We selected relevant studies published between 1 January 1990 and 21 March 2022 to assess the association between PPI and GC and CRC risk by searching PubMed, Embase, Cochrane Library, Web of Science. The search strategy appears in Supplementary Figure S1. Manual searches of references were conducted to identify other reports from a list of review articles and original research.

Studies were included if they met the following criteria: 1) case-control or cohort study; 2) the study compared at least two independent groups (i.e., PPI use and PPI non-use groups); 3) clear results with gastric or colorectal cancer; 4) literature that can directly or indirectly provide raw data to calculate parameters such as risk ratio (RR), risk ratio (HR) or odds ratio (OR), and 95% confidence interval (CI); 5) studies written in English. Articles that meet one of the following criteria are excluded: 1) duplicate publications; 2) studies for which data of interest cannot be retrieved or calculated; 3) non-clinical trials (animal experiments, etc.); 4) systematic review articles, meta-analyses, editorials, protocols and case reports. Two independent investigators reviewed study titles and abstracts and searched for studies that met inclusion criteria for a comprehensive assessment. Any disagreements at this stage were resolved by discussion and consensus with the third reviewer.

Data extraction was performed independently by two investigators, results were compared, and discussions were conducted to resolve disagreements. The following information was extracted: first author name, year of publication, region, study design, observation period, percentage of sex, size and mean age of the included population, lag time, primary and secondary outcomes. Two investigators independently assessed the methodological quality of included studies using the Newcastle-Ottawa Scale. Three areas of research were evaluated, including selection of participants, comparability of study groups, and identification of outcomes of interest, to assign star ratings representing the quality of the study (Oremus et al., 2012). The overall quality score is nine points. We consider a study to be of high quality if it has a score of ≥7 (Supplementary Table S1).

To calculate the pooled RR with 95% CI, we used the adjusted RR and 95% CI reported in each article where possible. Study weights were calculated using the inverse variance method. Due to the low incidence (<10%) of gastric and colorectal cancers, RR, OR, and HR are considered equivalent measures for risk assessment (Grant, 2014). Data analysis was performed on adjusted ratios whenever available, and we used unadjusted ratios if only unadjusted data were available. The Cochrane chi-square test was used to judge the heterogeneity of the included studies according to the p-value and I² value. Heterogeneity among studies was classified as low (I²: <25%), moderate (I²: 50%–75%), and high (I²: >75%), respectively (Higgins et al., 2003). We considered I²>50% and p < 0.05 to represent significant heterogeneity, and selected a random-effects model (DerSimonian-Laird method) for analysis (Singh et al., 2017). In sensitivity analyses, the effect of individual studies on the pooled estimates was assessed by excluding included studies one by one. Furthermore, we performed a subgroup analysis to further explore potential sources of heterogeneity, subgroup analysis of GC including study design (case-control study, nested case-control study, cohort study), events (<1,000, ≥1,000), population (≤ 10,000, >10,000, >50,000), mean age (<65, ≥65), region (North America, Europe, Asia), Newcastle-Ottawa Scale (NOS) score (<7, ≥7), lag time (<1 year, ≥1 year), year of publication (<2010, ≥2010), duration of PPI use (<1 year, 1–3 years, >3 years), GC site (adenocarcinoma, cardia, non-cardia), eradication of H. pylori infection, subgroup analysis of CRC including study design (case-control study, nested case-control study, cohort study), events (<1,000, ≥1,000), population (≤ 10,000, >10,000, >50,000), mean age (<65, ≥65), region (North America, Europe, Asia), Newcastle-Ottawa Scale (NOS) score (< 7, ≥7), lag time (<1 year, ≥1 year), year of publication (<2010, ≥2010), duration of PPI use (<1 year, ≥1 year, ≥5 years), adjusted for CRC risk (use of NSAIDs/aspirin, BMI, follow-up duration). We assessed publication bias using Begg’s test and Egger’s test. All statistical analyses were performed using Stata 14.0 (Stata Inc, University of Texas Station, United States).

We retrieved 9,381 potentially relevant records through data-base searches, of which 2,219 duplicate records were removed. 152 records were retained for full-text review after reviewing the title and abstract (Figure 1.). Eventually, 22 eligible articles were included in our analysis (Garcia Rodriguez et al., 2006; Robertson et al., 2007; Yang et al., 2007; Tamim et al., 2008; van Soest et al., 2008; Chubak et al., 2009; Poulsen et al., 2009; Lai et al., 2013; Hwang et al., 2017; Wennerström et al., 2017; Cheung et al., 2018; Niikura et al., 2018; Brusselaers et al., 2019; Babic et al., 2020; Kuiper et al., 2020; Liu et al., 2020; Seo et al., 2021; Abrahami et al., 2022a; Abrahami et al., 2022b), Lai (Lai et al., 2013) and Lee (Lee et al., 2020) studies included two independent studies. Therefore, we analyzed the association between PPI use and gastric cancer, colorectal cancer risk in 24 studies, including 13 studies on gastric cancer and 11 on colorectal cancer.

Table 1 and table 2 summarized the characteristics of the included studies. The 24 studies had a total of 8,066,349 participants. Countries and regions of origin include the United States (Chubak et al., 2009; Babic et al., 2020; Lee et al., 2020), Denmark (Robertson et al., 2007; Poulsen et al., 2009; Wennerström et al., 2017), the United Kingdom (Garcia Rodriguez et al., 2006; Yang et al., 2007; Liu et al., 2020; Abrahami et al., 2022a; Abrahami et al., 2022b), Netherlands (van Soest et al., 2008; Kuiper et al., 2020), Canada (Tamim et al., 2008), Taiwan (Lai et al., 2013; Brusselaers et al., 2019; Lai et al., 2019; Lei et al., 2021), Hong Kong (Cheung et al., 2018), Japan (Niikura et al., 2018), and South Korea (van Soest et al., 2008; Seo et al., 2021). All studies in the analysis were adjusted for different potential confounders. Helicobacter pylori infection status is an important potential confounder for gastric cancer, and three studies were considered to eradicate H. pylori (Wennerström et al., 2017; Niikura et al., 2018; Brusselaers et al., 2019). Supplementary appendix two shows the Newcastle-Ottawa Scale quality assessment of the included studies. We considered a NOS score ≥7 to be high-quality studies, with three studies rated as moderate-quality studies (scores 5–6). The mean of eligible studies was 7.3, thus indicating a high quality of included studies.

The relationship between long-term PPI use and gastric cancer risk is shown in Figure 2. Thirteen studies of 4,431,863 participants showed that long-term PPI use was significantly associated with an increased risk of gastric cancer, with an overall RR of 1.82 (95% CI: 1.46–2.29), with high heterogeneity (p < 0.05; I² = 94.9%).

The use of PPI was always associated with the risk of gastric cancer in subgroup analysis (Table 3). The risk of gastric cancer in nested case-control study using PPI was higher than that in cohort study (RR = 2.15, 95% CI: 1.71–2.70; RR = 1.27, 95% CI: 0.94–1.72). We determined the GC site by positioning according to the International Classification of Diseases (ICD) code. Stratified analysis based on the GC site showed that there was a significant positive correlation between the use of PPI and the risk of non-cardia cancer (RR = 2.75, 95% CI: 2.09–3.62) compared with cardia and adenocarcinoma (RR = 1.72, 95% CI: 1.19–2.48; RR = 2.04, 95% CI: 0.73–5.73) (Figure 3). Hierarchical analysis based on the duration of PPI showed that there was a significant trend between the duration dependent effect of PPI use and the risk of gastric cancer (<1 year RR = 1.56, 95% CI: 1.30–1.86; 1–3 years RR = 1.75, 95% CI: 1.28–2.37; >3 years RR = 2.32, 95% CI: 1.15–4.66) (Figure 4). PPI use was still positively associated with gastric cancer risk after eradication of potential confounders of H. pylori infection (RR = 3.09, 95% CI: 2.74–3.49) (Figure 5). Subgroup analysis shows that the heterogeneity can be partly explained by the study design (RR = 1.52, 95% CI: 1.19–2.48, I²< 50%), region (RR = 2.04, 95% CI: 1.58–2.63, I²< 50%), population (RR = 1.58, 95% CI: 1.36–1.83, I²< 50%) and mean age (RR = 1.56, 95% CI: 1.34–1.81, I²< 50%)and all were related to the risk of gastric cancer. However, the results of subgroup analysis are still highly heterogeneous due to the influence of potential confounding factors. Sensitivity analysis shows that the study of Wennerström (Wennerström et al., 2017) has the greatest impact on the overall results (Supplementary Figure S1). After rejecting Wennerström’s study, the RR was 1.50 (95% CI: 1.31–1.71, I² = 49.8%). The results still showed that long-term use of PPI was associated with the risk of gastric cancer. Egger’s test and Begg’s test (Egger et al., 1997) showed no publication bias in the study (p _{Egger’s test} = 0.209, p _{Begg’s test} = 0.115).

The relationship between long-term PPI use and colorectal cancer risk is shown in Figure 6 in the Supplement appendix 3. The results of 11 studies conducted on 3,634,486 participants showed that long-term use of PPI had no significant correlation with the increased risk of colorectal cancer. The overall adjusted RR was 1.22 (95% CI: 0.95–1.55), which was highly heterogeneous (p < 0.05; I² = 96.7%).

Subgroup analysis (Table 4) showed that there was no correlation between PPI use and CRC risk in more than 10,000 participants (RR = 1.74, 95% CI: 0.77–3.93). PPI use was significantly associated with colorectal cancer risk in Asian populations (RR = 1.72, 95% CI: 0.96–3.09). However, no evidence was found that PPI was associated with colorectal cancer risk in the United States (RR = 1.02, 95% CI: 0.87–1.19, p = 0.190) and Europe (RR = 1.06, 95% CI: 0.99–1.13). The duration of PPI was not significantly associated with the risk of colorectal cancer (≤1 year RR = 1.00, 95% CI: 0.78–1.28; >1 year RR = 1.18, 95% CI: 0.91–1.54; ≥5 years RR = 1.06, 95% CI: 0.95–1.17) (Figure 7). Among the 11 studies we included, only one nested case-control study (Lai et al., 2013) from Taiwan was of low quality and showed significant correlation (RR = 1.70, 95% CI: 0.95–1.55). In the subgroup analysis of high-quality studies, there was no significant correlation between the use of PPI and CRC risk (RR = 1.19, 95% CI: 0.93–1.54). Sensitivity analyses showed that the Lai (Lai et al., 2013) and Lee (Lee et al., 2020) studies had the greatest impact on the overall results (Supplementary Figure S1). After rejecting the literature for Lee and Lai, the RR was 1.11 (95% CI: 0.97–1.28), and the results still suggest that long-term use of PPI is associated with the risk of developing gastric cancer. However, the summary estimate still indicated not significantly associated with the risk of colorectal cancer. We found no evidence of publication bias in the Egger’s test and Begg’s test results (p _{Egger’s test} = 0.991, p _{Begg’s test} = 0.087).

This meta-analysis included 12 studies on proton pump inhibitors and the risk of gastric cancer. We found that PPI was associated with 82% increased risk of gastric cancer. This is consistent with the research results of Zeng et al. (2021) (RR = 1.78, 95% CI: 1.38–2.31). Compared with the latest meta-analysis (Peng et al., 2022), we excluded database duplication and specific population studies, so our research results are more robust and accurate. The results of subgroup analysis showed that the risk of gastric cancer increased significantly with the extension of PPI prescription time, and the patients who used PPI for more than 3 years had the highest risk of gastric cancer. This result may be related to hypergastrinemia. Long term use of PPI leads to hypergastrinemia through strong acid inhibition. Long term hypergastrinemia may lead to gastric cancer associated with intestinal chromaffin cell proliferation (Delle Fave et al., 1998). It is well known that gastritis caused by H. pylori infection is an important factor leading to gastric cancer. Chronic H. pylori infection may induce gastric neuroendocrine tumours under the condition of hypergastrinemia (Rais et al., 2022). Subgroup analysis showed that the use of PPI and the risk of gastric cancer increased nearly threefold after eradication of potential confounders of H. pylori infection. Long term use of proton pump inhibitors after eradication of H. pylori may create a gastric environment for N-nitrosamine formation and gastric cancer development (Kobayashi et al., 2019). One well-known role of nitrosamines is to increase the risk of gastric adenocarcinoma (Corleto et al., 2014). According to the GC site stratification, the long-term PPI users had a high risk of non-cardia cancer, while the increased risk of cardiac cancer and adenocarcinoma was not obvious. Cardiac cancer is considered to be different from non-cardia cancer in etiology. Most non-cardiac gastric cancer is associated with atrophic gastritis and peptic ulcer caused by helicobacter pylori infection (Hansen et al., 2007). This may explain the correlation between the long-term use of PPI and the occurrence of non-cardiac cancer. Cardiac cancer may be related to gastroesophageal reflux and obesity. Distal esophageal adenocarcinoma and cardiac cancer share a common precancerous background of intestinal metaplasia caused by chronic inflammation, so it is difficult to distinguish them in epidemiological studies (Imamura et al., 2021). Gastroesophageal reflux disease is one of the indications of PPI, which may cause significant confusion. The real effect of PPI on the risk of cardiac cancer may be weakened by the treatment of gastroesophageal reflux, which is a common risk factor for distal esophageal adenocarcinoma. This may be the reason why the risk of cardiac cancer is not significantly increased.

The results from 11 observational studies showed that use of PPI did not significantly increase the risk of colorectal cancer (RR = 1.22, 95% CI: 0.95, 1.55). This is consistent with Ma et al. (Ma et al., 2020) (RR = 1.26, 95% CI: 0.90–1.73). Compared with the previous meta-analysis, our study included three additional studies and more than two million additional patients. This is the most comprehensive study to investigate the relationship between PPI and risk of CRC. Seeing that the number of colorectal cancer participants from different regions and a wide range of subgroups, it is possible to conduct a more robust hierarchical analysis based on a standardized time frame that adjusted for CRC risk and PPI duration. The results of subgroup analysis showed that the risk of colorectal cancer did not increase significantly with the extension of PPI duration. This is the same as Ann et al. (Ahn et al., 2012). However, this is contrary to the findings of Abrahami et al. (Abrahami et al., 2022a). This may be due to the different characteristics of the study design and different study populations, as well as some uncontrollable confounding factors, so this result should be carefully interpreted. The nine studies included in this system review did not provide convincing evidence to prove the causal relationship between PPI use and CRC, and only two studies (Lai et al., 2013; Lei et al., 2021) found that there was a correlation between PPI and CRC. There is a research mechanism indicating that there is a biological relationship between the use of PPI and the risk of colorectal cancer. First of all, gastrin is an effective growth factor for normal and malignant gastrointestinal tissues, and has nutritional and tumorigenic effects. Secondly, long-term use of PPI will also lead to imbalance of intestinal microflora, decrease of abundance and diversity of intestinal microflora, and increase of pathogenic bacteria related to CRC carcinogenesis (Jackson et al., 2016; Si et al., 2021), which is conducive to the transformation and progress of colorectal cancer. However, there is also conflicting evidence that PPI may play an anti-tumor role through anti-inflammatory, antioxidant and antimutagenic mechanisms (Kim et al., 2010; Han et al., 2014).

It is worth noting that the early symptoms of gastric cancer are different from the late symptoms. Gastric cancer may have existed before PPI was used. Because of vague symptoms, including acid regurgitation, abdominal pain and heartburn, patients were treated with PPI before being diagnosed with gastric cancer. However, colorectal cancer does not cause reflux like symptoms. Therefore, patients with colorectal cancer usually do not need PPI treatment before being diagnosed with colorectal cancer. The significant association between the use of PPI and the increased risk of gastric cancer may indicate the contingency of clinical medication, rather than a causal relationship.

Our meta-analysis has several advantages. In this meta-analysis, the quality of the included studies was strictly evaluated, and the adjusted aggregate effect values were used to ensure the robustness of the results. Our study is the most comprehensive meta-analysis among the retrospective studies on the relationship between PPI use and the risk of gastric cancer and colorectal cancer compared with the previous systematic review and meta-analysis, including a large number of studies and participants.

There are some limitations in our study. There are still many potential uncontrollable confounding factors in our studies. The original study we included did not report the relationship between PPI dose and gastric cancer, and could not conduct subgroup analysis on specific the cumulative defined daily dose intervals. Biological gradient (dose response) is one of the important criteria for Bradford Hill to confirm causality. Our study does not meet the Bradford Hill criteria, so causality cannot be confirmed. It’ difficult to group and analysis based on the individual type of PPI, because most of the studies in these analyses do not report the type of PPI used. In addition, many of the studies we included did not provide information on over-the-counter PPI and precancerous lesions, which may affect the assessment.