Sage Journals: Discover world-class research

Abstract

Introduction:

The use of proton-pump inhibitors (PPI) may be associated with an increased risk of gastric cancer (GC).

Objective:

To review and meta-analyse available literature investigating the association between PPI use and GC risk.

Methods:

Two independent reviewers systematically searched Ovid MEDLINE, EMBASE, and Cochrane Library (inception to July 2020) for case-control and cohort studies assessing the association between PPI use and GC according to a predefined protocol in PROSPERO (CRD42018102536). Reviewers independently assessed study quality, extracted data, and meta-analysed available and newly calculated odds ratios (ORs) using a random-effects model, and stratified for GC site (cardia versus non-cardia) and PPI duration (<1 year, 1–3 years, >3 years).

Results:

We screened 2,396 records and included five retrospective cohort and eight case-control studies comprising 1,662,881 individuals in our meta-analysis. In random-effect models, we found an increased GC risk in PPI users [OR: 1.94, 95% confidence interval (95% CI): 1.47–2.56] with high statistical heterogeneity (I² = 82%) and overall moderate risk of bias. Stratified analyses indicated a significant risk increase in non-cardia (OR: 2.20, 95% CI: 1.44–3.36, I² = 77%) with a similar non-significant trend in cardia regions (OR: 1.77, 95% CI: 0.72–4.36, I² = 66%). There was no GC increase with longer durations of PPI exposure (<1 year: OR: 2.29, 95% CI: 2.13–2.47, I² = 0%; 1–3 years: OR: 1.46, 95% CI: 0.53–4.01, I² = 35%; >3 years: OR: 2.08, 95% CI: 0.56–7.77, I² = 61%).

Conclusion:

We found a twofold increased GC risk among PPI users, but this association does not confirm causation and studies are highly heterogeneous. PPI should only be prescribed when strictly indicated.

Keywords

stomach neoplasms proton-pump inhibitors meta-analysis systematic review

Introduction

Gastric cancer (GC) is one of the most frequent cancers worldwide with more than 1 million new patients per year, constituting 5.7% of all cancer cases worldwide.¹ Mortality of GC remains high with >780,000 deaths annually, corresponding to 8.2% of all cancer-related deaths.¹ Common GC risk factors are manifold and include environmental factors such as Helicobacter pylori (HP) infections, dietary habits, smoking and obesity, precursor lesions such as atrophic gastritis and intestinal metaplasia, and host-related factors.^2–5 Nevertheless, there is a decreasing trend in lifetime GC incidence possibly related to a lower burden of HP infection.⁶

Proton-pump inhibitors (PPI) are one of the most frequently prescribed drugs worldwide⁷ and considered relatively safe. However, previous meta-analyses^8–13 of mostly large retrospective cohort and case-control studies found a significant association between PPI use and an up to 2.5-fold increased GC risk across very heterogeneous study populations and PPI indications. Despite all potential limitations of retrospective studies, there is an urgent need for further investigation. Due to new evidence from Sweden,¹⁴ Taiwan,¹⁵ the United States,¹⁶ and the United Kingdom¹⁷ since the last meta-analyses, we performed an updated systematic literature review and meta-analysis.

Materials and methods

Data sources and search strategy

We conducted this systematic literature review according to a predefined protocol in PROSPERO, an international prospective register of systematic reviews (CRD42018102536) and adhered to the Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA statement).¹⁸

We developed search strategies for EMBASE, Ovid MedLine, and the Cochrane Library (Supplemental Appendix Figure 1), searched all databases from inception until 18 July 2020, and screened bibliographies of included records for additional publications.

Study selection, quality assessment, and data extraction

We screened titles and abstracts according to predefined eligibility criteria. We included cohort and case-control studies in any language assessing the association between PPI use and GC risk comprising individuals with a considerable PPI use and a control group with no or only rare PPI use. Moreover, we extracted burden of PPI exposure by any measure of duration, frequency, and dosage.

The primary outcome of our study was the risk of incident GC at any location. Whenever possible, we extracted adjusted odds ratio (OR) and respective 95% confidence interval (CI), as adjusted OR was the most frequent effect size in our included studies. In case only other point estimates, such as HR, standardized incidence ratios (SIRs), and RR were available, we extracted all necessary raw data to calculate the unadjusted OR. In case of missing or incomplete information, we contacted authors for additional information on the exposure and outcome definitions, distribution of exposure, GC cases, and controls. Only one potentially eligible study could not be considered in our meta-analysis after full-text review due to missing raw data to calculate OR and no response after contacting the corresponding author.¹⁹

We excluded studies on other malignant gastric tumours (i.e. mucosa-associated lymphatic tissue lymphoma, gastrointestinal stromal, and neuroendocrine tumours) or gastric manifestations from other malignancies such as hematologic neoplasms, as well as those with exclusively paediatric populations (<18 years). Finally, cross-sectional studies, case reports, and pooled estimates from meta-analyses were excluded; however, suitable individual studies from meta-analyses were considered.

Two independent reviewers (D.S., D.G. or B.M.) screened titles and abstracts. Consensus for emerging discrepancies between reviewers was achieved after discussion with a senior author (B.M.). The inter-observer agreement regarding study inclusion was good for the screening of abstracts and titles (kappa = 0.87) and excellent for full-text articles (kappa = 1). A reviewer (D.S.) extracted data on a standardized form followed by verification by the other reviewers (N.K., B.M.). Two independent reviewers (D.S., B.M.) assessed study quality using a tool for assessing risk of bias in non-randomized studies of interventions (ROBINS-I).²⁰ Different bias domains (confounding, selection of study participants, classification of interventions, deviation from intended interventions, missing data, outcome measurements, selection of reported results) were analysed for every individual study. The risk of bias for every domain and the whole study was estimated according to predefined questions and classified as ‘low’, ‘moderate’, ‘serious’, and ‘critical’ according to published instructions in the Cochrane Handbook for Systematic Reviews of Interventions version 6.2.²¹

Ethical considerations

This meta-analysis included only anonymised patient data upon informed written consent from published cohort and case-control studies approved by their respective institutional review boards complying with the 1975 Declaration of Helsinki.

Statistical approach

All calculations were performed using R version 4.0.4 and the package ‘meta’ version 4.19-0.²² For our meta-analysis, we extracted adjusted OR and respective 95% CI and calculated new unadjusted OR in case of different point estimates (see above). Consequently, we calculated pooled OR by random-effect models using inverse variance weights and the Hartung–Knapp–Sidik–Jonkman (HKSJ) method.^23–26 As an additional approach, we calculated pooled OR from provided raw binary outcome data by random-effect models using the Mantel–Haenszel and HKSJ methods. Heterogeneity between studies was assessed by means of I² statistics. We chose a random-effects approach due to an expected high heterogeneity and disproportional weight of a single study with a very high number of participants²⁷ in an experimental fixed effects model.

In addition, we performed stratified analysis for study design (case-control versus retrospective cohort studies), GC site (cardia/proximal GC versus non-cardia/distal GC), and duration of PPI intake (<1 year, 1–3 years, >3 years).

Moreover, we calculated the pooled prevalence of GC among PPI users and non-users in five retrospective cohort studies.^17,28–31

Finally, we assessed the risk of publication bias using Egger’s test.³²

Results

General characteristics of included studies

We identified 2,396 records during initial screening. After exclusion of 397 duplicates, 1,966 articles based on titles and abstracts, and 19 articles after full-text review, 14 studies remained for the qualitative synthesis (Supplemental Appendix Figure 2). Baseline characteristics of the individual studies are presented in Table 1. One large Danish case-control study¹⁹ did not provide sufficient information to be included in our meta-analysis; however, it was considered in the qualitative synthesis.

Table 1.

Main characteristics of the included case-control and cohort studies.

Study	Study design	Observation period	Overall population in meta-analysis	Origin of study sample	Sex (female, %)	Age (years);min–max (years)	Follow-up (years); person-years	PPI (n)	Non-PPI (n)	Non-GC (n)	GC (n)	Lag time	ROBINS-I
García Rodríguez et al.³³	Case-control study	1 January 1994–31 December 2001	10,293	Subjects registered in the general practitioners’ research database, United Kingdom	N/A	N/A; 40–84	N/A; 4,340,207^a	442	9,851	9,786	507	1 year	Moderate
Tamim et al.³⁴	Case-control study	1 January 1995–31 December 2003	8,229	Subjects registered in the health insurance plan databases, Québec, Canada	47.9	75.5 (mean); N/A	N/A; N/A	1,299	6,930	7,158	1,071	6 months	Moderate
Poulsen et al.²⁸	Retrospective cohort study	1 January 1990–31 December 2003	36,268	Subjects registered in the Danish National Health Care System, North Jutland County, Denmark	53	N/A; 40–84	N/A; 2,397,759^b	18,790	17,478	36,107	161	1 year	Moderate
Brusselaers et al.^14,29	Retrospective cohort study	1 July 2005–31 December 2012	815,700	Subjects registered in high-quality and nationwide registries, Sweden	58.5	N/A; ⩾ 18	4.9 (mean); 3,866,836^c	795,490	20,210	813,759	1,941	1 year	Low
Cheung et al.³¹	Retrospective cohort study	1 January 2003–31 December 2012	63,397	Subjects after HP eradication registered in Clinical Data Analysis and Reporting System of the Hong Kong Hospital Authority, Hong Kong	53.5	54.7 (median); ⩾ 18	7.6 (median); 483,260	3,271^d	60,126^d	63,244^d	153^d	6 months	Moderate
Niikura et al.³⁰	Retrospective cohort study	1998–2017	533	Subjects after HP eradication registered in the database of University Tokyo Hospital, Tokyo, Japan	44	N/A; N/A	6.9 (mean)^e; N/A	118	415	512	21	N/A	Severe
Lai et al.¹⁵	Case-control study	2000–2013	1,298	Subjects registered in the National Health Insurance Programme, Taiwan	34.4	65.6 (mean); 20–84	N/A	539	759	649	649	1 year	Moderate
Peng et al.³⁵	Case-control study	1996–2011	2,122	Subjects with GORD registered in the National Health Insurance Research Data Base, Taiwan	N/A	N/A; ⩾ 20	N/A	1,693	429	1,061	1,061	1 year	Moderate
Chien et al.²⁷	Case-control study	2000–2010	241,244	Subjects enrolled in the National Health Insurance Programme, Taiwan	41.9	69 (mean); N/A	6.6 (mean); N/A	12,535	228,709	219,188	21,956	3 years	Moderate
Lee et al.³⁶	Case-control study	1 January 1996–31 December 2016	11,776	Subjects from the Kaisers Permanente Northern California Population, United States	25.3	N/A ⩾ 18	N/A	937	10,839	10,543	1,233	2 years	Moderate
Crane et al.³⁷	Case-control study	1971–2000	242	Subjects from the Rochester Epidemiology Project, Olmsted County, United States	36.5	73 (mean);	N/A	2	240	121	121	N/A	Moderate
Wennerstrom et al.¹⁹	Case-control study	1 January 1995–31 December 2011	1,179,202^c	Subjects registered in the Danish Prescription Drug Registry, the Danish Civil Registration System and the Danish Cancer Registry, Denmark	55.5	N/A; N/A	N/A; N/A	1,179,202	N/A	1,177,855^c	1,347^c	1 year	Moderate
Liu et al.¹⁷	Case-control study	1 January 1999–30 April 2011	8,055	Subjects registered in the Primary Care Clinical Information Unit database in Scotland	42.9	N/A All age groups	5.1 years (median) N/A	1,542	6,513	6,607	1,448	1 year	Moderate
Liu et al.¹⁷	Retrospective cohort study	1 January 1999–30 September 2014	471,779	Subjects registered in the United Kingdom Biobank	54	N/A All age groups	4.6 years (median) N/A	46,146	425,633	471,529	250	1 year	Moderate

EGD, esophago-duodeno-gastroscopy; GC, gastric cancer; GORD, gastro-oesophageal reflux disease; HP, Helicobacter pylori; HR, hazard ratio; H2RA, histamine 2 receptor antagonist; IRR, incidence rate ratio; n, number; N/A, not available or not applicable; OR, odds ratio; PPI, proton-pump inhibitors; ROBINS-I, risk of bias in non-randomized studies of interventions; SIR, standard incidence ratio.

Only numbers for both GC and oesophageal adenocarcinoma provided; no separate calculation for GC cases possible.

Not available in original publication. Calculated from Table 2: Addition of unexposed and PPI (2 + prescr.) groups both for person-years and GC cases.

Numbers available for PPI users, but not for control group in original publication.

Multivariate analyses without trimming.

No separate information on PPI and H2RA users available.

Our final meta-analysis summarizes eight case-control^{15–17,27,33–35,37} and five retrospective cohort studies^17,28–31 with an overall population of 1,662,881 individuals from Europe (Denmark, Sweden, United Kingdom), Canada, United States, and Asia (Hongkong, Japan, Taiwan).

Three studies^17,19,28 assessed GC risk in relation to the number of PPI prescriptions, three studies according to the number of defined daily doses (DDDs)^16,17,34; five studies reported the duration of PPI use,^{14,16,17,29,31,33} among those one study with additional information on PPI frequency.³¹ The five remaining studies did not provide detailed analyses either on PPI frequency, dosage, or duration.^{15,27,30,34,37}

In two studies,^14,28,29 there was no information on the proportion of cases and controls in unexposed individuals. Thus, we re-defined the PPI-unexposed group as individuals on histamine 2 receptor blocker (H2RA). In the Swedish cohort, updated PPI exposure and GC data¹⁴ were combined with the H2RA population in the original paper²⁹ for the main analysis and the stratified analyses by PPI duration. Eight studies^{14,15,19,28,29,31,33,35,34} defined GC by localization according to International Classification of Disease (ICD) codes, three studies^29,33,37 provided information on GC histology; however, one of those without ICD codes.³⁷ Three studies did not specify the diagnostic criteria for GC.^27,30,36

Results from the quality assessment

The methodological quality of studies using the ROBINS-I tool²⁰ varied with an overall low risk of bias in 1 study,^14,29 moderate risk in 11 studies^{15–17,27,28,31,33–35,37} and severe risk in 1 study³⁰ (Supplemental Appendix Table 1). Egger’s test³² showed a significant risk of publication bias (intercept: 3.98, 95% CI: 2.46–5.50, p < 0.01).

Results from quantitative analyses

Association between GC and PPI use

Our meta-analysis of all 13 studies¹⁵–17,27 –31,33 –35,37 indicated an increased GC risk in PPI users (OR: 1.94, 95% CI: 1.47–2.56, I² = 82%, p < 0.01) in random-effect models (Figure 1). The pooled GC prevalence of five retrospective cohort studies^17,28–31 was 0.6% (95% CI: 0.1–4%) among PPI users and 0.2% (95% CI: 0.04–1%) among non-users, both with high statistical heterogeneity (I² = 99%, p < 0.01, Supplemental Appendix Figure 4). On a study level, Niikura et al.³⁰ report a GC prevalence rate of 11% among PPI users, which is 22-fold higher than in another Asian collective from Hong Kong³¹ (0.5%). Our secondary analysis using newly calculated unadjusted OR from raw binary outcome data from all 13 studies¹⁵–17,27 –31,33 –35,37 yielded similar results (OR: 2.05, 95% CI: 1.54–2.64, I² = 91%, p < 0.01). Stratified analyses according to study design showed a higher GC risk estimate in the five included retrospective cohort studies^17,28–31 (OR: 2.76, 95% CI: 1.56–4.88, I² = 63%, p = 0.03) than in the eight included case-control studies¹⁵–17,27,33 –35,37 (OR: 1.59, 95% CI: 1.23–2.05, I² = 81%, p < 0.01, Supplemental Appendix Figure 3).

Figure 1.

Association between PPI use and GC risk.

Association between GC and PPI use according to GC site

Seven out of 13 studies contributed to a stratified meta-analysis according to GC site.^{17,27,29,33–35,37} There was a significant risk increase in non-cardia GC (OR: 2.20, 95% CI: 1.44–3.36, I² = 77%) with a non-significant trend towards an increased risk in the cardia (OR: 1.77, 95% CI: 0.72–4.36, I² = 66%, Figure 2).

Figure 2.

Association between PPI use and GC risk stratified by GC site.

Association between GC and PPI use according to PPI duration

Overall, 3 out of 13 studies^14,29,31,33 could be meta-analysed for the duration of PPI intake. There was no trend for a duration-dependent effect of PPI use (<1 year: OR: 2.29, 95% CI: 2.13–2.47, I² = 0%, p = 0.97; 1–3 years: OR: 1.46, 95% CI: 0.53–4.01, I² = 35%, p = 0.21; >3 years: OR: 2.08, 95% CI: 0.56–7.77, I² = 61%, p = 0.08; Figure 3).

Figure 3.

Association between PPI use and GC risk stratified by duration of PPI use.

Results from qualitative analyses

GC risk according to location: non-cardia versus cardia

A meta-analysis regarding location-dependent GC risk was not possible, as most studies did not provide stratified information regarding GC location. Two additional studies could not contribute to the stratified meta-analysis according to GC site due to missing information.^19,31 While a retrospective cohort study from Hong Kong³¹ found a significant GC risk increase for PPI users in the non-cardia (HR: 2.56, 95% CI: 1.46–4.49), but not cardia region (HR: 1.24, 95% CI: 0.35–4.34), a large Danish case-control study¹⁹ obtained a significant risk increase in both regions (non-cardia GC: HR: 4.68, 95% CI: 4.14–5.29; cardia GC: HR: 2.51, 95% CI: 2.26–2.79).

GC risk according to age and sex

Two studies provided analyses accounting for age and sex.^15,29 Stratified meta-analyses according to age and sex were not feasible due to missing data in the Taiwanese study,¹⁵ which precludes the calculation of an OR with a meta-analysis. In a large retrospective Swedish cohort,²⁹ the highest relative GC risk was found in PPI users < 40 years (SIR: 22.76, 95% CI: 15.94–31.52), whereas relative GC risk was lowest in those >70 years (SIR: 2.76, 95% CI: 2.61–2.92). The relative GC risk was significantly higher in male (SIR: 3.65, 95% CI: 3.45–3.85) than in female PPI users (SIR: 3.07, 95% CI: 2.87–3.28). In contrast, a Taiwanese case-control study¹⁵ found no inter-sex difference in GC risk among PPI users (OR: 0.98, 95% CI: 0.77–1.25) and no GC risk increase with increasing life years (OR: 1.00, 95% CI: 0.99–1.01). In the analysis of two British databases,¹⁷ there was no significantly increased GC risk in males and females after full adjustment in the United Kingdom Biobank (male: HR: 1.14, 95% CI: 0.70–1.87; female: HR: 1.73, 95% CI: 0.86–3.45), whereas the Primary Care Clinical Informatics Unit (PCCIU) database showed a significantly increased GC risk in women, but not men (male: OR: 1.26, 95% CI: 0.97–1.62; female: OR: 1.84, 95% CI: 1.38–2.47).

GC risk according to frequency and intensity of PPI use

No meta-analysis was feasible for frequency and intensity of PPI use due to very heterogeneous quantifications and stratifications of PPI intake ranging from number of prescriptions,^17,19,28 DDD^16,17,34 to insufficient or no definition.^{15,27,30,34,37} For instance, Liu et al.¹⁷ used other subgroup cut-offs for daily dosages per year in the PCCI cohort (1–183, 184–365, 366–1095, ⩾1096) than Tamim et al.³⁴ (0–41, 42–119, 120–479, >480 daily dosages per year). Lee et al.³⁶ provided only mean daily dosages (0.75, 0.76–1.49, >1.5) without referring to a yearly dosage, so that a valid meta-analysis of these heterogeneous dosage units was not feasible. In contrast to the meta-analysis by Jiang et al.,¹⁰ we did not include Poulsen et al.²⁸ because the authors provided stratified analyses for the length of follow-up (with different time frames), but not for the duration of PPI intake.

Nevertheless, some individual studies deserve special attention. Cheung et al.³¹ stratified for PPI frequency in context with duration. Only daily PPI intake was significantly associated with an increased GC risk during a prescription period of at least 1 year (HR: 5.04, 95% CI: 1.23–20.61), 2 years (HR: 6.65, 95% CI: 1.62–27.26), and 3 years (HR: 8.34, 95% CI: 2.02–34.41). On the contrary, weekly and less than daily PPI intake did not significantly alter GC risk.

The retrospective cohort study by Poulsen et al.²⁸ found a trend towards a slightly increased GC risk with rising number of prescriptions; however, this effect was significant only in the subgroup with at least 15 prescriptions (incidence rate ratio: 2.1, 95% CI: 1.0–4.7). Furthermore, a Danish case-control study¹⁹ found a gradually increasing GC risk with rising number of prescriptions (1–4 prescriptions: HR: 1.91, 95% CI: 1.64–2.22; ⩾15 prescriptions: HR: 3.56, 95% CI: 2.82–4.46) in the cardia but not in other regions. In the PCCIU database, Liu et al.¹⁷ found the highest GC risk increase in PPI users with the lowest DDD and numbers of PPI prescriptions (1–183 DDD versus non-users: OR: 1.84, 95% CI: 1.43–2.38; 1–6 prescriptions: OR: 1.85, 95% CI: 1.44–2.37), whereas this effect mitigated to the highest DDD and numbers of PPI prescriptions (⩾1096 DDD versus non-users: OR: 1.30, 95% CI: 0.91–1.85; ⩾ 36 prescriptions versus non-users: OR: 0.97, 95% CI: 0.64–1.47). This trend was significant for DDD (p for trend = 0.02), but not for the number of prescriptions (p for trend = 0.07).

GC risk according to HP infection status

The literature summarized in our review is highly heterogeneous regarding the consideration of HP infection status; hence, no meta-analysis was possible. Brusselaers et al.²⁹ showed a substantially increased relative GC risk in PPI users with HP infection (SIR: 9.76, 95% CI: 8.87–10.71) compared with other PPI indications (SIR: 3.55, 95% CI: 3.27–3.86). Two studies^30,31 specifically included individuals after HP eradication; therefore, a direct intra-study assessment of HP effects was impossible. However, there was no evidence of stronger PPI effects than in the other studies. Wennerstrom et al.¹⁹ conducted a sensitivity analysis showing no difference in effect size upon censoring patients at the time of HP eradication. Poulsen et al.²⁸ incorporated HP infections as a confounder in their multivariate analysis, but the change in effect size could not be compared as unadjusted results were lacking.

Discussion

In our meta-analysis, we found a significant twofold increased overall risk of GC of PPI users compared with controls. Stratification according to GC site indicated a significantly increased non-cardia GC risk while there was a non-significant trend towards an increased cardia GC risk. Because the adenocarcinoma of the cardia and distal oesophagus are sometimes difficult to discriminate due to the overlapping pattern of local expansion, an accurate classification can be difficult also in epidemiologic studies.^38,39 Due to different risk factors for adenocarcinoma at proximally or distally to oesophago-gastric junction,⁴⁰ an accurate identification of risk factors at this site is challenging. In our meta-analysis, the true effect of PPI to the carcinogenesis of cardiac GC may be weakened by the treatment of gastro-oesophageal reflux, which represents a common risk factor for distal oesophageal adenocarcinoma. However, there was no clear duration-dependent risk increase in PPI users with discrepant results in individual studies.

Our meta-analysis includes global evidence from five retrospective cohorts/registries and eight case-control studies with an overall population of 1,662,881 individuals. Most studies^{14–17,19,27–29,31,33–35} used large populations from national and regional databases in Asia, Europe, and North America, which increases generalizability. In comparison with the latest meta-analyses^10–13 including up to eight studies on the association between PPI use and GC risk, we could include five additional studies and more than 500,000 additional patients. This enabled more robust stratified analyses according to GC site and standardized time frames for PPI duration compared with previous meta-analyses. Two recent meta-analyses deserve special mention. Lin et al.¹² included only a fraction of our studies (eight instead of 13 studies in our meta-analysis) with a literature search in 2018. The same applies to the stratified analysis according to GC site where they meta-analysed five studies compared with seven studies in our updated meta-analysis. In addition, this meta-analysis has some methodological shortcomings: Stratified analyses according to PPI duration were not homogeneous in terms of PPI duration. While we used standardized cut-offs for PPI duration (<1 year, 1–3 years, ⩾3 years), Lin et al. mixed different PPI durations in various categories enabling for the integration of Poulsen et al.²⁸ and Lai et al.¹⁵ in some but not all subgroups. However, this approach is potentially misleading due to the missing standardization of PPI durations. Furthermore, only the subgroups with a PPI use ⩾3 years and ⩾1 year were considered from the retrospective cohort study by Cheung et al.³¹ so that one of the largest retrospective cohort studies was only partly considered in this stratified analysis. Finally, they used apparently inconsistent inclusion criteria for HP subgroups in their stratified meta-analysis for HP status with current and past HP infection: while the authors classified two entire cohorts with probable – but not confirmed – HP eradication as ‘past HP infection’ groups,^30,31 only a fraction of the study population was analysed in the other studies,^15,29 which markedly enhances the risk of misclassification and selective reporting which impacts the generalizability of the data.

The other recent meta-analysis by Song et al.¹³ meta-analysed only a fraction of our included studies (5 versus 13 studies). In contrast to our meta-analysis on PPI exposure, the main exposure was acid-suppressive drugs and only one subgroup analysis of five studies for PPI use was available. Therefore, the authors were not able to present detailed stratified analysis on cardia and non-cardia GC and duration-dependent GC risk for PPI users.¹³

Stratified analyses according to study design indicated an increased GC risk not only in case-control studies but also in cohort studies, which are typically less prone to selection and recall bias. Our meta-analysis thus reflects the growing body of evidence for an association of PPI usage with GC in observational studies in highly different settings.

Addressing the risk of reverse causation, 11 out of 13 studies incorporated a lag time for GC diagnosis removing PPI usage limited to the previous 6^31,34 or 12 months,^{14,15,17,19,28,29,33,35} 2 years,^16,17 3 years,^17,27 or 4 and 5 years¹⁷ before GC diagnosis. Furthermore, 10 out of 13 studies excluded patients with a history of any cancer^{19,28,29,33–35} or history of GC.^15–17,31

Nevertheless, this work encountered several limitations due to very heterogeneous original studies concerning study design, quality, cases and controls definitions, PPI indications, quantification of PPI exposure, and effect sizes, which may also result in highly heterogeneous pooled OR and prevalence in the included studies. First, some original publications did not provide risk estimates or did only report incomplete information on the exposure and outcome distribution within the study population. We attempted to overcome this issue by calculating new unadjusted OR based on the published figures, and by contacting the corresponding authors for additional information. Nevertheless, unadjusted point estimates exhibit the risk of confounding and should be interpreted cautiously. However, in a sensitivity analysis comparing newly calculated OR with remaining studies, we found highly similar results. Second, in two studies,^14,28,29 we could not retrieve information on the unexposed group so that we used H2RA users as such. In a sensitivity analysis excluding these studies, however, the results remained unaffected. Third, although all studies indicate an increased GC risk among PPI users compared with controls, its extent varies, reflected by the high statistical heterogeneity in our and previous meta-analyses. Leave-one-out analyses, however, revealed no disproportional influence of any single study, and Graphical Display of Study Heterogeneity plots⁴¹ demonstrated no overt sub-clusters.

Fourth, we could only include retrospective cohort and case-control studies, which are subject to residual confounding. Fifth, immortal time bias⁴² leading to a misclassification of PPI users and non-users remains an issue in 11 out of 13 studies; however, 2 studies accounting for this issue still found a significant association of PPI exposure and GC risk.^14,15 Sixth, a publication bias cannot be excluded based on the Eggers test.

Seventh, missing data for important risk factors in the individual studies precluded adjustments for common GC risk factors, which may mitigate this association. Despite representing one of the most important confounders in the association between PPI and GC risk,⁴³ most studies did either not adjust for HP infection status^17,33,34,37 or analysed study populations after documented HP therapy (however, without proven eradication).^30,31 Therefore, there might be synergistic carcinogenic effect by HP infection and PPI through hypergastrinaemia^44,45 or early dyspeptic symptoms related to gastritis, precancerous lesions such as intestinal metaplasia, dysplasia, and early GC might favour regular PPI use. This may also explain the markedly increased GC risk in individuals with PPI use <1 year without further relevant risk increase in subgroups with longer PPI use. However, considering the decreasing prevalence of HP in Western countries, confounding by HP is unlikely to be the main explanation of the association of PPI usage and GC risk. Eighth, most studies failed to control for other common GC risk factors,^2–5 such as smoking^{15,19,27,29,30,34,35,37} and obesity,^{15,19,27–30,34,35,37}, and no study adjusted for dietary salt and nitroso compound intake.^{3,5,14,17,27–30,33–37}

Another inherent limitation of most included studies is the missing information on precancerous lesions such as intestinal dysplasia and metaplasia⁵ in most included studies,^{15,17,19,27,28,31,33,34,37} which may both favour regular PPI use and the development of GC. This could lead to an overestimation of the effect due to residual confounding. Finally, no universally accepted definition of long-term PPI usage is available and definitions of long-term PPI use were missing in 10 included studies^{14,17,27–30,34–37} and highly heterogeneous in the remaining studies,^3,33,35 so that we strictly avoided this term in our meta-analysis. However, our data did not show a duration-dependent risk increase in GC risk among PPI users.

In the randomized controlled COMPASS trial, Moayyedi et al.⁴⁶ found no statistically significant difference in the risk for gastrointestinal tumours between the pantoprazole and placebo group (HR: 1.04, 95% CI: 0.77–1.40); however, the study suffers from several methodological shortcomings and conflicts of interest in the authorship, as previously discussed.⁴⁷ For instance, it was underpowered for safety outcomes, comprised a highly selected study population, and did not provide information on drop-out rates as well as separate analyses on GC risk. In addition, two of three study arms included daily aspirin intake by study design, which in combination with regular PPI intake was associated with a considerably decreased GC risk increase compared with PPI alone in retrospective analyses.^48,49

Even though our meta-analysis demonstrates a clear association between PPI use and GC risk, causality cannot be proven. Large randomized controlled trials in this field are ethically challenging and not feasible. However, this meta-analysis includes evidence from large registry studies with a retrospective cohort design, but with prospective data collection, therefore the highest level of evidence available by now. Yet, future research should focus on a comprehensive and accurate prospective documentation of GC risk factors, endoscopic/histopathological findings, co-medication, and co-treatments, which would enable application of sophisticated causal inference techniques⁵⁰ as used in other studies on risks of PPI.⁵¹

Conclusion

Our meta-analysis found an almost twofold increased GC risk among PPI users compared with controls. Despite a clear association between PPI exposure and GC risk, causation cannot be proven. Moreover, there was substantial heterogeneity in study designs, populations, definitions of exposure and outcome, and residual confounding as well as a potential publication bias remain of concern. Indications for PPI treatment should be verified, PPI use avoided, and non-indicated PPI unprescribed whenever possible.

Supplemental Material

sj-docx-1-tag-10.1177_17562848211051463 – Supplemental material for Association between proton-pump inhibitors and the risk of gastric cancer: a systematic review with meta-analysis

Supplemental material, sj-docx-1-tag-10.1177_17562848211051463 for Association between proton-pump inhibitors and the risk of gastric cancer: a systematic review with meta-analysis by Daniel Segna, Nele Brusselaers, Damian Glaus, Niklas Krupka and Benjamin Misselwitz in Therapeutic Advances in Gastroenterology

Footnotes

Acknowledgements

Niklas Krupka and Benjamin Misselwitz contributed equally to this work and share last authorship.

Author contributions

D.S.,B.M., and N.B. designed the concept of the meta-analysis. D.S., D.G., and B.M. designed and performed the literature research and screened abstracts and manuscript and extracted the data. N.K. performed the statistical analysis. D.S., B.M., N.K., and N.B. wrote the paper. All authors approved the paper and contributed important intellectual content.

Conflict of interest statement

The authors declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: B.M. reports grants (MSD) and personal fees (MSD, Vifor, Takeda), and D.S. reports travelling fees (AbbVie, Vifor, Gilead), all outside the submitted work. The other authors have nothing to disclose.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Daniel Segna

Nele Brusselaers

Supplemental material

Supplemental material for this article is available online.

References

Bray

Ferlay

Soerjomataram

, et al. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2018; 68: 394–424.

Gonzalez

Pera

Agudo

, et al. Smoking and the risk of gastric cancer in the European Prospective Investigation into Cancer and Nutrition (EPIC). Int J Cancer 2003; 107: 629–634.

Lee

Chiang

Chou

, et al. Association between helicobacter pylori eradication and gastric cancer incidence: a systematic review and meta-analysis. Gastroenterology 2016; 150: 1113–1124.e5.

Yang

Zhou

Chen

, et al. Overweight, obesity and gastric cancer risk: results from a meta-analysis of cohort studies. Eur J Cancer 2009; 45: 2867–2873.

Song

Ekheden

Zheng

, et al. Incidence of gastric cancer among patients with gastric precancerous lesions: observational cohort study in a low risk Western population. BMJ 2015; 351: h3867.

Karimi

Islami

Anandasabapathy

, et al. Gastric cancer: descriptive epidemiology, risk factors, screening, and prevention. Cancer Epidemiol Biomarkers Prev 2014; 23: 700–713.

Forgacs

Loganayagam

Overprescribing proton pump inhibitors. BMJ 2008; 336: 2–3.

Ahn

Eom

Jeon

, et al. Acid suppressive drugs and gastric cancer: a meta-analysis of observational studies. World J Gastroenterol 2013; 19: 2560–2568.

Tran-Duy

Spaetgens

Hoes

, et al. Use of proton pump inhibitors and risks of fundic gland polyps and gastric cancer: systematic review and meta-analysis. Clin Gastroenterol Hepatol 2016; 14: 1706–1719.e5.

10.

Jiang

Wen

, et al. Relationship between long-term use of proton pump inhibitors and risk of gastric cancer: a systematic analysis. J Gastroenterol Hepatol 2019; 34: 1898–1905.

11.

Wan

, et al. Long-term proton pump inhibitors use and risk of gastric cancer: a meta-analysis of 926 386 participants. Gut 2019; 68: 762–764.

12.

Lin

J-L

Lin

J-X

Zheng

C-H

, et al. Long-term proton pump inhibitor use and the incidence of gastric cancer: a systematic review and meta-analysis. J Gastric Surg 2020; 2: 1–11.

13.

Song

Jeon

Squires

The association between acid-suppressive agent use and the risk of cancer: a systematic review and meta-analysis. Eur J Clin Pharmacol 2020; 76: 1437–1456.

14.

Brusselaers

Lagergren

Engstrand

Duration of use of proton pump inhibitors and the risk of gastric and oesophageal cancer. Cancer Epidemiol 2019; 62: 101585.

15.

Lai

Lin

, et al. Proton pump inhibitors and risk of gastric cancer in a case-control study. Gut 2019; 68: 765–767.

16.

Lee

Merchant

Schneider

, et al. Proton pump inhibitor use and risk of gastric, colorectal, liver, and pancreatic cancers in a community-based population. Am J Gastroenterol 2020; 115: 706–715.

17.

Liu

McMenamin

ÚC

Johnston

, et al. Use of proton pump inhibitors and histamine-2 receptor antagonists and risk of gastric cancer in two population-based studies. Br J Cancer 2020; 123: 307–315.

18.

Moher

Liberati

Tetzlaff

, et al. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. BMJ 2009; 339: b2535.

19.

Wennerstrom

ECM

Simonsen

Camargo

, et al. Acid-suppressing therapies and subsite-specific risk of stomach cancer. Br J Cancer 2017; 116: 1234–1238.

20.

Sterne

Hernán

Reeves

, et al. ROBINS-I: a tool for assessing risk of bias in non-randomised studies of interventions. BMJ 2016; 355: i4919.

21.

Sterne

JACHM

McAleenan

Reeves

, et al. Chapter 25: assessing risk of bias in a non-randomized study. In: Cochrane handbook for systematic reviews of interventions, 2021, p. 6.2, https://training.cochrane.org/handbook/current/chapter-25

22.

Balduzzi

Rücker

Schwarzer

How to perform a meta-analysis with R: a practical tutorial. Evid Based Ment Health 2019; 22: 153–160.

23.

Hartung

JKG

. A refined method for the meta-analysis of controlled clinical trials with binary outcome. Stat Med 2001; 20: 3875–3889.

24.

Hartung

An alternative method for meta-analysis. Biom J 1999; 46: 901–916.

25.

Sidik

KJJ

. A simple confidence interval for meta-analysis. Stat Med 2002; 21: 3153–3159.

26.

Sidik

KJJ

. On constructing confidence intervals for a standardized mean difference in meta-analysis. Commun Stat Simul Comput 2003; 32: 1191–1203.

27.

Chien

Huang

Shao

, et al. Proton pump inhibitors and risk of periampullary cancers: a nested case-control study. Int J Cancer 2016; 138: 1401–1409.

28.

Poulsen

Christensen

McLaughlin

, et al. Proton pump inhibitors and risk of gastric cancer: a population-based cohort study. Br J Cancer 2009; 100: 1503–1507.

29.

Brusselaers

Wahlin

Engstrand

, et al. Maintenance therapy with proton pump inhibitors and risk of gastric cancer: a nationwide population-based cohort study in Sweden. BMJ Open 2017; 7: e017739.

30.

Niikura

Hayakawa

Hirata

, et al. Long-term proton pump inhibitor use is a risk factor of gastric cancer after treatment for Helicobacter pylori: a retrospective cohort analysis. Gut 2017; 22: 22.

31.

Cheung

Chan

Wong

AYS

, et al. Long-term proton pump inhibitors and risk of gastric cancer development after treatment for Helicobacter pylori: a population-based study. Gut 2018; 67: 28–35.

32.

Egger

Davey Smith

Schneider

, et al. Bias in meta-analysis detected by a simple, graphical test. BMJ 1997; 315: 629–634.

33.

García Rodríguez

Lagergren

Lindblad

Gastric acid suppression and risk of oesophageal and gastric adenocarcinoma: a nested case control study in the UK. Gut 2006; 55: 1538–1544.

34.

Tamim

Duranceau

Chen

, et al. Association between use of acid-suppressive drugs and risk of gastric cancer. Drug Saf 2008; 31: 675–684.

35.

Peng

Huang

Lin

, et al. Association between proton pump inhibitors use and risk of gastric cancer in patients with GERD. Gut 2019; 68: 374–376.

36.

Lee

Merchant

Schneider

, et al. Long-term proton pump inhibitor use and risk of gastrointestinal cancers. Gastroenterology 2019; 156: 163.

37.

Crane

Locke

3rd Harmsen

, et al. Subsite-specific risk factors for esophageal and gastric adenocarcinoma. Am J Gastroenterol 2007; 102: 1596–1602.

38.

DeMeester

SR.

Adenocarcinoma of the esophagus and cardia: a review of the disease and its treatment. Ann Surg Oncol 2006; 13: 12–30.

39.

Imamura

Watanabe

Oki

, et al. Esophagogastric junction adenocarcinoma shares characteristics with gastric adenocarcinoma: literature review and retrospective multicenter cohort study. Ann Gastroenterol Surg 2021; 5: 46–59.

40.

Carr

Zafar

Saba

, et al. Risk factors for rising incidence of esophageal and gastric cardia adenocarcinoma. J Gastrointest Cancer 2013; 44: 143–151.

41.

Olkin

Dahabreh

Trikalinos

TA.

GOSH: a graphical display of study heterogeneity. Res Synth Methods 2012; 3: 214–223.

42.

Suissa

Proton-pump inhibitors and increased gastric cancer risk: time-related biases. Gut 2018; 67: 2228–2229.

43.

Beckh

Koop

Lankisch

, et al. Gastroenterologie compact. Stuttgart; New York: Georg Thieme Verlag, 2013, https://www.thieme-connect.de/products/ebooks/book/10.1055/b-002-57172

44.

Waldum

Rehfeld

JF.

Gastric cancer and gastrin: on the interaction of Helicobacter pylori gastritis and acid inhibitory induced hypergastrinemia. Scand J Gastroenterol 2019; 54: 1118–1123.

45.

Lundell

Vieth

Gibson

, et al. Systematic review: the effects of long-term proton pump inhibitor use on serum gastrin levels and gastric histology. Aliment Pharmacol Ther 2015; 42: 649–663.

46.

Moayyedi

Veldhuyzen van Zanten

SJO

Hookey

, et al. Proton pump inhibitors and gastric cancer: association is not causation. Gut 2019; 68: 1529–1530.

47.

Simin

Liu

Fornes

, et al. Safety of proton pump inhibitors questioned based on a large randomized trial of patients receiving rivaroxaban or aspirin. Gastroenterology 2020; 158: 1172–1173.

48.

Brusselaers

Lagergren

Maintenance use of non-steroidal anti-inflammatory drugs and risk of gastrointestinal cancer in a nationwide population-based cohort study in Sweden. BMJ Open 2018; 8: e021869.

49.

Cheung

Leung

WK.

Modification of gastric cancer risk associated with proton pump inhibitors by aspirin after Helicobacter pylori eradication. Oncotarget 2018; 9: 36891–36893.

50.

Hernán

Robins

JM.

Using big data to emulate a target trial when a randomized trial is not available. Am J Epidemiol 2016; 183: 758–764.

51.

Xie

Bowe

Yan

, et al. Estimates of all cause mortality and cause specific mortality associated with proton pump inhibitors among US veterans: cohort study. BMJ 2019; 365: l1580.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

1.63 MB