Introduction

Antithrombotic drugs, nonsteroidal anti-inflammatory drugs (NSAIDs), low-dose aspirin, and other antiplatelet agents are known to induce upper gastrointestinal (GI) bleeding (UGIB) [1, 2]; however, these drugs also cause injury in the small intestine and the large intestine [3–5]. Use of proton-pump inhibitors (PPIs) is an effective and preventive treatment for UGIB [6, 7], and they have been widely used for treating gastroesophageal reflux disease and functional dyspepsia, as well as high-risk GI injury related to antithrombotic drugs [8]. However, the protective and adverse effects of PPIs in lower GI tract injury remain unknown.

The findings of a recent video capsule endoscopy study suggest that PPI use increases the risk of aspirin-induced small intestinal damage [9]. Since enterobacteria and cytokines are of pathogenetic importance in NSAID-induced enteropathy, acid suppression by PPIs has the potential to cause lower GI tract injury through bacterial overgrowth [10–12]. Apart from PPI-induced-small intestinal damage, PPIs may cause lymphocytic infiltration or a marked thickening of the subepithelial collagen layer of the colonic epithelium (i.e., microscopic colitis) that can be exacerbated by the use of antithrombotic drugs [13, 14]. Furthermore, PPIs are known to interact with clopidogrel and warfarin by affecting their metabolism, which can increase the risk of GI events [15–19]. A population-based study conducted in Western countries indicated a decreasing trend in the incidence of upper GI tract events but an increasing trend in the incidence of lower GI tract events with PPI therapy [20]. Therefore, although there has been increasing concern over the effects of PPI use on the lower GI tract, the association between PPI use and lower GI bleeding (LGIB), a serious GI event, has not been fully investigated in large-scale observational studies. Moreover, no data are available on the effect of PPI–antithrombotic drug interactions with regard to LGIB risk.

In this study, we prospectively investigated colonoscopy as a first-line modality for diagnosing LGIB by excluding UGIB mainly with upper endoscopy and by assessing the use of PPIs with antithrombotic drugs on the day of colonoscopy. The specific objectives of this study were to determine the effects of PPI use on LGIB and to explore whether PPIs affect interactions between LGIB and the use of NSAIDs, low-dose aspirin, clopidogrel, or warfarin.

Methods

Study design, setting, and participants

We conducted a prospective, single-center, case–control study of adults who underwent colonoscopy between August 2010 and April 2014 in the Endoscopy Unit of the National Center for Global Health and Medicine (NCGM). NCGM has 900 beds and is the largest emergency hospital in metropolitan Tokyo, Japan. The Institutional Review Board of NCGM approved this study. The method and estimates of the odds ratio (OR) associated with agents that increase the risk of LGIB have been reported elsewhere [5].

Cases

The criteria for case inclusion were as follows: (1) more than 18 years old, (2) Japanese nationality, (3) outpatient onset of acute, continuous, or frequent LGIB, and (4) emergent hospitalization (not in-hospital bleeding). During the study period, 430 patients were eligible as cases; however, we excluded some patients for the following reasons: (1) unknown use of medications (n = 19), (2) not being independent in daily life activities (n = 3), (3) not able to understand written documents (n = 2), (4) middle GI tract bleeding (n = 13) [21], or (5) LGIB with bleeding of unknown origin or hemorrhoidal bleeding that did not meet the criteria for clinically significant upper and lower GI tract events (n = 69) [22]. Duplicate data were allowed in the exclusion criteria. This left 355 patients with LGIB for analysis.

Controls

The criteria for the inclusion of community control subjects were as follows: (1) more than 18 years old, (2) Japanese nationality, (3) no apparent episode of GI bleeding within the past 1 year, (4) admission to the hospital or outpatient attendance for various reasons (e.g., functional dyspepsia or irritable bowel syndrome), and (5) treated by a general practitioner in the same geographic area covered by the hospital. During the study period, 11,867 patients who underwent colonoscopy (59 % underwent upper GI tract endoscopy during same period) were eligible as controls. However, we excluded patients for the following reasons: (1) unknown use of medications (n = 193), (2) could not complete questionnaires owing to lack of time or refusal (n = 3,201), (3) not being independent in daily life activities (n = 72), or (4) having ulceration or a hemorrhagic lesion that involved the stomach, the duodenum, or the small or large intestine, as detected by endoscopy (n = 420). Duplicate data were allowed in the exclusion criteria. This left 8,221 patients without any evidence of GI bleeding for analysis as controls.

Diagnostic criteria for LGIB

An electronic high-resolution video endoscope (model CFH260; Olympus Optical, Tokyo, Japan) was used after sufficient bowel preparation to diagnose colorectal disease. LGIB was defined as bleeding from a source distal to the ileocecal valve found on endoscopy [21]. The sources of bleeding were a vascular lesion or a nonvascular lesion found on colonoscopy. Vascular lesions were defined by the presence of active bleeding, an adherent clot, or a visible vessel [23, 24], and were identified in cases of diverticular bleeding, radiation telangiectasia, angiectasia, rectal ulcer, and hemorrhoid as well as after polypectomy/endoscopic mucosal resection. Nonvascular lesions were defined as the presence of inflammation, such as colitis or inflammatory bowel disease, or the presence of a tumor, such as colorectal cancer. The diagnostic criteria for diverticular bleeding were divided into definitive and presumptive bleeding [23, 24]. A definitive diagnosis was based on colonoscopic visualization of a colonic diverticulum with stigmata of recent hemorrhage such as active bleeding, an adherent clot, or a visible vessel [23, 24]. A presumptive diagnosis was based on the presence of (1) fresh blood localized to the colonic diverticula in the presence of a potential bleeding source on total colonoscopy, (2) bright red blood in the rectum confirmed by objective color testing and colonoscopy, demonstrating a single potential bleeding source in the colon and complemented by negative findings on upper GI tract endoscopy or negative findings on capsule endoscopy, or (3) negative findings on nasogastric lavage [23, 24]. We performed upper GI tract endoscopy in 116 patients and capsule endoscopy in 49 patients during the same hospitalization. Overt LGIB of unknown origin or hemorrhoidal bleeding was defined as a clinically significant drop in hematocrit of 10 % or more and/or in hemoglobin concentration of 2 g/dL or more from baseline levels according to previous criteria [22]. Patients who did not meet these criteria were excluded from the study.

Data sources and measurement

After informed consent had been obtained, a detailed questionnaire was completed by staff during a face-to-face interview with each patient in the Endoscopy Unit on the same day as the colonoscopy. In addition, medical research staff blinded to both the bleeding outcomes and the colonoscopy results cross-checked patient medical records for unanswered questionnaire items to avoid omissions [5, 25, 26]. Patients were asked to indicate which drugs, if any, they had used on the basis of drug pictures provided on the questionnaire [5, 25, 26]. Use of a drug was defined as intermittent or regular oral administration within 1 month prior to the interview. Patients were asked about their use of eight types of NSAIDs (loxoprofen, diclofenac, naproxen, etodolac, zaltoprofen, meloxicam, lornoxicam, and celecoxib), two kinds of asprin agents (enteric-coated aspirin and buffered aspirin), nine types of non-aspirin antiplatelets (ticlopidine, clopidogrel, cilostazol, dipyridamole, sarpogrelate hydrochloride, ethyl icosapentate, dilazep hydrochloride, limaprost alfadex, and beraprost), three kinds of anticoagulants (warfarin, dabigatran etexilate, and rivaroxaban), acetaminophen, and oral corticosteroids. PPI therapy included treatment with omeprazole, esomeprazole, lansoprazole, or rabeprazole; all doses of these drugs were considered as PPI use. All patients with a history of taking aspirin, non-aspirin antiplatelets, anticoagulants, or PPIs were regular users (more than 1 month). Comorbid conditions were calculated according to the Charlson comorbidity index [27], a weighted scoring index.

Statistical analysis

Because no studies have explored the association between LGIB and PPI use to date, we calculated the case number (LGIB cases and non-LGIB cases) for this study on the assumption that many of those exposed to oral NSAIDs would also have received coadminstered PPIs. Previous studies investigating the association between LGIB and use of NSAIDs reported that approximately 20 % of the control group was exposed to NSAIDs [4, 28–30]. When an OR of 2.0 is assumed for detection, the expected rate of exposure to NSAIDs in LGIB cases is approximately 33 %. To calculate the differences in a case–control study with a 1:4 case–control ratio and an α value of 5 % at a power of 80 %, there need to be 126 cases (LGIB) and 504 controls (non-LGIB), which was fulfilled by our study population. Patient characteristics were compared between LGIB and non-LGIB patients using Pearson’s chi square test, Fisher’s exact test, or the Mann–Whitney U test, as appropriate. We used logistic regression analysis to compute the OR and 95 % confidence interval (CI) as an estimate of an LGIB event associated with drug exposure. In multivariate analysis, we included the factors that showed a significant association with LGIB in univariate analysis—namely, age, sex, Charlson comorbidity index score, and NSAID or aspirin therapy. As confounders, we included factors that showed a significant association with PPI use in univariate analysis—namely, alcohol drinking, smoking, and therapy with clopidogrel, warfarin, acetaminophen, or corticosteroids. We further investigated interactions between PPIs and NSAIDs, PPIs and aspirin, PPIs and clopidogrel, and PPIs and warfarin and their effect on LGIB by using a multivariate logistic model adjusted for the aforementioned confounding factors. P values less than 0.05 were considered significant. All statistical analysis was performed using Stata, version 13 (StataCorp, College Station, TX, USA).

Results

During the study period, 355 patients with LGIB and 8,221 patients without GI bleeding met the inclusion and exclusion criteria and were subject to analysis. Endoscopy revealed various diseases in the lower GI tract (Table 1).

Table 1 Sources of lower gastrointestinal bleeding (n = 355)
Full size table

Patient characteristics are shown in Table 2. Factors significantly associated with LGIB were age more than 65 years, Charlson comorbidity index score of 2 or greater, and the use of NSAIDs, aspirin, non-aspirin antiplatelets (including clopidogrel), or anticoagulants (including warfarin). Therapy with omeprazole was associated with LGIB, but therapy with other PPIs showed no association.

Table 2 Characteristics of patients with or without lower gastrointestinal bleeding (LGIB) (n = 8,576)
Full size table

The characteristics of PPI users and nonusers are shown in Table 3. PPI use was significantly associated with age more than 65 years, male sex, being a current alcohol drinker, Charlson comorbidity index score of 2 or greater, and concomitant use of NSAIDs, aspirin, clopidogrel, warfarin, acetaminophen, or corticosteroids. Current smoking was marginally associated with PPI use.

Table 3 Characteristics of patients with lower gastrointestinal bleeding with or without proton-pump inhibitor (PPI) use (n = 8,576)
Full size table

Crude and adjusted ORs for LGIB associated with PPI use are shown in Table 4. Multivariate analysis revealed no significant associations between PPI use and LGIB. In the interaction model (Table 5), no significant interactions were observed between PPIs and NSAIDs (interaction OR 1.40; 95 % CI 0.75–2.63; p = 0.293), aspirin (interaction OR 1.09; 95 % CI 0.62–1.93; p = 0.767), clopidogrel (interaction OR 0.99; 95 % CI 0.36–2.71; p = 0.985), or warfarin (interaction OR 1.52; 95 % CI 0.58–4.02; p = 0.398).

Table 4 Odds ratio (OR) for lower gastrointestinal bleeding associated with proton-pump inhibitor (PPI) use and with the type of PPI (n = 8,576)
Full size table
Table 5 Odds ratios (OR) for lower gastrointestinal bleeding associated with proton-pump inhibitor use among nonsteroidal anti-inflammatory drug (NSAID), low-dose aspirin, clopidogrel, and warfarin users (n = 8,576)
Full size table

Discussion

In this study, we explored the effects of PPI use on LGIB risk, particularly related to the concomitant use of antithrombotic drugs. Multivariate analysis revealed no association between PPI use and LGIB, and this finding remained unchanged irrespective of the type of PPI—namely, omeprazole, esomeprazole, lansoprazole, or rabeprazole. Although PPIs are known to exhibit important interactions with antithrombotic agents [15–19], we found no significant interactions between PPI use and concomitant use of NSAIDs, aspirin, clopidogrel, or warfarin that affected the development of LGIB. Our findings suggest that PPI use has little influence on bleeding outcomes in lower GI tract events, even among patients considered to be at a high risk of GI bleeding, such as those receiving antithrombotic therapy.

PPIs are potent drugs that profoundly suppress gastric acid secretion. Gastric acid can kill most bacteria, and long-term suppression of gastric acid can lead to bacterial overgrowth in the small intestine [31, 32]. Recent studies have indicated the apparent importance of bacterial overgrowth in the pathogenesis of lower GI tract injury related to the use of NSAIDs or aspirin. In an experimental study, Wallace et al. [10] demonstrated that a reduction in gastric injury caused by PPI use was accompanied by a marked exacerbation of small intestinal ulceration and bleeding. Endo et al. [9] prospectively evaluated small-bowel mucosal breaks using capsule endoscopy in aspirin users and found that PPI use was an independent risk factor for the development of mild mucosal changes in the small intestine. However, other experimental studies have provided contradictory results regarding the association between small intestinal injury and PPI use [33, 34].

In routine clinical practice, GI bleeding events and outcomes—not mucosal injury—are very important, because patients with severe LGIB are usually emergently hospitalized and typically require further examination and/or blood transfusions [29]. However, the effects of PPIs on LGIB have never been studied. Casado Arroyo et al. [35] conducted a cohort study among patients receiving dual antiplatelet therapy and PPI co-therapy and demonstrated that LGIB occurred more frequently than UGIB in these patients (74 % LGIB vs 26 % UGIB). However, the number of LGIB events in their study was relatively scarce (n = 6), and they did not perform a comparison of LGIB development with or without PPI use. Thus, a prospective, large-scale endoscopy-based study was essential for identifying whether PPI use does actually exacerbate lower GI tract mucosal ulcer or LGIB.

In our study, we found that PPI use did not affect the risk of LGIB in clopidogrel users. Moreover, multivariate analysis showed no interaction between clopidogrel and PPI use for the risk of LGIB. Treatment with aspirin and clopidogrel now forms the mainstay of atherosclerotic cardiovascular disease therapy [36], and concomitant PPI therapy is recommended for minimizing antiplatelet-related GI bleeding [37]. However, several studies have suggested that PPIs reduce the antiplatelet effects of clopidogrel owing to the drug–drug interaction, which increases the risk of cardiovascular events [15, 16]. The possible competitive metabolic effects of PPIs on cytochrome P450 2C19 could inhibit the conversion of clopidogrel to its active metabolite and consequently reduce its effectiveness [15, 16, 38]. However, few studies have attempted to identify whether such combination therapy increases the risk of GI bleeding. Van Boxel et al. [17] performed a retrospective cohort study with a large number of new clopidogrel users, and demonstrated that new clopidogrel users receiving concomitant PPI therapy were at increased risk of peptic ulcer disease (n = 38, 0.7 %) compared with those not using a PPI (n = 27, 0.2 %), with an adjusted hazard ratio of 4.8 (95% CI 1.2–19.2); this was inconsistent with our findings. However, since the number of GI events in their study was small and the 95 % CI was rather wide, their results should be interpreted with caution. The reason underlying this discrepant result is unclear, but it possibly involves the different genetic polymorphisms of cytochrome P450 between Asians and Westerners related to drug metabolism that affect the GI bleeding risk [39].

In addition to the PPI–clopidogrel interaction, it has been reported that patients receiving a PPI concomitantly with warfarin have higher rates of adverse drug events compared with patients receiving warfarin alone [18]. The US Food and Drug Administration has suggested that PPIs increase the international normalized ratio when used concomitantly with warfarin because of their being metabolized by cytochrome P450 2C19. These findings suggest that concomitant use of PPIs and warfarin may facilitate bleeding. However, in our study, PPI use by warfarin users was found to have no association with LGIB, and no significant PPI–warfarin interaction was found for LGIB risk. Only a few studies have evaluated the association of PPIs and warfarin with bleeding outcomes. Hata et al. [40] evaluated cardiovascular surgery cases, and showed that PPI use was associated with postcardiotomy bleeding. They also found that lansoprazole use compared with rabeprazole use was associated with higher rates of bleeding events, including three UGIB cases and three LGIB cases [19]. However, overall bleeding events in their study were scare (n = 10) and did not appear to reach statistical power for multivariate analysis; thus, further studies involving larger numbers of patients are needed to clarify this issue.

The strengths of the present study were as follows. First, the number of GI events was large (n = 368), all subjects underwent colonoscopy along with other endoscopic examinations, which is essential for diagnosing bleeding sources in the lower GI tract [41], and strict exclusion criteria were applied in cases of bleeding of unknown origin or hemorrhoidal bleeding [22]. Second, the collection of drug information was rigorous; we used a questionnaire with photographs of the drugs, and research staff cross-checked the medical records on the day of colonoscopy [5, 25, 26]. Third, detailed data on antithrombotic, acetaminophen, or corticosteroid use along with the comorbidity score enabled us to perform multivariate analysis after appropriate adjustment. We used multivariate analyses to control for confounders as an alternative to age- and sex-matching analysis, because there is currently no evidence supporting the associations of age and sex with both cases (LGIB) and exposure (PPI use) [42].

We also recognize some limitations. First, the diagnosis of LGIB can be influenced by the timing of colonoscopy [41], but we did not assess this. Second, the multiple interview items asked in the colonoscopy setting did not provide sufficient information on drug frequency or dose or duration of NSAID exposure. Third, low dietary fiber intake, physical inactivity, and obesity are potential influencers of LGIB [43, 44], but no data on these variables were collected in the present study.

In conclusion, our findings revealed that PPI use did not lead to an increased risk of LGIB, regardless of the type of PPI used. Further, LGIB risk was not affected by PPI use, irrespective of concomitant therapy with NSAIDs, low-dose aspirin, clopidogrel, or warfarin. Thus, the PPI–antithrombotic drug interaction does not appear to be harmful or require a change in the management of acute LGIB.