Association Between Acute Gastroenteritis and Continuous Use of Proton Pump Inhibitors During Winter Periods of Highest Circulation of Enteric Viruses

IMPORTANCE An increased risk of acute bacterial enteric infections has been reported among patients receiving proton pump inhibitor (PPI) therapy. The risk of acute gastroenteritis (AGE) of viral origin associated with continuous PPI exposure has been less studied. OBJECTIVE To investigate the association between continuous PPI therapy and AGE occurrence during winter epidemic periods when the circulation of enteric viruses is the highest. DESIGN, SETTING, AND PARTICIPANTS A matched cohort study was performed using a prospectively collected drug dispensing database from a large panel of community pharmacies in continental France. All patients recorded in the database during the 2015 to 2016 winter season, with documented age, sex, and use of an identifiable regular panel pharmacy, were eligible for the study. Each patient exposed to continuous PPI therapy was matched to 3 unexposed patients, according to year of birth, sex, and identifiable regular panel pharmacy. Analyses were performed between January 2017 and December 2018. AGE epidemic was the main outcome. Episodes of AGE were identified using a previously validated algorithm based on drug dispensing data. Relative risks of AGE were estimated usingamultivariablelog-binomialmodeladjustedforage,sex,andtreatmentsforchronicconditions. RESULTS There were 233596 patients receiving PPI therapy (median [interquartile range] age, 71 [62-81] years; 55.8% female) and 626887 matched patients not receiving PPI therapy (median [interquartile range] age, 70 [61-80] years; 56.3% female) included in the analyses. At least 1 AGE epidemic episode was identified in 3131 patients


Introduction
Proton pump inhibitors (PPIs) are drugs widely prescribed to reduce gastric acidity. Although they are considered globally safe, observational studies 1-3 have reported significant adverse effects associated with their long-term use, such as osteoporotic-related fractures, vitamin B 12 deficiency, kidney disease, or infections, particularly pulmonary and digestive tract infections. Numerous studies [4][5][6][7][8] have reported an association between PPI use and bacterial enteric infections, such as isolated and recurrent Clostridium difficile infections. By reducing the secretion of hydrochloric acid produced by the stomach, PPIs may promote the growth of the gastrointestinal microflora, increase bacterial translocation, affect the gastrointestinal microbiome, and weaken the immune system. 9 The association between continuous PPI use and acute gastroenteritis (AGE) caused by enteric viruses has been less studied. To our knowledge, only 1 study 10,11 has investigated the association between PPI use and norovirus infections, but it was based on a small sample of inpatients and did not distinguish between continuous and short-term PPI therapy.
In the temperate northern hemisphere, peaks of AGE activity are observed every winter. They are mainly caused by infections with enteric viruses. [12][13][14] This study aimed to investigate the association between continuous PPI exposure and the occurrence of AGE during winter epidemics when the circulation of enteric viruses is the highest, using community pharmacy drug dispensing data issued for ambulatory care prescriptions, collected in a large French nationwide database during the 2015 to 2016 winter season.

Study Design and Data Sources
Considering the association between PPI prescriptions and patient age and the large volume of data available, a matched retrospective cohort design using prospectively collected ambulatory careprescribed drug dispensing data from community pharmacies was considered appropriate to investigate this association. 15,16 The study was conducted using the LTD (Longitudinal Treatment Dynamics) database, which was established in 2012 and contains pseudonymized prescribed drug dispensing data collected from a panel of approximately 7000 community pharmacies in continental France. The pharmacies included in the panel (referred to hereafter as "panel pharmacies") represent approximately 30% of all French community pharmacies and are representative in terms of geographical spread in continental France and covered population age. 17,18 Only drugs dispensed in association with prescriptions for ambulatory care were considered in this study. This database contains information on patients (anonymized unique identifier, year of birth, and sex), dispensed drugs (name, unique identifier, EphMRA code [a drug identification code according to the anatomical classification system developed by the European Pharmaceutical Market Research Association 19 ], dosage, volume, and pack form), dispensing events (prescribing and dispensing dates, the department of the dispensing pharmacy, and dispensed quantity), and prescribers (specialty and geographical location).
Drug dispensing data collection and processing within the LTD database were approved by the French Data Protection Authority on October 21, 2011; the approval was updated in July 2018 to comply with the European General Data Protection Regulation and applicable laws. Patient consent was not required because the data used in this study were fully deidentified. This study complies with the guidelines for Reporting of Studies Conducted Using Observational Routinely Collected Data for Pharmacoepidemiology (RECORD-PE). 20

Study Period
The study was performed using data collected during the 2015 to 2016 winter season, which was cut into 3 subperiods (Figure 1): an epidemic period (ie, a 5-week period centered around the week with the highest AGE incidence), a pre-epidemic period (ie, the 100 days before the AGE epidemic period),

Study Population
The population eligible for the study included all patients with at least 1 drug dispensing event recorded during each of the pre-and postepidemic periods, documented year of birth and sex, and an identifiable regular panel pharmacy (ie, predominantly frequented by the patient during the study period) situated in continental France. These conditions were imposed to reduce the risk of including patients who died or migrated to a nonpanel pharmacy before or during the epidemic period.

Definition of the Exposure
To ascertain exposure status, all PPI dispensing events collected during the pre-epidemic and epidemic periods were extracted from the database. All PPIs available in France were considered, including omeprazole, pantoprazole, lansoprazole, rabeprazole, and esomeprazole.
Proton pump inhibitor therapy was classified as continuous if 3 strict criteria were simultaneously fulfilled 28 : a persistence criterion implying no PPI therapy discontinuation (ie, the delay between any 2 successive dispensing events did not exceed the days' supply estimated for the first dispensing event plus a grace period of 5 days), an adherence criterion based on the medication   possession ratio (estimated as the total PPI days' supply dispensed divided by the number of days observed), and another adherence criterion based on the medication possession ratio (estimated as the total PPI days' supply dispensed, excluding the last refill, divided by the number of days between the first and the last refill). The PPI days' supply dispensed was estimated as the dispensed quantity divided by the daily dose recommended in the drug monograph for the main indication for an adult.
Each adherence criterion was fulfilled if the corresponding medication possession ratio was greater than or equal to 1. Figure 2 presents an example of how each criterion was evaluated.
Exposed patients were those receiving continuous PPI therapy initiated at least 30 days before the epidemic period (hereafter referred to as "continuous PPI users"). Patients receiving PPI therapy who did not fulfill at least 1 of these criteria were excluded from the study. Unexposed patients were those with no PPI dispensing events recorded in the database during the pre-epidemic and epidemic periods (hereafter referred to as "non-PPI users").

Definition of the Outcome
The outcome of interest was the occurrence of at least 1 AGE episode during the epidemic period. It was coded as a binary variable (Ն 1 AGE case vs none). The detection of AGE episodes from prescribed drug dispensing data was done by using an algorithm that identifies drug dispensing events corresponding to AGE treatments, taking into account patient characteristics, the delay between the prescription and drug dispensing event, and the type, number, and quantity of drugs dispensed. 29,30 The types of drugs potentially prescribed to treat an AGE episode included oral rehydration salts, antiemetics, probiotic antidiarrheals, intestinal antipropulsives, intestinal absorbents, intestinal anti-infectious agents, and antispasmodic agents. Coprescriptions of these drugs and antibiotics, antineoplastic agents used in cancer therapy, gastric antacids, drugs used for inflammatory bowel diseases, or antiemetics in injectable form were considered to treat a disease other than AGE. This algorithm and the considered drug lists have been fully described and validated in a previous study 30 that showed a strong overall agreement between the weekly AGE incidence estimated from the LTD drug dispensing database and the weekly AGE incidence estimated using clinical data collected at the Sentinelles network during winter seasons (Pearson correlation All PPI dispensing events collected during the pre-epidemic and epidemic periods (segments A to B and B to D, respectively) were extracted from the database and analyzed (segment D to E represents the postepidemic period). Assuming that 4 PPI dispensing events have been observed for this patient during these periods, the PPI days' supply (DS; DS1, DS2, DS3, and DS4; the blue areas) was estimated for each of these dispensing events. A 5-day grace period between 2 successive dispensing events (yellow areas) was allowed before the treatment was considered interrupted. Criterion 1 is the persistence criterion in which there was no PPI therapy discontinuation (ie, all yellow areas are Յ5 days). Criterion 2 is the adherence criterion in which the medication possession ratio (MPR) is greater than or equal to 1, where MPR = (DS1 + DS2 + DS3 + DS4) / the number of days observed. The number of days observed is the number of days between the first PPI dispensing event observed and the end of the epidemic period. Criterion 3 is the adherence criterion in which MPR is greater than or equal to 1, where MPR = (DS1 + DS2 + DS3) / the number of days between the first and the last refill observed. Patients whose AGE status could not be determined were excluded from the study. To limit the misclassification of long-term treatments as indicating AGE episodes, patients for whom the algorithm detected more than 4 AGE episodes during the pre-epidemic and epidemic periods were also excluded from the study.

Confounding Factors
The confounding factors considered were age, sex, and treatments for the most common chronic conditions, which were identified on the basis of disease-specific anatomical therapeutic chemical classes: diabetes (class A10), cardiovascular diseases (classes B01, C01A, C01B, C01DA, C01DX, C01E, C02, C03, C07, C08, C09, and C10), obstructive airway diseases (class R03), and conditions requiring a psychotropic treatment (class N05 and N06). The presence of these conditions was inferred if disease-specific drugs had been dispensed at least once during the pre-epidemic or epidemic period.

Validation and Sensitivity Analysis
The study was replicated for the 2016 to 2017 AGE winter season to check the reproducibility of results. Three sensitivity analyses were conducted. First, to investigate the presence of a potential selection bias related to past PPI use, the analyses were performed using a restricted population among exposed and unexposed patients: recent PPI users (ie, no PPI dispensing events recorded in the database earlier than 180 days before the pre-epidemic period) and never PPI users (ie, with no PPI dispensing events ever recorded in the database), respectively. Second, the analyses were performed using a relaxed persistence criterion: the time lag between 2 subsequent PPI dispensing events should not exceed twice the days' supply estimated for the first dispensing event plus a grace period of 5 days. This criterion accounts for the winter holidays, which may affect an individual's drug refill habits. Third, an analysis evaluating the association between AGE and histamine 2 receptor antagonists, another acid-suppression medication less potent than PPI, was performed using the same methods.

Statistical Analysis
Non-PPI users were frequency matched to the continuous PPI users by groups of exact year of birth, sex, and identifiable regular panel pharmacy, using random sampling with no replacement for a ratio of 3 non-PPI users to 1 continuous PPI user.

Results
There were 233 596 continuous PPI users and 626 887 non-PPI users included in the study, after applying inclusion and exclusion criteria (Figure 3). Among the patients eligible for the study, 28 15 to 44 years. Those age groups represented less than 5% of the exposed population identified in   Most studies investigating the association between PPI exposure and infectious gastroenteritis have focused on adult populations, which are the most exposed to PPI. Less is known about this association in the youngest populations, for whom the highest AGE incidence is usually observed. 27 The current study was performed in the overall population, including children and young adults. No association between continuous PPI use and AGE occurrence was observed in the main analysis for patients younger than 45 years. However, continuous PPI use is less common in these age groups.  They represented less than 5% of the exposed population identified in the database. Thus, the results regarding these subgroups should be interpreted with caution.

JAMA Network Open | Pharmacy and Clinical Pharmacology
Another strength of this study is the use of a strict definition of exposure, based on 3 restrictive adherence and persistence criteria. Because PPI treatment coverage was estimated according to the dose recommended in the drug monograph for the main indication for an adult, misclassification of exposure status may still occur, particularly among children, for whom dosage depends on age and weight. However, such misclassification of exposure, if any, would tend toward classifying actual continuous PPI users as occasional PPI users and excluding them from the study, rather than the opposite, which would bias toward the null.

Interpretation of Results and Comparison With Previous Studies
Results reported in this study are in line with those reported in a Swedish study 10 investigating the association between PPI use and norovirus infection in hospitalized adults (odds ratio, 1.73; 95% CI, Although the pathophysiological association between PPI use and viral enteric infections is still unclear, it has been shown that PPI use induces changes in the gut microbiome and that microbiota changes can affect enteric virus pathogenesis, suggesting a possible biological association. 11,36 A direct effect of gastric acid suppression on the risk of enteric viral infections seems plausible, because the same association was observed between histamine 2 receptor antagonists and AGE.

Research and Clinical Perspectives
Observational studies based on nationwide databases provide valuable estimates of the association between continuous PPI therapy and potential adverse events. More studies in different settings are needed to confirm the association reported here. Investigating the dose-response relationship and the pathophysiological mechanisms of the interaction between PPI use and the risk of contracting enteric viruses may help infer a causal association.
One-half of continuous PPI prescriptions are inappropriate. 37 Associations such as those reported in this study are yet another reason to reexamine unnecessary and non-evidence-based indications for PPI therapy. As regions head into enteric viral infection season, this would be an opportune time to reassess the ongoing need for PPI therapy in patients, especially older adults, and potentially deprescribe.

Limitations
The study has several limitations, mainly related to PPI exposure and AGE status ascertainment. First, diagnoses were not available, and ascertainment of the AGE status relied solely on drug dispensing data. The drugs used to treat AGE are not disease specific, and treatment indication depends on patient characteristics. To overcome this limitation, AGE episodes were identified on the basis of a previously validated algorithm. 29,30 In addition, to reduce the risk of misclassifying other chronic diseases as AGE episodes, patients for whom the algorithm detected more than 4 AGE episodes were excluded from the analysis. Fifth, information on additional potential confounders or effect modifiers (eg, sociodemographic characteristics, food consumption, potentially contagious contacts, body mass index, smoking, and alcohol consumption) were not available. Only the comorbidities whose treatment rely on disease-specific drug classes could be measured and accounted for. They were indeed more frequent in PPI users. These comorbidities could result in more-frequent contact with the health care system and, thus, a greater likelihood of receiving a diagnosis of and treatment for AGE; patients who are not as ill may not seek medical care for viral gastroenteritis. Thus, the presence of comorbidities might be a confounder of the investigated association, on which the models were adjusted. Imbalances observed between exposed and unexposed patients for the measured confounders may also be present in other unmeasured confounding factors. However, the association estimated in this study was significant (particularly among elderly patients), and it remained relatively stable after adjustment for the measured confounders.

JAMA Network Open | Pharmacy and Clinical Pharmacology
Sixth, even though patients with antibiotics prescribed along with AGE drugs were not counted as AGE cases, other antibiotic dispensing events were not taken into account. Seventh, confounding by indication cannot be ruled out, but we could not find any published evidence linking AGE risk and disorders for which PPIs are indicated (eg, gastroesophageal reflux or peptic ulcer disease).

Conclusions
Continuous exposure to PPI therapy among individuals aged 45 years and older was associated with an increased risk of developing AGE during periods of highest circulation of enteric viruses. Despite the limitations related to exposure and outcome ascertainment and potential residual confounding, the results reported in this study support the hypothesis that continuous PPI use is associated with an increased risk of infections with enteric viruses and motivate the need for further studies to confirm this association and investigate the pathophysiological mechanisms.