia; Hospital-based – Individual drug dispension Adverse event reporting by investigators 192 134 537 92 67 258 100 67 279 Variable 
 Matshushita et al. [30Three clinical trials (individual. patient data) Japan; Hospital-based – Individual drug dispension Adverse event reporting by investigators 95 13 724 43 7375 52 6349 Variable 
 Sato et al. [31RCT (post-hocJapan; Hospital-based 1991–1995 Individual drug dispension – 263 179 84 1,2,5 

a1, age, 2, sex, 3, race, 4, BMI, 5, smoking/alcohol, 6, H. pylori, 7, peptic ulcers, 8, other medications (aspirin/NSAIDs), 9, other lipid lowering agents, 10, healthcare utilization, 11, socioeconomic status, 12, comorbidities, 13, calender year, 14=region.

C–C, case–control; CTT, cholesterol treatment trialists’ Collaboration; RCT, randomized, controlled trials; NR, not reported.

study selection

Studies considered in this meta-analysis were either RCTs or observational studies that met the following inclusion criteria: (i) evaluated and clearly defined exposure to statins, (ii) reported GC incidence and (iii) reported relative risks (RR) (for cohort studies) or odds ratio (OR) (for case–control studies) or provided data for their calculation. Inclusion was not otherwise restricted by study size, language or publication type. Articles were excluded from the analyses if there were insufficient data for determining an estimate of RR/OR and a 95% confidence interval (CI), though these were included in the systematic review and described qualitatively. When there were multiple publications from the same population, only data from the most recent comprehensive report were included.

data abstraction and quality assessment

Data were independently abstracted onto a standardized form by both the authors. The following data were collected from each study: study design, time period of study/year of publication, country of the population studied, primary outcome reported, type, dose and duration of statin use, information source for exposure measurement, total number of persons in each group (exposed to statins versus not exposed), OR and 95% CIs with and without adjustment for potential confounders and potential confounders used for adjustment. Data on the following confounding risk factors for GC were extracted from each study: age, sex, race, cigarette smoking, body mass index (BMI), history of H. pylori, diet, other medication use [aspirin/non-steroidal anti-inflammatory drugs (NSAID)] and alcohol use. Conflicts in data abstraction were resolved by consensus, referring back to the original article.

The methodological quality of case–control and cohort studies was assessed by both the authors independently using the Newcastle–Ottawa scale, with very good agreement (κ = 0.77; 95% CI, 0.65–0.90) [17]. In this scale, observational studies were scored across three categories: selection (four questions) and comparability (two questions) of study groups, and ascertainment of the outcome of interest (three questions), with all questions with a score of one, except for comparability of study groups, where separate points were awarded for controlling age and/or sex (maximum two points). A score of ≥7 was suggestive of a high-quality study. The methodological quality of randomized trials was assessed using the Cochrane Collaboration's tool for assessing the risk of bias in randomized trials [15]. Any discrepancies were addressed by a joint re-evaluation of the original article.

outcomes assessed

The primary analysis focused on assessing the risk of GC among users of statins. A priori hypotheses to explain potential heterogeneity in the direction and magnitude of effect among different studies included location of study (Asian population versus Western population), study design (observational studies versus clinical trials) and study setting (hospital-based versus population-based).

In the three studies which provided data on relationship between dose/duration of statin use and risk reduction in GC [11, 12, 14], long duration of statin use was defined as: more than median cumulative defined daily dose of statins [11] or >2 years of statin use [12] or >4–6 years [14]. Likewise, short duration of statin use was defined as: less than median cumulative defined daily dose of statins [11] or 0.5–1.0 years of statin use [12] or 1–2 years [14].

data synthesis and analysis

We used the random-effects model described by DerSimonian and Laird to calculate summary OR and 95% CI [18]. We assessed heterogeneity between study-specific estimates using two methods [19]. First, the Cochran's Q statistical test for heterogeneity, which tests the null hypothesis that all studies in a meta-analysis have the same underlying magnitude of effect, was measured. Because this test is underpowered to detect moderate degrees of heterogeneity [20], a P value of < 0.10 was considered suggestive of significant heterogeneity. Second, to estimate what proportion of total variation across studies was due to heterogeneity rather than chance, I2 statistic was calculated. In this, a value of >50% was suggestive of considerable heterogeneity [21]. Once heterogeneity was noted, between-study sources of heterogeneity were investigated using subgroup analyses by stratifying original estimates according to study characteristics (as described above). In this analysis also, a P value for differences between subgroups of <0.10 was considered statistically significant (i.e. a value of P < 0.10 suggested that stratifying based on that particular study characteristic partly explained the heterogeneity observed in the analysis). The presence of publication bias was assessed quantitatively using Egger's regression test (publication bias considered present if P ≤ 0.10) [22] and qualitatively, by visual inspection of funnel plots of the logarithm of ORs versus their standard errors [23]. Additionally, using the Rosenthal's ‘fail-safe N’ method, we estimated the number of unpublished null studies (which fail to show an association between statins and risk of GC) needed to remove the significance from the findings of this meta-analysis [24]. All P-values were two-tailed. For all tests (except for heterogeneity and publication bias), a probability level of <0.05 was considered statistically significant.

Since outcomes were relatively rare, ORs were considered approximations of RR. We also calculated the number needed to treat with statins to prevent one case of GC in the general population, based on the current estimates of incidence of GC. All calculations and graphs were carried out using Comprehensive Meta-Analysis version 2 (Biostat, Englewood, NJ).

results

search results

Eleven studies met the defined inclusion criteria for this meta-analysis (seven case–control, one cohort, three post-hoc analysis of 26 clinical trials) [11–14, 25–31]. An abstract search from meeting proceedings did not yield any additional articles. These cumulatively reported 5581 cases of GC in 5 459 975 patients. Of the total, 962 192 patients were classified as statin users (16.7%). There were two Taiwanese studies from the same cohort [11, 32] and hence, only the most comprehensive report from these was included [11].

characteristics of included studies

The characteristics of these studies are shown in Table 1. The earliest study period began in 1985, and the latest ended in 2009. Seven studies were carried out in the Western population (four in Europe, two in United States and one was collaboration of multi-center trials across Europe, North America and Australia); [13, 14, 25–29] four studies were in the Asian population (two in Japan, one in South Korea, one in Taiwan) [11, 12, 30, 31]. The studies used both lipophilic and hydrophilic statins.

Chiu et al. carried out a case–control study on Taiwanese patients >50 years of age with new diagnosis of GC from 2005 to 2008 [11], while Huakka et al. [28], Graaf et al. [27] and Vinogradova et al. [14] carried out nationwide population-based nested case–control studies on statin exposure and the risk of various cancers, in Finland, the Netherlands and UK, respectively, through record linkage between the pharmacy dispensing records and the national cancer registries. Marelli et al. carried out a propensity-matched cohort analysis of the incidence of cancer in older adults who had or had not used statins, using the General Electric Centricity electronic medical records database of >11 million patients [29]. Friedman et al. used the pharmacy information management system and cancer registry of the Kaiser Permanente Medical Care Program of northern California, United States to estimate the risk of cancer in patients exposed to statins [26]. Lee et al. carried out a single-center, matched case–control, hospital-based study in patients with diabetes mellitus to assess the association between GC and statin use [12]. Matsushita et al. carried out an analysis of individual patient data from three large-scale clinical trials in patients with hyperlipidemia in Japan [30]. Emberson et al. carried out an individual patient data analysis of 22 RCTs of statins versus controls conducted by the Cholesterol Treatment Trialists’ (CTTs) collaboration to assess cancer incidence with statin exposure [25].

quality of included studies

Six observational studies were considered high-quality (Newcastle–Ottawa score ≥7) [11, 13, 14, 27–29]. Tables 2 and 3 depict the methodological quality of all studies. Though most studies adjusted or matched for demographic variables, they did not consistently account for other potential confounders: BMI (3 of 11), smoking and/or alcohol use (3 of 11), history of H. pylori (2 of 11) and other medication use (aspirin/ NSAIDs) (4 of 11). None of the studies adjusted for dietary risk factors for GC. For outcome ascertainment, most studies relied on medical diagnostic codes for GC, others reported record linkage through the cancer registry; both strategies had been validated to have high accuracy in the respective databases. In order to account for detection bias, three studies adjusted for the frequency of healthcare utilization in statin users and non-users [11, 13, 27]. In all included studies, a temporal relation of development of GC to statins was established by excluding cases of GC developing before exposure to statins, though there was variable duration of minimum statin use before new incident GC cases were included for analysis (0 months–2 years were reported). The overall quality of included RCTs was moderate to high [25, 30, 31]. However, all clinical trial studies were post-hoc analyses of RCTs carried out to assess the safety and efficacy of statins in the management of hyperlipidemia and/or coronary heart diseases.

Table 2.

Newcastle–Ottawa scale for assessment of quality of included studies—case–control studies (each asterisk represents if individual criterion within the subsection were fulfilled)

Quality assessment criteria Acceptable(*) Chiu et al. [11Huakka et al. [28Kaye et al. [13Graaf et al. [27Vinogradova et al. [14Lee et al. [12Marelli et al. [29
Selection 
 Is the case definition adequate? Yes, with independent validation – – – – – – 
 Representativeness of cases? Consecutive or obviously representative series of cases – 
 Selection of controls? Community controls – 
 Definition of controls? No history of gastric cancer (GC) 
Comparability 
 Study controls for age/gender Yes 
 Study controls for at least three additional factors Race, smoking, body mass index (BMI), history of Helicobacter pylori, diet, other medication use (aspirin/NSAIDs), alcohol use, healthcare utilization – – – 
Exposure 
 Ascertainment of exposure? Secure record, structured interview by a healthcare practitioner, blind to case–control status 
 Same method of ascertainment of cases/controls? Yes 
 Non-response rate? Same for both the groups – – 
 Overall quality score (maximum = 9)   
Quality assessment criteria Acceptable(*) Chiu et al. [11Huakka et al. [28Kaye et al. [13Graaf et al. [27Vinogradova et al. [14Lee et al. [12Marelli et al. [29
Selection 
 Is the case definition adequate? Yes, with independent validation – – – – – – 
 Representativeness of cases? Consecutive or obviously representative series of cases – 
 Selection of controls? Community controls – 
 Definition of controls? No history of gastric cancer (GC) 
Comparability 
 Study controls for age/gender Yes 
 Study controls for at least three additional factors Race, smoking, body mass index (BMI), history of Helicobacter pylori, diet, other medication use (aspirin/NSAIDs), alcohol use, healthcare utilization – – – 
Exposure 
 Ascertainment of exposure? Secure record, structured interview by a healthcare practitioner, blind to case–control status 
 Same method of ascertainment of cases/controls? Yes 
 Non-response rate? Same for both the groups – – 
 Overall quality score (maximum = 9)   

NSAIDs, Non-steroidal anti-inflammatory drugs.

Table 3.

Newcastle–Ottawa scale for assessment of quality of included studies—cohort studies (each asterisk represents if individual criterion within the subsection were fulfilled)

Quality assessment criteria Acceptable(*) Friedman et al. [26
Selection 
 Representativeness of exposed cohort? Representative of average adult in community (age/sex/being at risk of disease) 
 Selection of the non-exposed cohort? Drawn from same community as exposed cohort 
 Ascertainment of exposure? Secured records, structured interview 
 Demonstration that outcome of interest was not present at the start of the study? Only incident cases of gastric cancer (GC) 
Comparability 
 Study controls for age/sex? Yes – 
 Study controls for at least 3 additional risk factors? Race, smoking, body mass index (BMI), history of Helicobacter pylori, diet, other medication use (aspirin/NSAIDs),alcohol use, healthcare utilization – 
Outcome 
 Assessment of outcome? Independent blind assessment, record linkage 
 Was follow-up long enough for outcome to occur? Follow-up >3 years 
 Adequacy of follow-up of cohorts? Complete follow-up, or subjects lost to follow-up unlikely to introduce bias – 
 Overall quality score (maximum = 9)   
Quality assessment criteria Acceptable(*) Friedman et al. [26
Selection 
 Representativeness of exposed cohort? Representative of average adult in community (age/sex/being at risk of disease) 
 Selection of the non-exposed cohort? Drawn from same community as exposed cohort 
 Ascertainment of exposure? Secured records, structured interview 
 Demonstration that outcome of interest was not present at the start of the study? Only incident cases of gastric cancer (GC) 
Comparability 
 Study controls for age/sex? Yes – 
 Study controls for at least 3 additional risk factors? Race, smoking, body mass index (BMI), history of Helicobacter pylori, diet, other medication use (aspirin/NSAIDs),alcohol use, healthcare utilization – 
Outcome 
 Assessment of outcome? Independent blind assessment, record linkage 
 Was follow-up long enough for outcome to occur? Follow-up >3 years 
 Adequacy of follow-up of cohorts? Complete follow-up, or subjects lost to follow-up unlikely to introduce bias – 
 Overall quality score (maximum = 9)   

NSAIDs, non-steroidal anti-inflammatory drugs.

risk of gastric cancer

On meta-analysis of all studies assessing the risk of GC, the use of statins was associated with a statistically significant 30% reduction in GC incidence (unadjusted OR, 0.70; 95% CI, 0.51–0.97) (Figure 2). There was, however, considerable heterogeneity observed across studies (Cochran's Q-test, P < 0.01; I2 = 93%). This risk reduction with statins persisted even after adjusting for potential confounders (adjusted OR, 0.68; 95% CI, 0.51–0.91) (Figure 2), though the heterogeneity persisted (Cochran's Q-test, P < 0.01, I2 = 89%). For the remainder of the analysis, the maximally adjusted OR reported in each study was used.

Figure 1.

Flowsheet summarizing study identification and selection.

Figure 1.

Flowsheet summarizing study identification and selection.

Figure 2.

Summary of unadjusted and adjusted odds ratios (OR) assessing the risk of gastric cancer (GC) with statin exposure. No significant differences were observed in the summary estimates of adjusted and unadjusted OR (Pinteraction = 0.70), suggesting the absence of significant confounding, and confirming an independent relationship between statin use and the risk of GC. Significant heterogeneity was observed in the overall analysis (Cochran's Q P < 0.01; I2 = 89%). This was primarily attributed to a single study by Lee et al., which was positive outlier. On exclusion of this study, the summary adjusted and unadjusted ORs were 0.84 (0.78–0.90) and 0.88 (0.83–0.95), respectively. After excluding this study, the previously observed heterogeneity resolved (Cochran's Q-test P = 0.76, I2 = 0%).

Figure 2.

Summary of unadjusted and adjusted odds ratios (OR) assessing the risk of gastric cancer (GC) with statin exposure. No significant differences were observed in the summary estimates of adjusted and unadjusted OR (Pinteraction = 0.70), suggesting the absence of significant confounding, and confirming an independent relationship between statin use and the risk of GC. Significant heterogeneity was observed in the overall analysis (Cochran's Q P < 0.01; I2 = 89%). This was primarily attributed to a single study by Lee et al., which was positive outlier. On exclusion of this study, the summary adjusted and unadjusted ORs were 0.84 (0.78–0.90) and 0.88 (0.83–0.95), respectively. After excluding this study, the previously observed heterogeneity resolved (Cochran's Q-test P = 0.76, I2 = 0%).

On restricting analysis to high-quality observational studies [11, 13, 14, 27–29], statin use continued to be associated with reduced risk of GC (n = 6 studies; adjusted OR, 0.83; 95% CI, 0.76–0.90). Moreover, the results were consistent across studies with minimal heterogeneity (Cochran's Q-test P = 0.71; I2 = 0%).

subgroup analysis

We carried out pre-planned stratified analyses of studies based on study design, location and setting (Table 4). None of these stratifications could account for the heterogeneity observed in the overall analysis. Statin use was associated with a reduced risk of GC in observational studies, and a non-significant reduction in risk was in post-hoc analyses of RCTs. Statin use was associated with a reduced risk of GC in a subset of studies carried out in the Western population; there was a similar trend in the Asian studies, though this was not statistically significant and considerably heterogeneous.

Table 4.

Subgroup analysis of all studies

Grouping variable Subgroups No. of studies Unadjusted OR 95% CI Adjusted OR 95% CI Heterogeneity between groups (Pinteraction)a 
All studies 
Study design Observational 0.66 0.45–0.99 0.65 0.45–0.93 0.25 
RCT (post-hoc analyses) 0.85 0.65–1.07 0.83 0.66–1.05 
Study location Western 0.90 0.84–0.97 0.86 0.79–0.93 0.18 
Asian 0.51 0.21–1.25 0.50 0.23–1.09 
Study setting Population-based 0.89 0.82–0.96 0.84 0.78–0.91 0.38 
Hospital-based 0.56 0.20–1.55 0.56 0.22–1.39 
After excluding Lee et al. [12
Study design Observational 0.89 0.82–0.96 0.84 0.78–0.91 0.91 
RCT (post-hoc analyses) 0.85 0.65–1.07 0.83 0.66–1.05 
Study location Western 0.90 0.84–0.97 0.86 0.79–0.93 0.09 
Asian 0.71 0.56–0.91 0.68 0.53–0.87 
Study setting Population-based 0.89 0.82–0.96 0.85 0.79–0.91 0.85 
Hospital-based 0.85 0.65–1.07 0.83 0.66–1.05 
Grouping variable Subgroups No. of studies Unadjusted OR 95% CI Adjusted OR 95% CI Heterogeneity between groups (Pinteraction)a 
All studies 
Study design Observational 0.66 0.45–0.99 0.65 0.45–0.93 0.25 
RCT (post-hoc analyses) 0.85 0.65–1.07 0.83 0.66–1.05 
Study location Western 0.90 0.84–0.97 0.86 0.79–0.93 0.18 
Asian 0.51 0.21–1.25 0.50 0.23–1.09 
Study setting Population-based 0.89 0.82–0.96 0.84 0.78–0.91 0.38 
Hospital-based 0.56 0.20–1.55 0.56 0.22–1.39 
After excluding Lee et al. [12
Study design Observational 0.89 0.82–0.96 0.84 0.78–0.91 0.91 
RCT (post-hoc analyses) 0.85 0.65–1.07 0.83 0.66–1.05 
Study location Western 0.90 0.84–0.97 0.86 0.79–0.93 0.09 
Asian 0.71 0.56–0.91 0.68 0.53–0.87 
Study setting Population-based 0.89 0.82–0.96 0.85 0.79–0.91 0.85 
Hospital-based 0.85 0.65–1.07 0.83 0.66–1.05 

afor adjusted OR.

No, number; OR, odds ratio; RCTs, randomized, controlled trials.

A trend toward a dose–duration response was seen on pooled analysis of three studies which provided sufficient data for this analysis [11, 12, 14]. A ‘long’ duration of statin was significantly more likely to have chemopreventive effect (adjusted OR, 0.35; 95% CI, 0.16–0.76) than ‘short’ duration of statin use (adjusted OR, 0.73; 95% CI, 0.51–1.05). Sufficient information was not available to carry out stratified analysis based on the age, gender, location of GC (cardia or non-cardia) or on the type of statins (lipophilic versus hydrophilic).

sensitivity analysis

To assess whether any one study had a dominant effect on the meta-analytic OR, each study was excluded and its effect on the main summary estimate and Cochran's Q-test P value for heterogeneity was evaluated. While no study significantly affected the summary estimate, exclusion of the study by Lee et al. resulted in resolution of the previously observed marked heterogeneity in the analysis [12]. The favorable and strong effect sizes observed in this single study were causing heterogeneity in the strength, but not the direction, of overall association. On analysis after excluding this study, the summary OR for both unadjusted (OR, 0.88; 95% CI, 0.83–0.95) and adjusted analysis (OR, 0.84; 95% CI, 0.78–0.90) remained significant and no heterogeneity was observed in the analysis (for unadjusted OR, Cochran's Q-test, P = 0.56, I2 = 0%; for adjusted OR, Cochran's Q-test, P = 0.74, I2 = 0%). Subgroup analyses were repeated after excluding this study (Table 4). The previously statistically non-significant association between statin use and GC risk in the Asian population became statistically significant.

On replacing one Taiwanese study [11] with another study from Taiwan from the same cohort [32], the overall results (adjusted OR, 0.64; 95% CI, 0.47–0.88), as well as in the subgroup of Asian studies (adjusted OR, 0.38; 95% CI, 0.17–0.85), were unchanged.

publication bias

There was no evidence of significant publication bias, both quantitatively (Egger's regression test, P = 0.54) and on visual inspection of the funnel plot (data not shown). Furthermore, we estimate that 136 unpublished null studies (which fail to show a chemopreventive association between statins and risk of GC) would be needed to remove the significance from the findings of this meta-analysis.

number needed to treat

Due to significant heterogeneity between studies and varying incidence of GC across different regions of the world, a single summary estimate for the number needed to treat with statins to prevent one case of GC could not be inferred. Using an age-adjusted incidence rate of GC in men in East Asia of 42.4 per 100 000 person years and a 32% reduction in GC risk with statin use, 7372 East Asian men would need to be treated with statins to prevent one case of GC per year [1]. Similarly, using an age-adjusted incidence rate of GC of 9.0 per 100 000 person years in Western Europe and a 14% reduction in GC risk with statin use, 79,371 men would need to be treated to prevent one case of GC [1].

discussion

With the high incidence of GC and very poor prognosis associated with the diagnosis, identification of potential chemopreventive agents is highly desirable. While aspirin and NSAIDs [33, 34] and H. pylori eradication therapy may be associated with reduced risk of GC [35], they are not without significant side-effects. In this comprehensive meta-analysis of all existing studies in almost 5.5 million patients with 5581 cases of GC, we found that statin use is associated with a 32% reduction in the risk of GC, after adjusting for multiple confounding factors. Given considerable heterogeneity attributed to one study, a 16% risk reduction with statin use may be a more conservative and consistent estimate. Such a chemopreventive effect of statins has also been observed for some inflammation-associated cancers (including hepatocellular cancer [6], esophageal cancer [36]), but not for others (such as sporadic colorectal cancer [37], breast cancer [38] and prostate cancer [38]) This is comparable with 22% reduction in the risk of GC seen with aspirin/NSAID use (OR, 0.78; 95% CI, 0.69–0.87) seen in a previous meta-analysis of observational studies [34]. Importantly, the latter may confounded by indication, since patients with gastric ulcers (a group at high risk for GC) may be less likely to be prescribed aspirin/NSAIDs, thereby overestimating the apparent chemopreventive effect of these medications.

The strengths of this analysis include the comprehensive and systematic literature search of both observational studies and RCTs, consistency of association between statins and GC, ability to evaluate the potential influence of measured confounders on the summary estimates and evaluation of duration–response effect. The likelihood of important selection or publication bias in our meta-analysis is small. During the identification and selection process, we did not exclude any article because of methodological characteristics. Our results are consistent with a previous small meta-analysis. Kouppala et al. included only two cohort studies in their analysis and observed a significant chemopreventive effect of statin use (OR, 0.59; 95% CI 0.40–0.88) [38]. In our analysis, we added numerous additional studies, including data from RCTs, to bring the evidence base up to date, and provide a more robust pooled estimate on the effect of statins on GC risk. With the larger number of studies, we were able to carry out multiple subgroup analyses, evaluate heterogeneity and the presence of publication bias.

anti-neoplastic effect of statins

GC has two histological subtypes that have distinct carcinogenic pathways [39]. Intestinal or well-differentiated tumors progress through atrophic gastritis followed by intestinal metaplasia, dysplasia and carcinoma, whereas diffuse or poorly differentiated carcinomas are not characterized by preceding steps other than the chronic gastritis associated with H. pylori infection. Among the genetic changes implicated in the multi-stage gastric carcinogenesis, amplification of c-myc and mutations of K-ras have been commonly reported [40]. In mice models, atorvastatin has been shown to block MYC phosphorylation and activation, suppressing tumor initiation and growth through a HMG-CoA reductase-dependent pathway [41]. The anti-neoplastic effects of statins have also been demonstrated in human GC cell lines through reduced cell division and increased apoptosis. Using human GC cell line HGT-1, Follet et al. have shown that lovastatin strongly suppresses genes involved in cell division on whole transcriptome microarrays [10]. Lovastatin also induced upregulation of cell-cycle inhibitor p21 and suppression of anti-apoptotic survivin and Mcl-1 proteins in these GC cell lines.

differences in observational studies and clinical trials

The chemopreventive effect of statins was seen primarily in observational studies which accounted for a large majority of the included GC cases (5290 cases, 94.8%). RCTs included in the study did not demonstrate any significant chemopreventive effect of statins though there is a trend toward statistical significance (adjusted OR, 0.83; 95% CI, 0.66–1.05). Importantly, the RCTs included in the meta-analysis represented post-hoc analyses of 26 clinical trials carried out on the effect of statins on cardiovascular morbidity [25, 30, 31]. By design, the patients enrolled in these RCTs were not at high risk of development of GC. Also, these studies were not adequately powered to detect a significant difference in GC incidence. Moreover, since the occurrence of cancer was not the primary objective of these trials, patients were not routinely screened for development of GC; this might have affected the detection rate of GC. The follow-up duration in these RCTs was short. These factors may explain why current clinical trials of statins do not demonstrate a statistically significant chemopreventive effect of statins against GC.

On the other hand, the chemopreventive effect of statins seen in observational studies may also over-estimate its true effect. Observational studies lack the experimental random allocation of the intervention necessary to test exposure-outcome hypotheses optimally. Despite adjusting for numerous covariates, it is not possible to eliminate the potential of residual confounding. It is possible that the observed decreased risk of GC seen in statin users may relate to a ‘healthy user’ bias [42]. Patients taking statins may be more likely to be prescribed preventive medications and/or be more compliant with preventive health measures, when compared with patients not taking statins. The latter poorly compliant patients may have other unhealthy lifestyle practices predisposing them to GC. In our analysis, on restricting analysis to studies which adjusted for frequency of healthcare utilization, statins continued to show a stable and consistent chemopreventive effect against GC (adjusted OR, 0.68; 95% CI, 0.52–0.90) [11, 13, 27]. Hence, given the strength and consistency of association along with in vitro data that suggest biological plausibility of preventive effects of statins on GC, ‘healthy user’ effect is unlikely to be the primary driver of results. Observational studies are inherently not able to definitely establish when the intervention may exert its influence, the minimum effective dose–duration and frequency of medication intake that is required to achieve a benefit. Although we found a potential duration benefit, differing exposure definitions between studies make recommendations regarding dosing regimens difficult.

limitations

First, the included studies did not provide data based on the location or histological subset of GC, which are associated with different pathogenesis and prognosis. Second, all studies did not adjust for the same confounders. Importantly, studies failed to adjust for H. pylori infection which is strongly associated with GC risk. However, though biologically plausible, multiple RCTs and meta-analyses have failed to show a clinically significant reduction in GC risk with H. pylori eradication [35, 43–45]. Moreover, in our analysis, there was no significant difference in the pooled analysis of unadjusted and maximally adjusted data from each study, suggesting that any difference attributable to using different confounders for adjustment is likely small. Another potential limitation that particularly applies to case–control studies evaluating GC is recall bias. However, in most studies pharmacy drug prescriptions information was used, and hence, the effects of this are likely minimal. The significant heterogeneity observed in the meta-analysis could be explained by excluding a single study. In their single-center study of a diabetic cohort in high-risk South Korean population, Lee et al. observed a marked 70%–80% reduction in GC risk [12]. This marked decrease in risk may be related to detection bias in this referral center study as well as potential confounding by concomitant use of other putative chemopreventive effect of aspirin and metformin [46]. Finally, we did not contact authors of RCTs to solicit unpublished data, and this may have resulted in reporting bias. However, we estimated that 136 unpublished null studies (which fail to show a chemopreventive association between statins and risk of GC) would be needed to remove the significance from the findings of this meta-analysis.

conclusion

Based on this comprehensive meta-analysis, statins appear to have chemopreventive effects against GC in both Asian and Western population. However, since the observed magnitude of GC risk reduction associated with statins is relatively modest, especially in the Western population, the number needed to treat to prevent one case of GC would be large. A definitive, randomized chemoprevention trial is needed to more rigorously assess the effects of statins on incident GC, but would be lengthy, logistically challenging and resource intensive. To facilitate further clarification of statins’ GC chemopreventive potential, clinical evaluation in an enriched patient population (i.e. East Asian men with history of gastric atrophy or ulcers) may be the first best step. Meanwhile, in patients with borderline eligibility for statin therapy, other risk factors for GC may be informative with respect to balancing risks versus benefits for clinical decision-making.

disclosure

The authors have declared no conflicts of interest.

references

1
Jemal
A
Bray
F
Center
MM
, et al.  . 
Global cancer statistics
CA Cancer J Clin
 , 
2011
, vol. 
61
 (pg. 
69
-
90
)
2
Lello
E
Furnes
B
Edna
TH
Short and long-term survival from gastric cancer. A population-based study from a county hospital during 25 years
Acta Oncol
 , 
2007
, vol. 
46
 (pg. 
308
-
315
)
3
Cunningham
D
Allum
WH
Stenning
SP
, et al.  . 
Perioperative chemotherapy versus surgery alone for resectable gastroesophageal cancer
N Engl J Med
 , 
2006
, vol. 
355
 (pg. 
11
-
20
)
4
Cunningham
SC
Kamangar
F
Kim
MP
, et al.  . 
Survival after gastric adenocarcinoma resection: eighteen-year experience at a single institution
J Gastrointest Surg
 , 
2005
, vol. 
9
 (pg. 
718
-
725
)
5
Vakil
N
Talley
N
van Zanten
SV
, et al.  . 
Cost of detecting malignant lesions by endoscopy in 2741 primary care dyspeptic patients without alarm symptoms
Clin Gastroenterol Hepatol
 , 
2009
, vol. 
7
 (pg. 
756
-
761
)
6
Singh
S
Singh
PP
Singh
AG
, et al.  . 
Statins are associated with a reduced risk of hepatocellular cancer: a Systematic review and meta-analysis
Gastroenterology
 , 
2012
, vol. 
144
 (pg. 
323
-
332
)
7
Bansal
D
Undela
K
D'Cruz
S
, et al.  . 
Statin use and risk of prostate cancer: a meta-analysis of observational studies
PLoS One
 , 
2012
, vol. 
7
 pg. 
e46691
 
8
Nielsen
SF
Nordestgaard
BG
Bojesen
SE
Statin use and reduced cancer-related mortality
N Engl J Med
 , 
2012
, vol. 
367
 (pg. 
1792
-
1802
)
9
Demierre
MF
Higgins
PD
Gruber
SB
, et al.  . 
Statins and cancer prevention
Nat Rev Cancer
 , 
2005
, vol. 
5
 (pg. 
930
-
942
)
10
Follet
J
Corcos
L
Baffet
G
, et al.  . 
The association of statins and taxanes: an efficient combination trigger of cancer cell apoptosis
Br J Cancer
 , 
2012
, vol. 
106
 (pg. 
685
-
692
)
11
Chiu
HF
Ho
SC
Chang
CC
, et al.  . 
Statins are associated with a reduced risk of gastric cancer: a population-based case–control study
Am J Gastroenterol
 , 
2011
, vol. 
106
 (pg. 
2098
-
2103
)
12
Lee
J
Lee
SH
Hur
KY
, et al.  . 
Statins and the risk of gastric cancer in diabetes patients
BMC Cancer
 , 
2012
, vol. 
12
 pg. 
596
 
13
Kaye
JA
Jick
H
Statin use and cancer risk in the General Practice Research Database
Br J Cancer
 , 
2004
, vol. 
90
 (pg. 
635
-
637
)
14
Vinogradova
Y
Coupland
C
Hippisley-Cox
J
Exposure to statins and risk of common cancers: a series of nested case–control studies
BMC Cancer
 , 
2011
, vol. 
11
 pg. 
409
 
15
Higgins
JPT
Green
S
 
Cochrane Handbook for Systematic Reviews of Interventions, Version 5.1.0 (updated March 2011). The Cochrane Collaboration 2012. Available at www.cochrane-handbook.org (4 December 2012, date last accessed)
16
Stroup
DF
Berlin
JA
Morton
SC
, et al.  . 
Meta-analysis of observational studies in epidemiology: a proposal for reporting. Meta-analysis of Observational Studies in Epidemiology (MOOSE) group
JAMA
 , 
2000
, vol. 
283
 (pg. 
2008
-
2012
)
17
Wells GA
SB
O'Connell
D
Peterson
J
, et al.  . 
The Newcastle–Ottawa Scale (NOS) for assessing the quality of nonrandomised studies in meta-analyses
 
Available at http://www.ohri.ca/programs/clinical_epidemiology/oxford.asp (10 December 2012, date last accessed)
18
DerSimonian
R
Laird
N
Meta-analysis in clinical trials
Control Clin Trials
 , 
1986
, vol. 
7
 (pg. 
177
-
188
)
19
Higgins
JP
Thompson
SG
Deeks
JJ
, et al.  . 
Measuring inconsistency in meta-analyses
BMJ
 , 
2003
, vol. 
327
 (pg. 
557
-
560
)
20
Thompson
SG
Pocock
SJ
Can meta-analyses be trusted?
Lancet
 , 
1991
, vol. 
338
 (pg. 
1127
-
1130
)
21
Guyatt
GH
Oxman
AD
Kunz
R
, et al.  . 
GRADE guidelines: 7. Rating the quality of evidence—inconsistency
J Clin Epidemiol
 , 
2011
, vol. 
64
 (pg. 
1294
-
1302
)
22
Egger
M
Davey Smith
G
Schneider
M
, et al.  . 
Bias in meta-analysis detected by a simple, graphical test
BMJ
 , 
1997
, vol. 
315
 (pg. 
629
-
634
)
23
Easterbrook
PJ
Berlin
JA
Gopalan
R
, et al.  . 
Publication bias in clinical research
Lancet
 , 
1991
, vol. 
337
 (pg. 
867
-
872
)
24
Rosenthal
R
The ‘file drawer problem’ and tolerance for null results
Psychol Bull
 , 
1979
, vol. 
86
 (pg. 
638
-
641
)
25
Emberson
JR
Kearney
PM
Blackwell
L
, et al.  . 
Lack of effect of lowering LDL cholesterol on cancer: meta-analysis of individual data from 175,000 people in 27 randomised trials of statin therapy
PLoS One
 , 
2012
, vol. 
7
 pg. 
e29849
 
26
Friedman
GD
Flick
ED
Udaltsova
N
, et al.  . 
Screening statins for possible carcinogenic risk: up to 9 years of follow-up of 361,859 recipients
Pharmacoepidemiol Drug Saf
 , 
2008
, vol. 
17
 (pg. 
27
-
36
)
27
Graaf
MR
Beiderbeck
AB
Egberts
AC
, et al.  . 
The risk of cancer in users of statins
J Clin Oncol
 , 
2004
, vol. 
22
 (pg. 
2388
-
2394
)
28
Haukka
J
Sankila
R
Klaukka
T
, et al.  . 
Incidence of cancer and statin usage—record linkage study
Int J Cancer
 , 
2010
, vol. 
126
 (pg. 
279
-
284
)
29
Marelli
C
Gunnarsson
C
Ross
S
, et al.  . 
Statins and risk of cancer: a retrospective cohort analysis of 45,857 matched pairs from an electronic medical records database of 11 million adult Americans
J Am Coll Cardiol
 , 
2011
, vol. 
58
 (pg. 
530
-
537
)
30
Matsushita
Y
Sugihara
M
Kaburagi
J
, et al.  . 
Pravastatin use and cancer risk: a meta-analysis of individual patient data from long-term prospective controlled trials in Japan
Pharmacoepidemiol Drug Saf
 , 
2010
, vol. 
19
 (pg. 
196
-
202
)
31
Sato
S
Ajiki
W
Kobayashi
T
, et al.  . 
Pravastatin use and the five-year incidence of cancer in coronary heart disease patients: from the prevention of coronary sclerosis study
J Epidemiol
 , 
2006
, vol. 
16
 (pg. 
201
-
206
)
32
Leung
HW
Chan
AL
Lo
D
, et al.  . 
Common cancer risk and statins: a population-based case–control study in a Chinese population
Expert Opin Drug Saf
 , 
2013
, vol. 
12
 (pg. 
19
-
27
)
33
Abnet
CC
Freedman
ND
Kamangar
F
, et al.  . 
Non-steroidal anti-inflammatory drugs and risk of gastric and oesophageal adenocarcinomas: results from a cohort study and a meta-analysis
Br J Cancer
 , 
2009
, vol. 
100
 (pg. 
551
-
557
)
34
Wang
WH
Huang
JQ
Zheng
GF
, et al.  . 
Non-steroidal anti-inflammatory drug use and the risk of gastric cancer: a systematic review and meta-analysis
J Natl Cancer Inst
 , 
2003
, vol. 
95
 (pg. 
1784
-
1791
)
35
Wang
C
Yuan
Y
Hunt
RH
The association between Helicobacter pylori infection and early gastric cancer: a meta-analysis
Am J Gastroenterol
 , 
2007
, vol. 
102
 (pg. 
1789
-
1798
)
36
Singh
S
Singh
AG
Singh
PP
, et al.  . 
Statins are associated with reduced risk of esophageal cancer, particularly in patients with Barrett's esophagus: a systematic review and meta-analysis
Clin Gastroenterol Hepatol
 , 
2013
 
January 26 [epub ahead of print], doi:10.1016/j.cgh.2012.12.036
37
Bonovas
S
Filioussi
K
Flordellis
CS
, et al.  . 
Statins and the risk of colorectal cancer: a meta-analysis of 18 studies involving more than 1.5 million patients
J Clin Oncol
 , 
2007
, vol. 
25
 (pg. 
3462
-
3468
)
38
Kuoppala
J
Lamminpaa
A
Pukkala
E
Statins and cancer: a systematic review and meta-analysis
Eur J Cancer
 , 
2008
, vol. 
44
 (pg. 
2122
-
2132
)
39
Tahara
E
Genetic pathways of two types of gastric cancer
IARC Sci Publ
 , 
2004
, vol. 
157
 (pg. 
327
-
349
)
40
Panani
AD
Cytogenetic and molecular aspects of gastric cancer: clinical implications
Cancer Lett
 , 
2008
, vol. 
266
 (pg. 
99
-
115
)
41
Cao
Z
Fan-Minogue
H
Bellovin
DI
, et al.  . 
MYC phosphorylation, activation, and tumorigenic potential in hepatocellular carcinoma are regulated by HMG-CoA reductase
Cancer Res
 , 
2011
, vol. 
71
 (pg. 
2286
-
2297
)
42
Patrick
AR
Shrank
WH
Glynn
RJ
, et al.  . 
The association between statin use and outcomes potentially attributable to an unhealthy lifestyle in older adults
Value Health
 , 
2011
, vol. 
14
 (pg. 
513
-
520
)
43
Lee
YC
Chen
TH
Chiu
HM
, et al.  . 
The benefit of mass eradication of Helicobacter pylori infection: a community-based study of gastric cancer prevention
Gut
 , 
2013
, vol. 
62
 (pg. 
676
-
682
)
44
Ford
AC
Moayyedi
P
Redundant data in the meta-analysis on Helicobacter pylori eradication
Ann Intern Med
 , 
2009
, vol. 
151
 pg. 
513
  
author reply 513–514
45
Fuccio
L
Zagari
RM
Eusebi
LH
, et al.  . 
Meta-analysis: can Helicobacter pylori eradication treatment reduce the risk for gastric cancer?
Ann Intern Med
 , 
2009
, vol. 
151
 (pg. 
121
-
128
)
46
Kato
K
Gong
J
Iwama
H
, et al.  . 
The antidiabetic drug metformin inhibits gastric cancer cell proliferation in vitro and in vivo
Mol Cancer Ther
 , 
2012
, vol. 
11
 (pg. 
549
-
560
)