Proton pump inhibitors and the risk of gastric cancer: a systematic review, evidence synthesis and life course epidemiology perspective

Abstract

Objectives

Since proton pump inhibitors (PPI) have been introduced, many concerns were raised regarding potential gastric carcinogenicity. We aim to summarise and weigh the epidemiological evidence and address possible causality.

Design

Systematic literature review, evidence synthesis and life-course assessment.

Data sources

PubMed, Web of Science and Cochrane database (from inception up to October 2024), and back- and forward citation tracking (Web of Science).

Eligibility criteria

Original studies and quantitative evidence syntheses assessing the association between PPIs and gastric cancer in humans, without language restrictions.

Data extraction and synthesis

Study design, definitions (and participant numbers) of PPI use and gastric cancer, study characteristics (setting, period, follow-up, lag-time), age and sex distribution presented in tables and evidence mapping.

Results

We identified 33 original studies, 21 meta-analyses, three umbrella meta-analyses, one individual patient data meta-analysis and a Markov model (2006–2023). PPIs were consistently associated with an increased gastric cancer risk with 20/21 meta-analyses reporting pooled relative risks between 1.3 and 2.9. Available trials were underpowered. Reverse causation/protopathic bias, residual confounding (by indication) and lag time seem the largest methodological challenges, as well as disentangling the effects of Helicobacter pylori and its’ eradication. Insufficient data are available on age and sex-specific risks, with no studies specifically addressing PPIs in young populations. We hypothesise a sensitive-period exposure model, in which PPI use during pregnancy and early life may be particularly damaging regarding long-term cancer risk. An exploration of Swedish cancer incidence data suggests potential cohort effects as overall gastric cancer risk decreased over time (1970–2022). The risk has increased in young (<40 years) men since the early 2000s, ~10 years after the introduction of Helicobacter pylori eradication and PPIs.

Conclusion

Although for older individuals with valid indications, the gastric cancer risk related to PPI use may be limited, we do argue for a more rational and evidence-supported use of PPIs in young populations.

WHAT IS ALREADY KNOWN ON THIS TOPIC?

• Proton pump inhibitors (PPIs) have been used for over 40 years, and epidemiological and mechanistic evidence has been mounting suggesting an association with an increased gastric cancer risk. Yet, this association is still heavily disputed, and PPIs remain among the most commonly prescribed drugs globally, with the use still increasing in several settings and subpopulations.

WHAT THIS STUDY ADDS

• There is strong epidemiological evidence for an association between PPIs and gastric cancer. Although it may be unlikely that PPIs are an independent, singular risk factor, they seem to enhance/accelerate the procarcinogenic effects of Helicobacter pylori. There seem to be age cohort effects, with increasing risks of gastric cancer over time in younger individuals, and in more recent birth cohorts.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• As unwarranted PPI use is abundant, there is room for improvement of prescription practices and need for PPI stewardship. The clinical implications for current older PPI users seem limited. We propose that PPI use at a younger age (childhood and young adults) and in more recent birth cohorts may be more harmful (sensitive period), particularly if exposed over longer periods of time (accumulated effect). Therefore, we argue for a more precautionary approach regarding (long-term) PPI prescription particularly in children and young adults.

Introduction

The first proton pump inhibitor (PPI), omeprazole, was discovered during the late 1970s and soon commercialised in Europe (first in Sweden in 1988) and the USA (1990).^{1 2} PPIs are now among the most commonly used drugs worldwide,³ often as maintenance therapy (up to 10%–30% of adults),⁴,⁹ although the main indications warrant short-term treatment (<4–8 weeks).¹⁰,¹² Up to 25%–70% of long-term PPI use is insufficiently justified,^{13 14} with a large cost to the patient and society.¹⁵,¹⁸ PPIs are even available over-the-counter in many countries (first in Sweden in 1999),¹⁹ although the American Food and Drug Administration (FDA) advised against this in 2000, based on concerns about the long-term benefits and risks.¹⁹

Already in 1985, rat studies showed gastric carcinoids, ‘attributable to pronounced hypergastrinaemia produced as a secondary effect of almost complete inhibition of acid secretion by the large omeprazole doses’.^{20 21} During the 1980–1990s, there were fierce warnings regarding the potential carcinogenic effects of long-term acid suppression.²²,²⁴ In the 1990s, potential genotoxic effects were suggested,²⁵ and case series described gastric glandular cysts in association with omeprazole exposure (1–5 years), with no sex predilection.^{26 27} Fundic gland polyps were reported to disappear after PPI cessation.²⁸,³⁰

The potential pathophysiological mechanisms have been discussed and disputed regularly.¹⁰,¹²³¹ In short, the PPI-induced prolonged and absolute gastric suppression could lead to secondary hypergastrinaemia, change the gastric pH³⁶ and also harm our ‘gastric barrier’ against microbes.³⁷ Gastric bacterial overgrowth may influence gastric carcinogenesis, as already discussed during the early 1990s in relation to PPIs,³⁶ and even suggested in 1895.³⁷ Gastric cancer is now considered a multifactorial disease with Helicobacter pylori as the most established risk factor, discovered in 1984, and formally recognised as a definite (gastric) carcinogen by the World Health Organization since 1994.³⁷,³⁹ To complicate population-based association studies on PPIs and gastric cancer, H. pylori eradication has been increasingly implemented around the same period as PPIs were introduced.⁴⁰ Long-term PPIs have been associated with increased gastric inflammation and development of atrophy among those with active H. pylori infections.⁴¹ This risk could however be reduced or eliminated by testing and eradicating H. pylori, through a combination of antibiotics and PPIs.^{41 42} More recently, PPIs have been described to affect the oral, gastric, gut and even lung microbiota, with stronger population-based effects than antibiotics.⁴³,⁴⁶

The first human epidemiological studies on PPIs and gastric cancer appeared during 2006–2009,⁴⁷,⁴⁹ suggesting an increased risk yet also addressing confounding by indication. Additionally, increased risks have been suggested for many more diseases including other gastro-intestinal cancers and gastro-intestinal infections.⁵⁰,⁵²

Our aim is to provide a comprehensive overview of all epidemiological evidence (original studies and evidence syntheses) regarding PPI and gastric malignancy and to discuss the challenges to establish causality and impact. We also explored potential cohort effects to assess if changes in environmental exposures (PPIs and H. pylori) have affected younger populations differently than older populations based on Swedish (gastric and overall gastro-intestinal) cancer incidence data (1970–2022).

Methods

Systematic literature review and evidence synthesis

This systematic literature search was based on PubMed, the Web of Science and Cochrane, using backward and forward citation tracking of relevant papers, last updated on 23 October 2024 (online supplemental 1), and reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines for systematic reviews (online supplemental 2). We aimed to identify all original human studies assessing the association between PPIs and gastric cancer, as well as all quantitative evidence syntheses. Inclusion and exclusion criteria, data extraction and evidence mapping are described in detail in online supplemental 1.

Power and measures of impact

Post hoc power analyses were used to assess how many person-years would be needed to obtain sufficient power (0.80) to obtain the most conservative estimate reported in the meta-analyses (online supplemental 3). Measures of impact were calculated to assess the absolute effect on the gastric cancer risk among PPI users and on population level, assuming causality (online supplemental 4).

Life course epidemiology and cohort effects

To explore trends over time and potential cohort effects, we extracted gastric cancer and overall gastrointestinal cancer incidence in Sweden, from 1970 to 2022, by sex and age (5 years groups and overall), expressed as number of cases per 100 000 individuals per year.⁵³ Data were extracted from the publicly available Statistical Cancer Database from the National Board of Health and Welfare in Sweden, using the ICD-7 categorisation of cancer (codes: 151 gastric cancer, 150–158 all gastrointestinal cancer).⁵³ We visualised PPI prescription trends (2006–2023), as available through the nationwide Prescribed Drug Registry (established in July 2005).⁵⁴ For power reasons, we defined and merged young individuals as everyone younger than 40 years.

We assumed the epidemiological definition of cohort effects, which presumes different distributions of cancer arising from a changing or new environmental cause affecting age groups differently. In other words, we do hypothesise a period effect that is differentially expressed through age-specific exposure/susceptibility (to PPI and H. pylori eradication on population level).⁵⁵ These cohort effects could be a consequence of an unequal distribution of environmental causes in the population (including no exposure to H. pylori eradication and PPIs before 1989) and/or if some age groups are in the midst of a critical developmental period during which long-lasting effects may be established on lifetime disease risk.⁵⁵ In life course epidemiology, this would be called a sensitive period model, with stronger, more permanent effects being assumed if exposure occurs during, for example, preconception and early life (figure 1).⁵⁶ An accumulation model focuses on the accumulated exposure and duration, assuming no differences in risk depending on the timing (age) of exposure.⁵⁶ Yet, we cannot assess this appropriately using the currently available Swedish data as the Prescribed Drug Registry was only established in July 2005, and no earlier information is available (PPIs available since 1988).⁵⁶ The other two main life course models (pathway model and social mobility model) do not seem applicable for this research question.⁵⁶

Figure 1. Sensitive period model and age cohort effects could affect the association between proton pump inhibitors and gastric cancer risk.

We applied two graphical approaches to assess age period cohort effects:⁵⁵ first, we examined age trends in gastric cancer plotted by year of observation (per 10 years). Second, we assessed the incidence of gastric cancer (5 year average) by increasing age for five different birth cohorts (born 1935 and every 10 years after), so-called cohort-stratified age trends.

Patient and public involvement

None.

Results

Our systematic search resulted in 64 included articles (figure 2): 33 original studies (28 observational studies,^{847 49 57},⁸⁰ five trials);⁸¹,⁸⁵ 26 evidence syntheses (21 meta-analyses,⁸⁶,¹⁰⁶ three umbrella meta-analyses (MA),^{50 107 108} an individual patient data meta-analysis (IPD-MA)¹⁰⁹ and a Markov model);¹¹⁰ and five additional case-control studies retrieved through the IPD-MA not previously citing eligible data (not described separately).¹¹¹,¹¹⁵

Figure 2. Flowchart describing the selection of relevant epidemiological evidence.PPIs, proton pump inhibitors.

The evidence map (figure 3) visualises which original studies are included in each evidence synthesis. Unfortunately, we could not retrieve the full text of two meta-analyses,^{90 92} cited in one umbrella meta-analysis,¹⁰⁸ and therefore do not know which original studies were included.

Figure 3. Evidence map showing which original studies were included in the different evidence syntheses regarding the association between proton pump inhibitors (PPIs) and gastric cancer. MA, meta-analyses; IPD-MA, individual patient-data A.

Observational studies

The 28 observational studies included a pharmacovigilance study (2023),⁶⁷ a Mendelian randomisation study (2023),⁸⁰ nine case-control studies (2006–2022),^{47 49 61 62 64 71 72 76} 16 cohort studies (2009–2023)^{848 57},60 63 65 66 68 and one study reporting both a cohort and case-control design (therefore mentioned twice in online supplemental 5).⁷³ Methodological heterogeneity is high, and all studies originated from countries with a very high human development index (ranked among the first 30 out of 193 countries). Only 13 of the studies reported the proportion of female PPI users (ranging between 8%–63%); and 10 studies reported the sex distribution for gastric cancer (17%–57% in women).

The case-control studies originated from Northern America,^{47 62 64 72} Taiwan,^{61 71 76} the UK^{49 73} and Italy⁶³ and included between 35⁶³ and 7681⁶¹ cases of gastric cancer and an exposure window maximally 20 years.⁷²

The cohort studies from Asia,^{60 66 68 74 75 77 78} Northern Europe,^{8 48 58 59 79} the UK,^{57 73} the USA^{69 70} and Israel⁶⁵ included between 118 and 2 179 048 individuals exposed to PPIs with a maximal follow-up of 28 years.⁵⁷

The pharmaco-vigilance study was based on FDA data,⁶⁷ while the Mendelian randomisation project included UK and Finnish data.⁸⁰ Some of the observational studies defined PPI exposure as any recorded PPI or one prescription, while other studies required a continued or cumulative exposure between 0.5 and 2 years.^{8 59 62 65 68 72 78} Of the 17 cohort studies, 11 compared PPI use with non-users, and seven (also) used an active-comparator design comparing PPIs with H2-receptor antagonists.^{48 57 58 66 69 70 78} Of the case-control studies, only one compared with H2-receptor antagonists.⁷³

Thirteen were population or community based, including all those exposed to PPIs in a certain region, while others were restricted to elderly^{47 65} or specific indications including H. pylori eradication,^{66 68 70 75} atrophic gastritis,⁶³ gastro-oesophageal reflux⁷⁶ or coronary interventions.⁷⁴

Experimental studies

The five identified trials (with 22%–27% female participants)⁸¹,⁸⁵ were all selected because they were included in at least one meta-analysis as none reported (or were designed to assess) gastric cancer as a primary outcome (online supplemental 5). Two do not report any case of gastric cancer,^{81 83} while one reported one case in the H2-receptor antagonist group and none in the PPI group,⁸⁵ and another reported four cases in the PPI group and none in the controls.⁸² The largest trial reported, respectively, 86 and 83 cases of gastrointestinal cancer (not specifically gastric cancer) in the PPI exposed and controls.⁸⁴ The four smallest trials (between 114 and 424 PPI users) had a maximal follow-up between 6 and 24 months,⁸¹,⁸³⁸⁵ while the follow-up in the largest trial (8791 PPI users) ranged between 2 days and 5 years (median 3 years).⁸⁴

Evidence synthesis studies

Of the 21 meta-analyses, 15 reported exclusively on gastric malignancies,^{8687 89},^{93 95 97} while the other six^{88 94 96 100 101 106} reported on other cancer types and/or other adverse events (table 1). These 21 meta-analyses included 2–19 original studies. The three umbrella meta-analyses identified 2–7 meta-analyses (indirectly including between 7 and 14 different original studies).^{50 107 108} The individual patient data meta-analysis¹⁰⁹ was based on five case-control studies described elsewhere previously, yet without addressing the association between PPIs and gastric cancer.¹¹¹,¹¹⁵ The Markov model¹¹⁰ was based on estimates of two original studies.^{8 60}

Table 1. Main characteristics of all included evidence syntheses assessing the association between proton pump inhibitor intake and gastric cancer.

First author	Year	Cancer types	Study design	Number of studies on gastric cancer	Statistical conclusion gastric cancer	Pooled effect
Ahn	2013	Gastric	SR/MA	4 (1 cohort, 3 case controls)	Increased risk	OR=1.39 (1.19–1.64)
Tran-Duy	2016	Gastric	SR/MA	3 (1 cohort, 2 case controls)	Increased risk	OR=1.43 (1.23–1.66)
Islam	2018	Adverse events	SR/MA	2 (2 case controls)	Increased risk	OR=1.78 (1.41–2.25)
Jiang	2019	Gastric	SR/MA	7 (4 cohorts, 3 case controls)	Increased risk	OR=2.50 (1.74–3.85)
Lv Cui-Cui	2019	Gastric	SR/MA*	6 (2 cohorts, 4 case controls)	Increased risk	RR=1.87 (1.00–3.22)
Wan	2019	Gastric	SR/MA	7 (4 cohorts, 3 case controls)	Increased risk	OR=2.10 (1.10–3.09)
Wang	2019	Gastric	SR/MA*	7 (3 cohorts, 4 case controls)	Increased risk	OR=1.80 (1.41–2.29)
Lin	2020	Gastric	SR/MA	8 (4 cohorts, 4 case controls)	Increased risk	RR=2.10 (1.17–3.97)
Song	2020	All incl. gastric	SR/MA	6 (3 cohorts, 3 case controls)	Increased risk	RR=1.53 (1.13–2.07)
Segna	2021	Gastric	SR/MA	13 (5 cohorts, 8 case controls)	Increased risk	OR=1.94 (1.47–2.56)
Zeng	2021	Gastro-intestinal	SR/MA	10 (6 cohorts, 5 case controls)	Increased risk	RR=1.78 (1.38–2.31)
Gao	2022	Gastric	SR/MA	18 (12 cohorts, 7 case controls)	Increased risk	OR=1.94 (1.43–2.64)
Poly	2022	Gastric	SR/MA	13 (8 cohorts, 6 case controls)	Increased risk	RR=1.80 (1.46–2.22)
Zeng	2022	Gastric	SR/MA (letter)	12 (9 cohorts, 4 case controls)	Increased risk	RR=1.81 (1.31–2.50)
Guo	2023	Gastric/colorectal	SR/MA	12 (8 cohorts, 5 case controls)	Increased risk	RR=1.82 (1.46–2.29)
Liu	2023	Gastric/colorectal	SR/MA	12 (9 cohorts, 4 case controls)	Increased risk	RR=2.04 (1.33–2.75)
Pan	2023	Gastric	SR/MA	15 (8 cohorts, 7 case controls)	Increased risk	OR=1.67 (1.37–2.00)
Peng	2023	Gastric	SR/MA	16 (8 cohorts, 8 case controls)	Increased risk	OR=1.75 (1.28–2.40)
Piovani	2023	Gastric	SR/MA	12 (9 cohorts, 4 case controls) + 2 trials	Increased risk	RR=1.33 (1.11–1.59)
Zheng	2023	Gastric cancer	SR/MA	3 trials	No association	OR=1.06 (0.79–1.43)
Tran	2024	Gastro-intestinal	SR/MA	11 (11 cohorts, 1 case control)	Increased risk	RR=2.88 (2.29–3.61)
Broide	2020	Gastric vs QALY and life expectancy	Markov	2 cohorts	PPI use justified for QALY increase (yet limited decrease in life expectancy)	n.r.
Salvo	2021	Safety	Umbrella MA	3 meta-analyses (4 cohorts, 3 case control)	Increased risk	OR=1.78 (1.41–2.25)
Zhang	2022	Cancer	Umbrella MA	8 meta-analyses (incl. 5 cohorts, 3 case controls)	Increased risk	OR=2.07 (1.30–3.29),
Bai	2023	Gastric	Umbrella MA	2 meta-analyses (incl. 8 cohorts, 3 case controls)	Increased risk	RR=1.74 (1.25–2.43)
Sassano	2023	Gastric	IPD-MA	case controls	Increased risk	OR=1.78 (0.76–4.14)

^*Meta-analyses included in umbrella meta-analysis (Zhang 2022) but not retrieved and likely in Chinese (numbers not checked)..

MA, meta-analysis; n.r., not reported; PPI, proton pump inhibitor; QALY, quality adjusted life year; RR, relative risk; SR, systematic review.

From all observational studies, three of the oldest papers (2006–2009)⁴⁷,⁴⁹ were most frequently included in the evidence syntheses. Four of the most recent studies (2022–2023) were not yet included in any meta-analysis.^{63 67 68 80}

Of all 21 meta-analyses, all but one showed a statistically significant increased risk of gastric cancer among PPI users versus controls. The only meta-analysis not reporting a significant association was exclusively based on three of the abovementioned clinical trials¹⁰⁵ and considered underpowered. The largest effect size was reported in the most recent meta-analysis (relative risk (RR)=2.88 (2.29–3.61)),¹⁰⁶ while the lowest was RR=1.33 (1.11–1.59)).¹⁰⁴ The umbrella meta-analyses selected the results of one of the included meta-analyses as the overall result for gastric cancer and therefore did not add new evidence.^{50 107 108} The IPD-MA reported a non-significant increased risk of RR=1.78 (0.76–4.14).¹¹¹,¹¹⁵ The Markov model concluded that PPI use in individuals with functional dyspepsia was associated with a decreased life expectancy of 58.4 days, yet a gain of 2.1 quality adjusted life years, with older people benefiting less than younger people.¹¹⁰

Challenges to establish causality and measures of effect

The Bradford-Hill criteria were used to weigh the level of causal evidence regarding the association between PPI and gastric cancer (table 2). Carcinogenesis is considered a multistep process (‘chronology of cancer’), from initiation to promotion and progression.¹¹⁶ The question remains when PPIs are most harmful, as it may take 10 years to develop gastric cancer after a carcinogenic exposure, through atrophic gastritis and intestinal metaplasia.¹¹⁶ This would imply that many of the included studies would be too short, particularly if PPIs would only affect cancer initiation and not cancer progression.¹¹⁷ H. pylori is a common (approximately half of the global population)¹¹⁸ and main risk factor for gastric cancer, attributable for approximately 90% of non-cardia gastric cancers, and 20% of cardia gastric cancers.¹¹⁹ Infection with H. pylori and the presence of premalignant lesions were not always ascertained before PPI initiation, which does make it difficult to disentangle different risk factors and the temporal effects. Yet, several of the included observational studies adjusted for or restricted by underlying indications,^{8 60 70 75 76} or exclusively selected individuals with baseline endoscopy results.⁶³ The effect of potential confounders was explored in many of the original studies by using multivariable regression modelling. Yet only one study explored if this association differs by age group, showing a dramatically higher relative risk (RR) in younger individuals (standardised incidence ratio (SIR) <40 years=22.76); than older individuals (SIR=2.76 among≥70 years).⁸ Sex-stratified results were scarce and inconsistent, with significantly lower risks among Swedish women (SIR=3.07) than in men (SIR=3.65)⁸ and (non-significantly) higher among UK women (RR=1.73) than men (RR=1.14).⁷³ Stratified results for therapy durations were, however, presented in many papers (number of prescriptions, time exposed), indications and anatomical sublocations. One argument often used to counterclaim that PPIs are carcinogenic is that some studies, particularly trials, do not show significant associations. Our post hoc power analysis used the global sex-stratified estimates of gastric cancer¹²⁰ (online supplemental 4). Overall, 7.6–15.8 million person-years would be needed to detect a 1.3 times higher risk, assuming a 10% prevalence of PPI use. Setting up new PPI trials to assess incident (new) PPI exposure and gastric cancer as the primary outcome would therefore be logistically and ethically challenging. Yet, trials may be useful for assessing short-term effects¹²¹,¹²³

Table 2. Bradford-Hill criteria: association or causation between proton pump inhibitors (PPIs) and gastric cancer?

	Evidence supporting causality?	(Pharmaco-) epidemiological challenge(s)
Strength of association	Highest pooled relative risk in meta-analyses was 2.9 -> ‘small to medium effect size’.131	Impact(absolute risk): even if relatively small relative increase in risk, potential large impact (population attributable fraction) as PPI use is common (assuming causality). Power:several studies were underpowered (too short follow-up, too low number of exposed individuals). Methodological heterogeneity: differences by anatomical site, definitions of PPIs, presence of premalignancy (often no gastroscopy before PPI initiation in younger individuals), indications, age groups, comorbidity, etc.
Consistency	Consistent: all but one meta-analysis showed statistically significant increased risks.	Several study designs: population based, specific indications, active comparator designs, different age groups, definitions of PPI exposure…which enables to explore potential selection bias (eg, only for certain subgroups). Information biasPPI intake: if available over-the-counter likely underestimation of use, and potential misclassification as non-users, diluting effect ->likely underestimation of true effect.
Experiment	Some trials were conducted with short follow-up/designed for other purposes/underpowered. Meta-analyses based on these trials were consequently also underpowered.	Are new trials justified to test PPI carcinogenicity with current level of evidence (precautionary principle), to expose people to PPI over longer periods of time? Is it justified to give placebo? Is there a valid ‘alternative treatment’ with H2RA being outcompeted almost entirely? Active comparator designs have been used in some observational studies, yet indications are not entirely the same.
Specificity	Not specific as it seems PPI also affects other (gastro-intestinal) cancer risks and other outcomes.	H2RAs also affects gastric acidity but the association with gastric cancer is less clear. Not only association with gastric cancer, but many other potential long-term consequences (from infections to kidney failure, and overall survival). Another reason to be extra vigilant (precautionary principle)?
Analogous explanations	Indication for PPI use instead of the drug exposure itself: also increased when compared with individuals with same indication; also increased risks in indications for which no increased risk was expected (eg, gastro protectant). Helicobacter pylori is a major risk factor for gastric cancer and may have a long-lasting carcinogenic effect even after eradication. Confounders: PPI users may be less healthy (lifestyle/comorbidities/poor diet): some of the included studies adjusted for this and these seem important confounders. Socioeconomic status and family history? Reverse causality/protopathic bias: individuals receive PPIs because of early (unrecognised) cancer symptoms/premalignancy.	Confounding by indication (yet also increased risk for indications not associated with gastric cancer; and H2RA association less clear). Residual confounding by lifestyle factors including diet, alcohol intake, chronic comorbidities, overall health? Not all studies adjusted for all of these confounders. Ascertainment bias (faster/more likely diagnosis among those exposed than unexposed because of closer surveillance) seems to be less likely, as it seems PPI users may be diagnosed later if prescribed PPIs for vague symptoms, leading to delayed diagnosis. Healthy user bias: individuals who are more health aware with more access to healthcare are more likely to be prescribed PPI, and this may weaken the association.
Coherence	Also increased risks described for other (gastro-intestinal) cancer types and other diseases.	Other drugs and environmental exposures may affect cancer risk (multicausality).
Biological plausibility	Different mechanisms described for potential carcinogenic effect based on laboratory studies (animal, in vivo/in vitro) incl. gastric atrophy, hyperacidity, pH changes, microbiome changes.	Not an epidemiological challenge, but different mechanistic pathways have been proposed.
Dose response	Dose-response effect or threshold effect (only if above certain dose/duration?); duration of effect (irreversible or reversible after treatment cessation?).	Survival bias (need to live long and healthy to achieve high dose/duration); what about intermittent use/time-changing exposure. Problem of time-varying exposure: is PPI effect irreversible (once exposed, risk remains) or does the risk decrease after cessation of therapy (as in smoking or menopausal hormone therapy)? Or is there a cumulative effect (and safe dose/duration?)? Problem of left-censoring (unknown exposure before start of study) also leading to an underestimated cumulative use and potential misclassification as non-users ->likely underestimation of true effect. Self-reported PPI use and suboptimal compliance may have contributed to misclassification.
Temporal relationship	Effect does not disappear when excluding first 6–12 months of follow-up (reverse causation); some studies removed 2–3 years. Yet, this lag time may not be sufficient?	Reverse causality/protopathic bias (PPIs prescribed for yet unrecognised early carcinogenic symptoms, delaying diagnosis). ‘Chronology of cancer’: if it indeed takes 10+years to develop gastric cancer, few of the original studies are long/powered enough.

Assuming causality, a relative increased risk of 1.3 (the most conservative meta-analysis estimate) shows that approximately one in four cancers in PPI users is attributable to PPIs and 3%–8% of all gastric cancers in the total population (online supplemental 5). These estimates should be interpreted with extreme caution as these are based on imprecise global estimates, and large differences are expected by population characteristics and underlying pathophysiology.

Trends and cohort effects of gastric cancer in Sweden

Gastric cancer incidence in Sweden has decreased dramatically since 1970, a trend not markedly affected by the introduction of H. pylori eradication and PPI treatment around 1990 (figure 4a). This decrease reduced ten years later (~2000). When looking at the youngest group (<40 years), this decrease between 1970 and 2000 was not as apparent, and in men, the incidence even started to increase around the early 2000 (figure 4b). Gastrointestinal cancer has overall increased since 1970, although there was a decreasing trend between 1985 and 2000. This was in both sexes, and also in the youngest group, with the steepest increase in men (figure 4c,d). The age-related gastric cancer incidence (figure 5a,b) was generally higher in 1970 than in later years, particularly in the older age groups. The 2020 incidence in the highest age groups is lower than previous time periods, yet not consistently for the younger groups (<60 years) in both sexes. When looking at birth cohorts (figure 5c,d), the youngest cohort, born in 1975, did not have the lowest incidence in the young age categories, with obviously no data available yet for older ages.

Figure 4. Trends in (a–b) gastric and (c–d) gastrointestinal (GI) cancer (1970–2022) and a proton pump inhibitor (PPI) use (2006–2023) in Sweden by sex, in all ages and in those younger than 40 years old.

Figure 5. (a–b) Incidence of gastric cancer in Sweden (1970–2022) by age and period and (c–d) by age and birth cohort, subdivided by sex.

Discussion

This study shows a vast and growing body of epidemiological research investigating the association between PPIs and gastric cancer, with all but one (underpowered) meta-analyses showing a 1.3–2.9 times increased risk among PPI users compared with non-users.

Our systematic search was based on three valid databases, and thorough back-and-forward citation tracking, through which we did identify some original studies not identified in published evidence syntheses. As two of the included meta-analyses were published in Chinese, we could not assess the included results (reported in one of the umbrella meta-analyses¹⁰⁹). The 28 included observational studies presented a large variety in study methodology, including the study settings, definition of PPI exposure, the included population (specific indications to population-based), duration of follow-up, anatomical locations (cardia, non-cardia) and others. All identified studies originated from highly affluent countries in Northern America, Europe (mainly Northern Europe) and Asia. Only adults or elderly individuals were investigated, and several studies were clearly underpowered (low number of individuals and/or too short follow-up). The large variation in the PPI definition was challenging, ranging from a few days of exposure/one prescription to years of (presumed) continued exposure. We acknowledge the challenges of investigating this association in regions lacking patient and drug registers, as both compliance and follow-up need to be monitored (or reliably estimated) over prolonged periods of time. Both PPI exposure and cancer diagnoses must be sufficiently reliable and valid. The absence of suitable comparators used for the exact same indications is also challenging with H2-receptor antagonists being the most reported comparator, although also reported to be associated with adverse health effects including gastric cancer.¹²⁴ PPIs clearly dominate the gastric antacid market, with up to five times more PPI users than H2-receptor antagonists in the UK⁵⁷ and even 40 times more in Sweden.^{8 125} Some studies used an active comparator design (H2-receptor antagonists).^{48 57 58 66 69 70 78} Of these included original studies, five showed higher effect sizes between 1.15 and 1.30 among PPI users than H2-receptor antagonists (although not significant in three studies).^{48 66 70} The remaining two studies did not show any association.^{58 78} According to one included 2020 meta-analysis,⁹⁴ H2-receptor antagonists also showed an increased risk of gastric cancer RR=1.32 (95% CI 1.08 to 1.62, five studies), compared with RR=1.53 (95% CI 1.13 to 2.07, seven studies) for PPIs. The older included meta-analysis (2013) showed similar effect sizes for H2-receptor antagonists (OR=1.40; 95% CI 1.24 to 1.59, 10 studies) and PPIs (OR=1.39; 95% CI 1.19 to 1.64, three studies).⁸⁶ Despite the methodological heterogeneity and remaining gaps of knowledge, we do present the most comprehensive overview of the available epidemiological evidence regarding the association between PPIs and gastric cancer. There is also the possibility of publication bias, with negative or unexpected findings being less likely to be accepted for publication. While all included RCTs reported null findings, most included meta-analyses which assessed this mathematically did not find statistical evidence for publication bias.^{86 94 96 97 99 100 103 106} Only three meta-analyses suggested some asymmetry,^{87 95 98} yet correcting for this did not change the positive association.⁹⁸

When addressing Swedish gastric cancer trends, the incidence was already decreasing markedly before the introduction of H. pylori eradication (and PPI commercialisation). Improved sanitation and fresh food have been described as important contributors to a decreasing global prevalence of H. pylori over time.¹²⁶ Yet, the overall widespread antibiotic consumption for other indications than H. pylori may also have reduced H. pylori prevalence as an unintended side effect, yet this hypothesis seems less likely regarding the emerging antimicrobial resistance—basically since the first antimicrobials were introduced.¹²⁷ To see a marked diversion in the trend in gastric cancer already around 1990 would have been unlikely, as cancer progression takes time, and the risk will not disappear instantly. Yet, as H. pylori eradication has shown to reduce gastric cancer,^{126 128} we would have expected to see this reflected on population level gastric incidence rates. On the contrary, this decreasing trend slowed down approximately 10 years later (+- 2020) and even increased in men younger than 40 years. The global reduction of H. pylori prevalence over time should especially have benefited younger cohorts regarding gastric cancer incidence, as fewer should have been infected, and progress into chronic infection and atrophic gastritis.¹²⁶ When looking at the age cohort period effects, the risks in older age groups (>60 years) have decreased since 1970, but this is not the case for younger groups and more recent birth cohorts. Although we do not know how these trends will develop over the next few decades, these results seem in line with our hypothesis that earlier exposure in life to PPIs may have a stronger effect on gastric cancer risk—as all generations now older than 60 could not have been exposed to PPIs before young adulthood. These potential cohort effects need further exploration in other settings with population-based registries enabling long-term follow-up for cancer, particularly if nationwide prescribed drug registries were established earlier like in Denmark (1995) and Finland (1994).¹²⁹

It remains to be explored how long potential effects of PPIs and long-term gastric acid inhibition may linger after treatment cessation, which complicates assessment of reverse causality. Yet, this might be several years based on families with hereditary mutations resulting in non-functioning proton pumps.¹³⁰ Yet, latency time may be shorter in those with H. pylori-related gastric atrophy.⁶⁰ Although it remains unclear what this means for our health, it has also been shown that PPIs do have a stronger effect on the gut microbiome than antibiotics on population level,⁴⁴ and antibiotics may already leave an imprint in the gut microbiome up to 8 years after exposure.¹³⁰

When weighing the evidence using the Bradford-Hill criteria, we argue that the evidence is quite convincing for a potential causal association between PPIs and gastric cancer, with ‘small (~OR=1.5) to medium (~OR=2.5) effect sizes’.^{131 132} The findings in clinical and epidemiological studies are also consistent with animal studies, in which profound acid inhibition induces gastric neoplasia in all species examined.¹³³ A limitation of the larger-scaled human studies, particularly if based on real-world data collection, is that gastric acidity (as measured by gastrin or chromogranin A) and even H. pylori status were not sufficiently and consistently explored and documented for individuals exposed to PPIs and their controls.

In 1994, H. pylori was categorised as a Class I carcinogen for gastric cancer,¹²⁶ and we question if PPIs should become listed as a ‘possible’ carcinogen. As H. pylori negative gastric is rare (<5%), it may be unlikely that PPI exposure by itself is a sufficient, single cause of gastric cancer, and the risk will also be affected by other confounding factors including diet, obesity, physical activity and smoking.^{39 134} It seems unlikely that confounding by indication can fully explain this association, as several included studies used an active comparator design^{48 57 58 66 69 70 78} stratified,^{8 49 57 61 64 71 77} adjusted^{77 78} or restricted^{63 66 67 70 74 75 77} by indication. Increased risks of gastric cancer were also found in groups only using PPI as ulcer prophylaxis.^{8 74}

PPIs may create a cancer-promoting micro-environment for H. pylori-induced changes to evolve more rapidly, pushing the patient up on the risk scale. Previous mechanistic research showed that reduced gastric acid production in individuals with H. pylori allowed the bacteria to expand their territory within the stomach and changed the distribution of gastritis.^{135 136} Prolonged gastric acid inhibition also hinders recovery of gastric atrophy after H. pylori eradication, with gastrin playing a key role in the regulation of gastric acidity.^{60 137} Similar mechanisms may also play for other gastric acid inhibiting drugs, such as the newer and even more potent vonoprazan.¹³⁸ The effects of vonoprazan cannot yet be assessed on a population level in Sweden as it was not yet recorded during the study period. Our earlier PPI utilisation study showed other anti-acid drugs became rare in Sweden (<1% of all prescriptions in 2023 of the A02B containing all drugs to treat gastro-oesophageal reflux and peptic ulcers), hindering the assessment of population-based long-term carcinogenic effects.¹²⁵

Even if PPIs are not independent carcinogens, we know PPIs are overprescribed (up to 70%), often as maintenance therapy, while the large majority of indications warrant short-term use. Yet, many physicians globally fail to re-assess the need for treatment, contributing to polypharmacy and potential drug-drug interactions¹³⁹ and unwarranted long-term health effects.¹³¹⁵,^{18 139} We do acknowledge PPIs can be life-saving drugs (eg, haemorrhaging peptic ulcers) and may improve quality of life, at least temporarily.^{35 110} Yet, PPI utilisation is still increasing globally and also in Sweden, among all age groups.^{125 140} Although the stomach is not as acidic during foetal life and the first months of life, we know insufficiently how and how long PPIs and other drugs impact the many physiological processes during early life.¹⁴¹,¹⁴⁵ Most research regarding gastric and other cancers is, however, restricted to adult and elderly populations. Although differences by sex and/or age were only explored in two of the included studies,^{8 73} we do worry about potential longer-term effects in younger individuals, particularly children (antenatal and early-life exposure)¹⁴³¹⁴⁵,¹⁴⁸ and young adults.¹⁴⁹ Our own recent study on antenatal and early-life exposure to antibiotics and PPIs did suggest that PPI use before age two might be associated with an overall increased risk of childhood cancer, although power was insufficient to address specific cancer types.¹⁴⁷ Even if gastric cancer incidence is decreasing globally (mainly attributed to the eradication of H. pylori)¹¹⁸, the incidence seems to increase in younger adults in low-incidence regions, and it is not ruled out yet if PPIs may be involved.¹¹⁹ Another issue specifically for younger individuals is the potential delayed diagnosis if PPIs are initiated for vague gastrointestinal symptoms without alarming symptoms such as weight loss. Besides the lower suspicion for cancer in younger individuals, there may be age cut-offs for urgent referral for upper gastrointestinal endoscopy. The American and Canadian Colleges of Gastroenterology guidelines suggest referral for gastroscopy in case of new onset dyspepsia from 60 years and above to exclude gastro-intestinal malignancy.¹⁵⁰ For those under age 60, it is only recommended to have non-invasive assessment of H. pylori but no assessment for malignancy, although family history and regional gastric cancer prevalence may be considered.¹⁵⁰ The National Institute for health and Care Excellence (NICE) guidelines recommend urgent referral for endoscopy in case of dyspepsia/reflux if they have dysphagia (all ages) or are at least 55 years old presenting with weight loss.¹⁵¹

The clinical implications of this study may be limited, and many scientific questions require further exploration. Yet, we do plead for better PPI stewardship because of the widespread documented over-utilisation of PPIs.^{125 152} Gastric cancer is not the only outcome linked to prolonged exposure to PPIs, and benefits of any medical treatment should be weighed against potential short- and long-term harm, considering alternatives including a healthier lifestyle and diet. Major open questions consider a potential safe dose and duration of PPI use (threshold effect?) regarding gastric cancer risks and other potential adverse events and what the effects are of prolonged exposure at a younger age (children, young adults, pregnancy). It does seem unlikely that short-term acid inhibition affects gastric carcinogenesis, yet it remains unclear how we could define a safe duration and accumulated dosage as there may be individual differences in susceptibility, and other adverse events have been associated with PPI exposure. Sex differences remain largely unexplored, while we do know gastric cancer incidence, time of diagnosis and prognosis differ.^{39 153 154} Efficacy and safety of drugs, including PPIs, may also differ drastically between sexes with a likely over-utilisation in women.¹⁵⁵ We do suspect the effects, but also the risk-benefit balance, differ by underlying risk factors (age, sex, family history, unhealthy diet, alcohol consumption, smoking, lifestyle, presence of premalignancy) and by region, as both PPI consumption and prescription practices, and gastric cancer incidence vary.

Conclusions

PPIs were associated with a 1.3 to 2.9 times increased RR of gastric cancer with all but one out of 22 meta-analyses providing statistically significant evidence. Considering the established role of H. pylori in gastric cancer development, it seems unlikely that PPIs are an independent, single-cause, carcinogen, but enhance the H. pylori carcinogenic pathway. Even if the overall balance between short and long-term harms and benefits pleads in favour of PPI for the individual regarding gastric cancer, many other adverse consequences have been associated with PPI use. It is important to evaluate how we use PPIs and to avoid long-term use if possible. Although PPIs are mostly used in older individuals representing the largest absolute risk, the relative risk for gastric cancer (and other potential adverse health effects) might be higher among younger individuals (children and young adults). Especially early life seems an important phase, with potential health consequences lasting over decades, and even potential intergenerational effects in case of antenatal exposure during pregnancy.

Data availability statement

All data relevant to the study are included in the article or uploaded as supplementary information.