Bin Xia, Man Yang, Long H. Nguyen, Qiangsheng He, Jie Zhen, Yuanyuan Yu, Mengyang Di, Xiwen Qin, Kuiqing Lu, Zi Chong Kuo, Yulong He, Changhua Zhang, Wenbo Meng, Jinqiu Yuan

Gastroenterology 2021 Dec;161(6): 1842-1852.e10

Background & Aims 

Proton pump inhibitors (PPIs) have a major impact on gut microbiome and immune function, which in turn, may increase the risk of inflammatory bowel disease (IBD). Our aim in this study was to evaluate PPI use and subsequent risk of IBD and subtypes (ie, Crohn's disease and ulcerative colitis).


This was a pooled analysis of the Nurses’ Health Study (NHS, n = 82,869), NHS II (n = 95,141), and UK Biobank (n = 469,397). We included participants with information on personal use of PPIs and free of IBD or cancer at baseline. We evaluated hazard ratios and 95% confidence intervals (CIs) with Cox regression adjusting for lifestyle factors, PPI indications, comorbidities, and other medications.


We documented 271 cases of IBD (median follow-up, 12 years) in the pooled NHS cohorts and 1419 cases (median follow-up, 8.1 years) in the UK Biobank. For both pooled NHS cohorts and UK Biobank, regular use of PPIs consistently showed a significantly positive association with IBD, Crohn's disease, and ulcerative colitis risk. Combined analyses of 3 cohorts showed that regular PPI users had an increased risk of IBD as compared with nonusers (hazard ratio, 1.42; 95% CI, 1.22–1.65; number needed to harm, 3770; 95% CI, 3668–4369). Direct comparison with H2 receptor antagonist, a less potent acid suppressor, showed that PPI use was also associated with higher IBD risk (hazard ratio, 1.38; 95% CI, 1.16–1.65).


Regular use of PPIs was associated with an increased risk of IBD and its subtypes. The findings should be interpreted with caution because the absolute risk was low and the clinical benefits of PPIs are substantial.



See “Regular use of proton pump inhibitor and the risk of inflammatory bowel disease: pooled analysis of 3 prospective cohorts,” by Xia B, Yang M, Nguyen LH, et al, on page 1842.

Proton pump inhibitors (PPIs) dramatically reduce gastric acid secretion and are widely used in the prevention and treatment of acid-related gastrointestinal diseases.1 Gastric acid secretion rapidly returns once PPIs are discontinued so these drugs are often used long-term. US data show that in 2011–2012 approximately 8% of adults had used prescription PPIs during the past 30 days. In line with these data, a Danish nationwide study reported a prevalence of PPI use of 7% in 2014, and, from 2002–2014, PPI use increased 4-fold.2 Given their common use, the safety of PPIs, especially during long-term treatment, is central. Observational studies have suggested that PPI use is associated with increased risk of several diseases, including pneumonia, bone fractures, kidney disease, myocardial infarction, dementia, Clostridium difficile infection, and other enteric infections.3 Given these adverse events and the observation that PPIs may increase all-cause mortality, this has led some to advocate a dramatic reduction in PPI prescribing,4 although other recommendations are more nuanced.5

Mechanistically, the link between PPI use and enteric infections could be mediated by gut microbial changes because PPI use is associated with lower bacterial diversity and increased abundance of oral bacteria.567 Gut microbial changes and gastrointestinal infections have both been implicated in the development of inflammatory bowel disease (IBD).8910

In this issue of Gastroenterology, Xia et al11 report on the risk of IBD in PPI users. They conducted a prospective observational study of data from the Nurses’ Health Study and the UK Biobank encompassing >600,000 individuals followed up for a median of 12 years. Using multifactor-adjusted Cox regression analysis, the authors showed that PPI users are at a 42% increased risk of IBD as compared with nonusers. The authors undertake several analytical strategies to mitigate confounding and bias, which are inherent in most observational studies. To minimize confounding, they adjusted for several potential confounders, including socio-demographic characteristics, lifestyle factors, and other medications. To minimize the risk of reverse causation (ie, symptoms of undiagnosed IBD resulting in PPI prescription), they lagged the analysis for 2 or more years. To avoid confounding by indication (ie, indications for PPI treatment rather than PPIs per se resulting in higher risk of IBD), they performed propensity score–matching and they conducted an active comparator analysis comparing the risk of IBD in patients treated with PPIs versus H2 receptor antagonists, which have similar indications but are less potent that PPIs. The results were robust and remained largely similar in all analyses. Yet, a limitation of the study, which the authors acknowledge, is the lack of information on dosing precluding examination of a dose-response relationship between PPI use and IBD. Furthermore, the absolute risk associated with PPI use is modest. Number needed to harm is 3770, meaning that when 3770 individuals are treated with PPIs for 1 year, 1 additional case of IBD is observed.

This article adds IBD to the growing list of harms that are associated with PPIs. Why then are PPIs still prescribed if they cause so much harm? The answer lies in an understanding of the strengths and limitations of modern epidemiology. The discipline of epidemiology blossomed in the 1950s and 1960s with the promise of discovering the cause of many diseases.12 Austin Bradford Hill realized making causal inferences on observational data was challenging and outlined a list of factors that would make this interpretation more or less likely (Figure 1).13 Sadly, although the citation of these factors has increased year on year,14 authors often primarily focus on biologic plausibility when discussing whether the association they report is causal. Hill’s article has been criticized,15 and some of the factors, such as coherence and analogy, have not stood the test of time. Nevertheless, strength of association, dose response, and consistency are important and often not commented on in observational studies.

Consistency is important because, for example, initial concerns that PPI therapy may increase the risk of COVID-19 infection were not born out by other studies.16 Xia et al11 cannot comment on consistency as they have made a novel observation although the use of 2 separate databases, which they meta-analyzed, partially addresses this problem. Dose response could also not be evaluated given the nature of the databases studied. The strength of the association was modest given a relative risk <2.0 and a number needed to harm of almost 4000. The reason for focusing on strength of association is that the more dramatic the effect, the less likely it is to be due to confounding factors. Xia et al11 are to be congratulated on the thorough analysis of the data evaluating whether confounding was an issue. However, this remains a problem with all observational studies as the authors acknowledge. Sophisticated statistics can help but cannot exclude that the association is due to residual confounding. The other item on Hill’s list is specificity, which was addressed by Xia et al11 with the H2 receptor antagonist comparison. Another interpretation for specificity would be that nature is complex and it is extremely rare for a medication to prevent all diseases or cause all diseases. If PPIs are associated with almost all diseases that are studied, it is likely that there is another explanation for these observations, such as sick people are more likely to suffer from other illness and sick people are more likely to be prescribed PPI therapy.17

Indeed, a double-blind, randomized clinical trial18 comparing pantoprazole with placebo over 3 years with more than 53,000 patient years follow-up found there was no association with any of the diseases described already in this editorial. Because randomized clinical trials are the most robust form of evidence and for most outcomes the 95% confidence intervals were tight, this suggests that PPIs are not as harmful as observational studies suggest. Interestingly, this trial reported a slightly higher risk of enteric infections, and this is the underlying mechanism proposed for how PPI therapy may increase the risk of IBD. Most associations for PPI and harm are likely to be residual or unmeasured confounding, whether this is also true for IBD will only be determined by further study.