2.1 Data Source

We used data from 1999 to 2022 from the VHA Corporate Data Warehouse (CDW) to conduct this retrospective cohort study. The VHA is an integrated health system that provides primary and specialty medical care to Veterans of the United States Armed Services and a selection of their family members [25]. It consists of 172 medical centers and 1138 outpatient clinics that serve 9 million patients annually. CDW contains electronic health record and insurance claims data, including demographics, diagnosis codes, procedures, lab results, prescription information, and free-text documents [26, 27].

2.2 Cohort Eligibility: Acute PUD Diagnosis

Among patients meeting VHA enrollment criteria, we assembled a cohort of patients diagnosed with acute PUD. Potential patients were identified by International Classification of Diseases (ICD) codes for acute PUD from inpatient or outpatient encounters (Table S1). The date of the first acute PUD diagnosis was considered the cohort entry date. Because there is no validated case-finding algorithm for acute PUD, we implemented several criteria to optimize the likelihood of identifying patients with clinically significant acute PUD. For inclusion, patients must have had an upper endoscopy in conjunction with their first acute PUD diagnosis and not have had prior diagnoses of chronic or unspecified PUD. Furthermore, because the outcome of interest was PPI prescriptions exceeding the approved treatment duration for PUD, we excluded patients with other indications for acute or chronic PPI use, such as Barrett's esophagus, and patients who had filled a PPI prescription between 365 and 14 days before the first acute PUD diagnosis. Finally, we identified a subset of patients who filled a prescription for a PPI in the VHA within 14 days of PUD diagnosis. Patients who did not fill a PPI prescription in the VHA within 14 days were not analyzed because they may obtain their prescriptions using non-VHA benefits or they may have never initiated treatment [28, 29]. A full description and rationale of the cohort eligibility is described in Figure 1 and Data S1.

2.3 Outcome: Time to a Filled PPI Prescription Exceeding the Approved Treatment Duration

FDA-approved prescribing instructions and PUD treatment guidelines recommend that PPI treatment last no longer than 8 weeks [21-24]. As such, we classified filled PPI prescriptions that would extend treatment beyond 8 weeks to exceed the approved duration. To determine if a PPI prescription had a duration exceeding the approved length, we used 60 days as the threshold instead of 8 weeks × 7 days = 56 days because prescriptions in the United States are often written in 30-day increments. A patient's backstock of PPI from previously filled prescriptions was not considered because providers do not routinely calculate remaining stock when making prescription decisions. Because active H. pylori infection is also an indication for up to 2 weeks of PPI and is causative of some cases of PUD, we accounted for H. pylori treatment by classifying filled prescriptions that ended beyond 60 days from the acute PUD diagnosis date but within 60 days of the first H. pylori diagnosis as having approved duration. The 60-day margin after H. pylori diagnosis instead of a two-week margin was implemented to allow for delays in treatment initiation after H. pylori diagnosis. H. pylori diagnosis was ascertained from a combination of ICD codes (ICD-9041.86 and ICD-10 B96.81), laboratory results (stool antigen, urea breath test, and Campylobacter-like organism rapid urease test), and free-text surgical pathology results. Recurrent and refractory H. pylori infections were not considered.

During the study period, several societies representing various medical specialties released highly variable recommendations for prophylaxis of gastrointestinal bleeding using PPIs. In a secondary analysis to assess the maximal effect of these guidelines on long-term PPI prescribing, we considered PPI prescriptions among patients who met any of the gastroprotection guidelines as having an approved duration. Individual gastroprotection guidelines were assessed the year after their introduction and later. The gastroprotection criteria were history of PUD and NSAID use (after 2009) [30-32]; history of gastrointestinal bleeding (after 2010) [33]; antiplatelet use and any of age greater than 65, anticoagulant use, NSAID use, or aspirin use (after 2010) [33]; concurrent aspirin and antiplatelet use (after 2017) [34]; concurrent aspirin and anticoagulant use (after 2018) [35]; or two or more of aspirin, NSAIDs, anticoagulants, or antiplatelets (after 2020) [36].

2.4 Patient, Provider, and Facility Factors

Because the objective of the study was to identify healthcare factors associated with filled PPI prescriptions exceeding the approved treatment duration among patients with PUD, we assessed patient, provider, and facility factors that could influence prescribing and filling of PPIs. Patient factors included age at PUD diagnosis, sex, self-reported race, self-reported ethnicity, calendar year of PUD diagnosis, whether the PUD diagnosis was made during a hospitalization, whether the PUD diagnosis was associated with gastrointestinal bleeding, the Charlson-Deyo comorbidity score, the VA Frailty Index, healthcare fragmentation through the usual provider of care index, and the VHA priority group at PUD diagnosis (an administrative categorization that influences VHA enrollment and the cost of copays based on a Veteran's military service, disability, income, and other benefits) [37-40]. Provider factors included provider type (physician versus advanced practice provider), provider specialty, and provider subspecialty at the time of PPI prescription. Facility factors included Veterans Integrated Services Network (VISN, an organizational unit of regional healthcare delivery in the VHA) and rurality of the VHA facility where the patient's PUD diagnosis was made. A conceptual model of the relationship between these factors and the study outcome is presented in Figure S1.

As described in the Outcome subsection, we ascertained whether each patient in the cohort met criteria for several gastroprotection guidelines. To do so, we identified aspirin, antiplatelet, anticoagulant, or nonsteroidal anti-inflammatory drug (NSAID) use at each opportunity to receive a PPI prescription (specific agents in Table S2). We also identified H2-receptor antagonist use, as these may serve as substitutes for PPIs. Methodology for medication exposure ascertainment is described in Data S1.

2.5 Statistical Analysis

This analysis included three steps. First, we calculated the proportion of patients in the acute PUD cohort who filled PPI prescriptions exceeding the approved treatment duration and described the characteristics of these patients and prescriptions. We reported counts and percentages for categorical variables and medians and interquartile ranges (IQR) for continuous variables. Second, to assess which healthcare factors explain the receipt of a PPI prescription exceeding the approved treatment duration for PUD, we developed a multivariable Andersen-Gill model using time-to-recurrent-event data to calculate the adjusted hazard ratio (aHR) for each factor [41]. The acute PUD cohort entry date was the start of follow-up and the fill dates of PPI prescriptions exceeding the approved duration were the outcomes. We chose the Andersen-Gill model for recurrent events over the single-event Cox Proportional Hazard model because patients can receive PPI prescriptions several times after an acute PUD diagnosis. Furthermore, limiting the analysis to only the first PPI prescription exceeding the approved treatment duration may not reflect factors associated with later prescriptions. Model building details are described in Data S1. Third, to understand the health system-wide impact of a particular factor on filled PPI prescriptions exceeding the approved treatment duration, we calculated the adjusted time-to-event population attributable fraction (details in Data S1) [42]. We interpreted the adjusted population attributable fraction as the proportion of PPI prescriptions exceeding the approved duration that would be eliminated from the health system if patients with the factor were treated equivalently to patients without the factor.

Because the exclusion diagnoses described in the Cohort Eligibility subsection are absolute or relative indications for PPI, patients who experienced these after PUD diagnosis were censored from the analysis. Additionally, to analyze only healthcare delivery that is potentially influenced by the PUD diagnosis, all patients were censored after 3 years. Furthermore, because of a network-wide PPI deprescribing initiative conducted in VISN 17 starting on August 1, 2013, patients who had their PUD diagnosis in VISN 17 were censored after this date [20]. In the secondary analysis, patients meeting gastroprotection criteria were censored at the time they qualified under any guideline. Figure S2 depicts the cohort eligibility, outcome, healthcare factor, and censoring assessment intervals.

Data extraction from VHA CDW was conducted using the R statistical programming environment (Vienna, Austria). Data analysis was conducted using Stata MP 18.0 (College Station, TX). Figure 1 and Figure S2 were designed in BioRender. This research was classified as exempt human subjects research by the VA Pittsburgh Healthcare System Institutional Review Board.

2.6 Sensitivity Analysis

We conducted four sensitivity analyses of the primary analysis to assess the impact of key study design elements. First, to determine the impact of limiting the study follow-up to 3 years, we performed a sensitivity analysis in which the maximum follow-up time was extended to 5 years. Second, to assess the maximal impact of not accounting for delayed gastric ulcer healing on repeat upper endoscopy in the main analysis, we performed a sensitivity analysis where PPIs prescribed in the 60 days after follow-up upper endoscopies were to have approved treatment duration. Third, to reduce potential heterogeneity in provider decision-making about concomitant antiplatelet and PPI prescriptions related to the 2009 U.S. Food and Drug Administration warning about drug–drug interactions, we conducted a sensitivity analysis limiting the cohort to patients diagnosed with PUD in 2010 and later [43]. Fourth, because there may be unmeasured confounding with indications for anticoagulants and antiplatelets that could influence the association of the VA Frailty Index with PPI prescriptions, we conducted a sensitivity analysis in which the VA Frailty Index was calculated without indications for anticoagulants or antiplatelets (atrial fibrillation, cerebrovascular disease, coronary artery disease, and peripheral vascular disease).

3.1 Factors Associated With a Filled PPI Prescriptions Exceeding the Approved Treatment Duration

In the final multivariable model, age, inpatient PUD diagnosis, VA Frailty Index, number of concurrent prescriptions at PUD diagnosis, anticoagulants, aspirin, and NSAIDs were associated with a filled PPI prescription exceeding the approved treatment duration, whereas female sex, Black race, and calendar year were inversely associated (Table 2, Table S5). The strongest identified associations were inpatient PUD diagnosis (aHR 1.32, 95% CI 1.25–1.39), NSAIDs (aHR 1.26, 95% CI 1.18–1.34), anticoagulants (aHR 1.25, 95% CI 1.13–1.38), Black race (aHR 0.83, 95% CI 0.78–0.88), and moderately frail VA Frailty Index (1.15, 95% CI 1.06–1.26). The final multivariable model for the secondary outcome accounting for gastroprotection guidelines was similar except that female sex was not associated (Table 2).

3.2 Adjusted Population Attributable Fraction

On the health-system level, inpatient PUD diagnosis had the highest adjusted population attributable fraction for filled PPI prescriptions exceeding the approved treatment duration for the majority of follow-up, peaking at 0.26 39 days after PUD diagnosis. The peak adjusted population attributable fractions for NSAIDS, anticoagulants, and aspirin were 0.18, 0.18, and 0.09, respectively. The peak-adjusted population-attributable fraction of VA Frailty Index levels relative to nonfrail was 0.06 for prefrail, 0.09 for mildly frail, and 0.12 for moderately frail (Figure 3). The adjusted population-attributable fractions for the secondary outcome accounting for gastroprotection guidelines were nearly identical to those for the primary outcome (Figure S3).

3.3 Sensitivity Analysis

The results of the four sensitivity analyses related to follow-up time, delayed gastric ulcer healing, PPI-antiplatelet drug–drug interactions, and the VA Frailty Index were consistent with the main analysis with respect to the most strongly associated factors (Table S6).

4.1 Plain Language Summary

Proton pump inhibitors (PPIs) are used to treat peptic ulcer disease (PUD), which are ulcers that occur in the stomach or the small intestine. Regulatory agencies recommend that PPIs be used for no more than 8 weeks, and recent studies suggest that long-term PPI use is associated with side effects like intestinal infection and reduced kidney function. We used patient data from the Veterans Health Administration in the United States to determine rates of PPI treatment for PUD exceeding the eight-week treatment duration and healthcare factors associated with this overuse. We identified that 35% of patients treated for PUD had PPI use exceeding the approved treatment duration. Markers of patient complexity, such as being diagnosed with PUD as an inpatient and use of medications that can potentially cause gastrointestinal bleeding, were associated with PPI prescriptions exceeding the approved treatment duration. These results may inform the development of deprescribing programs that assist prescribers and health systems in reducing inappropriate long-term PPI use.