Proton-pump inhibitors and the risk of antibiotic use and hospitalisation for pneumonia

Acid-suppressant therapy represents the second leading medication therapy worldwide, with sales of $26.9 billion in 2005.1 Proton-pump inhibitors (PPIs) are the most commonly used acid suppressants.

While PPIs are considered relatively safe, recent case–control studies suggest their use may be associated with an increased risk of respiratory tract infections2 and community-acquired pneumonia in adults3,4 and children.5 Suppression of gastric acid resulting in colonisation by ingested pathogens is considered a possible causal factor.4

The incidence of first episode of community-acquired pneumonia has been reported to be 2.45 per 100 person-years in those exposed to PPIs compared with 0.6 per 100 person-years in those not exposed.3 We aimed to confirm this association in the Australian veteran population by means of a retrospective cohort study, and to determine the overall incidence of community-acquired pneumonia by including multiple events (eg, all hospitalisations for pneumonia for each veteran).

The Department of Veterans’ Affairs (DVA) claims database contains details of all prescription medicines, medical and allied health services and hospitalisations subsidised by the DVA. The data file contains 80 million pharmacy records, 200 million medical and allied health service records and over six million hospital records for a treatment population of 310 000 veterans. The DVA maintain a client file, which includes data on sex, date of birth, date of death and family status. Date of death is sourced from family notifications, death notices and the Australian Government Births, Deaths and Marriages registries. Medicines are coded according to the World Health Organization anatomical and therapeutic chemical (ATC) classification6 and the Schedule of Pharmaceutical Benefits item codes.7 Hospitalisations are coded according to the WHO International classification of diseases, 10th revision (ICD-10).8

This historical cohort study included veterans aged 65 years or over at 1 January 2002, who had at least one medication prescribed in the 6 months before entry into the cohort and who were Gold Card holders (indicating they were eligible for all health services subsidised by the DVA). Veterans were excluded if they had been dispensed a histamine-2 receptor antagonist (H2RA) in the 6 months before the study. Subjects who began taking H2RAs after study entry were censored at the time of their first prescription of an H2RA.

Dosage information is not available in the dataset, so exposure to PPIs was defined as 36 days from the prescription dispensing date. This period was calculated from the data, and represents the period within which 75% of veterans return for a repeat dispensing of a PPI. During the 36 days after any dispensing of a PPI, veterans were considered exposed; at any other time they were treated as not exposed for the purpose of the analysis. Repeat dispensings within 36 days were considered as continuous exposure. Subjects not exposed to PPIs were followed from entry into the cohort until death or the end of the study on 30 December 2006.

The primary endpoint was any hospitalisation with a primary diagnosis of pneumonia (ICD-10 codes: J12–J19, inclusive). Secondary endpoints included:

• any hospitalisation with a primary diagnosis of bacterial pneumonia (ICD-10 codes: J13, J14, J15, J16);

• any dispensing of antibiotics commonly used for respiratory tract infections — amoxicillin (ATC code: J01CA04), amoxicillin with clavulanic acid (J01CR02), cefaclor (J01DC04), cefuroxime (J01DC02), erythromycin (J01FA01), roxithromycin (J01FA06), doxycycline (J01AA02), ciprofloxacin (J01MA02), oral moxifloxacin with gatifloxacin (Pharmaceutical Benefits Scheme codes: 08636M, 04329W, 04297E). Antibiotics dispensed as part of Helicobacter pylori eradication therapy were not included.

Hospitalisation rates were calculated as the cumulative number of hospitalisations in the exposed or unexposed period divided by the number of days at risk. Rate ratios were calculated with Poisson generalised estimating equations (GEE) to allow for clustering of patients, adjusting for age at entry into the cohort, sex, socioeconomic index of disadvantage for postcode of residence9 at study entry, number of comorbidities (assessed annually by means of the validated RxRisk-V10 score), season, residential aged-care status, use of tiotropium as a proxy indicator of chronic obstructive pulmonary disease, use of renin–angiotensin system medicines concurrent with frusemide as a proxy indicator of heart failure, number of prescriptions, number of prescribers, number of pharmacies, occupational therapy visits and speech pathology services. These last five categories were included as additional proxy measures of frailty and health status, and updated annually. Our data tables were constructed as day-by-day events, enabling covariates to be adjusted on the date at which they were first detected. Socioeconomic index scores were categorised as low at 964.1 or below, low–medium from 964.2–1001.1, medium–high from 1001.2–1057.5, and high at 1057.6 or above. All analyses were performed using SAS, version 9.12 (SAS Institute, Cary, NC, USA).

The final cohort consisted of 185 533 veterans, of whom 58% were men. Their mean age was 79.4 years (SD, 5.2 years). In the year before the study, the median number of prescriptions for all medicines dispensed per year was 36. The median number of prescribers was three and the median number of pharmacies was two. At study entry 6.8% of veterans were resident in aged care.

Unadjusted risk estimates were significant for all endpoints. After adjustment, there was a small, significant increased risk of hospitalisation for pneumonia and antibiotic use among those exposed to PPIs. No increased risk was observed for bacterial pneumonia (Box 1).

Subgroup analysis by age and index of socioeconomic disadvantage found the youngest cohort, aged 65–74 years, with either a low or a high index of disadvantage score had the highest risk estimates for hospitalisation for pneumonia for those exposed to PPIs (Box 2).

Our cohort study confirms the increased risk seen in previous case–control studies of both antibiotic use and hospitalisation for pneumonia associated with use of PPIs. The overall increased rate ratio (RR) observed in our study (RR, 1.16; 95% CI, 1.11–1.22) for pneumonia is in keeping with findings observed in the Dutch population (RR, 1.9; 95% CI, 1.4–2.6)3 and the Danish population (RR, 1.5; 95% CI, 1.3–1.7).4 There is some evidence that the increased risk of pneumonia is highest within the first 7 days of therapy (odds ratio [OR], 5.0; 95% CI, 2.1–11.7), with risk still present but diminishing over time (OR, 1.3; 95% CI, 1.2–1.4).4 Our study, which included ongoing exposure and repeat events, found risk levels more similar to those reported among continuous users. Only one other study assessed use of antibiotics, among a sample of 405 patients, based on patient self-report, and found a fourfold increase in risk.2 We also found increased risk in our much larger study, although the risk estimate was more conservative. When analysed by age and index of socioeconomic disadvantage, we found the highest risk estimates among veterans aged 65–74 years who had either high or low disadvantage scores. The only other study to report risk by age group also reported higher risk estimates in the younger age groups.4

We included bacterial pneumonia as a secondary endpoint, because relying on administrative data assumes the diagnoses coded are correct whereas some could be misclassified, and bacterial pneumonia is potentially a more robust indicator of pneumonia cases. Unlike previously published results,4 we did not find a significant association with this endpoint. However, only a small number of cases were coded as bacterial pneumonia; most were coded as pneumonia unspecified. Only 5.2% of people presenting to Australian general practice with pneumonia are referred to hospital,11 which suggests our study may have under-estimated the incidence of community-acquired pneumonia.

While the overall increased risk of pneumonia and hospitalisation with the use of PPIs was small, the prevalent use of PPIs means that many people are potentially affected. In 2005, about one million Australians were dispensed PPIs;12 based on the incidence estimates from our study, this suggests a potential 300 000 extra antibiotic prescriptions and 4000 extra hospitalisations for pneumonia annually associated with PPI use. The estimated 30-day fatality rate among older people admitted to hospital with community-acquired pneumonia in Australia is 18%.13

Our study relied on administrative claims data, so we were unable to adjust for confounders such as obesity, smoking and alcohol use. However, previous studies have shown that the odds ratio did not change after adjustment for these factors.2,4 Unmeasured confounders may have influenced the effect we observed, but one previous nested case–control study compared previous users of acid-suppressive medicines with current users to minimise the likelihood that the result was confounded, and found a level of risk3 similar to that in our study.

In the absence of dosage information, we used 36 days as the exposure period. PPIs are supplied under the Pharmaceutical Benefits Scheme in packs to last 30 days or less. For regular users of PPIs, the exposure times are reasonable allowing for a small number of omitted doses. Veterans who used the medicines intermittently may have been misclassified, as may those who did not take the medicines. However, on average, veterans are dispensed 9.3 prescriptions per year for PPIs, suggesting that most use is continuous.

Current guidelines for PPI use recommend step-down or intermittent therapy where maintenance therapy is required.14 Our findings in this study are only observational, but they do suggest that a more cautious approach to PPI therapy that adopts the lowest dose wherever possible may be in the best interests of patients.