Proton pump inhibitors and acute kidney injury: a nested case–control study

Proton pump inhibitors (PPI) are a widely-used class of drugs for the treatment of gastro-esophageal reflux disease and other acid-related disorders of the gastrointestinal tract. In 2009, more than 119 million PPI prescriptions were filled in the United States, accounting for nearly $14 billion dollars in prescription PPI sales, in addition to billions more in over-the-counter (OTC) sales [1]. The PPI class includes several different agents, such as esomeprazole, omeprazole, lansoprazole, pantoprazole, deslansoprazole, rabeprazole, etc., all of which possess a common mechanism of action for reducing parietal cell acid production by blocking H+/K + adenosine triphosphatase [2, 3].

As a class, PPIs have demonstrated a favorable safety profile [4–6]. Overall, complications have been infrequent and minor. The most common unwanted effects reported have been headaches, abdominal pain, and diarrhea [2]. In fact, the favorable adverse effect profile has led to movement of some of these agents to non-prescription status. However, as the use of PPIs has grown more widespread so have safety concerns. A growing body of literature suggests that this class of drugs may be linked to acute kidney injury, which can potentially lead to chronic injury or kidney failure. In 1992, a sentinel case-report identified the PPI omeprazole as a possible cause of acute interstitial nephritis (AIN) in an elderly woman [7]. Since that time, numerous biopsy-confirmed case reports and retrospective descriptive reports have been published suggesting a connection between PPI use and AIN [8–10]. Detailed summaries of these case reports have been published [9, 11, 12]. However, research in this area has been limited and consisted of small sample sizes without comparator groups. As a result, no estimate of the effect size is currently available. In turn, determining if an association between PPI use and AIN exists requires studies using larger sample sizes and control groups.

Drug-induced AIN is a form of acute kidney injury thought to result from an idiosyncratic, cell-mediated immunologic reaction [13, 14]. Drugs are the most commonly cited cause of AIN, but infectious and immune etiologies are also possible [15]. Recognition of AIN is elusive as the clinical manifestations are subtle. Classically, AIN has been described to present the following triad of events: fever, rash, and arthralgias [16–18]. Unfortunately, in case-reports of PPI-induced AIN, the triad of symptoms was present in a minority of subjects. Most complaints were nonspecific, including fever, fatigue/lethargy, weight loss, and nausea/vomiting [9, 11, 12]. Laboratory abnormalities were more consistently present in PPI-induced AIN cases, with elevated serum creatinine, pyuria, proteinuria, and hematuria observed in over 60% of the cases [9, 11, 19]. Because of varied clinical presentation and lack of accurate non-invasive tests, AIN can only be definitively diagnosed through a renal biopsy [13, 18].

In practice, PPI-induced AIN is likely under- or mis-diagnosed due to low suspicion (infrequent occurrence, limited information dissemination) and inconsistent clinical presentation [12]. Rather, AIN is more likely to be classified as some other form of acute or chronic renal injury. Recognizing this mis-classification, studies using claims data to investigate a link between PPI use and AIN require broadening ICD-9 codes beyond the single heading “AIN”. The objective of this study was to determine if an association exists between renal injury and PPI use and in turn estimate an effect size for this association using claims data.

This retrospective nested case–control study used claims data from a private insurer in a single Midwestern state to examine the relationship between renal disease and prescription PPI use. The nested case–control design has been employed in pharmacoepidemiologic studies, and its strengths and limitations have been previously described [20–22]. The insurer’s databases included 39 months (September 2002 through November 2005) of inpatient, outpatient, and prescription claims for a cohort of 184,480 patients. The database has been used in a number of studies and is a generally complete claims record for covered patients [23–25]. Patient identifiers were encrypted but could be linked across medical and prescription claims databases and years as long as the patient had coverage during these years. The study was approved by the University of Nebraska Medical Center Institutional Review Board.

The inclusion of a propensity score in the model did not change the relationship between PPI use and renal disease (OR 2.05, CI 1.52, 2.72). Analyses of H2 Blockers showed that there was not a statistically significant relationship between H2 Blocker use and renal disease (OR 1.37, CI 0.46, 4.13).

To account for the potential that a 90 day exposure status may include patients not currently using PPIs, sensitivity analyses using a 15 and 30 exposure window were also conducted. The results, not presented, were consistent with and confirmed the base case analyses. Similarly, inclusion of just incident users (first PPI prescription within 90 days) confirmed the results of the base case analyses.

Previous publications have suggested an association between the use of PPIs and acute interstitial nephritis [7–12]. Our study is the first to use a quasi-experimental design to support an association between renal disease and PPI exposure even after controlling for other known causes of renal disease. Additionally, our study provides the first estimate of an effect size for this relationship. We found that patients with an incident of renal disease diagnosis were nearly twice as likely to have been exposed to PPIs compared to those without renal disease.

Presently, AIN is not preventable due to its idiosyncratic nature. Therefore it is important that emphasis be placed on timely recognition. Early detection and treatment (i.e., removal of the offending agent and possible use of oral corticosteroids) have been shown to reduce the morbidity of AIN [29]. Healthcare professionals are recommended to have a heightened awareness of patient complaints or clinical manifestations associated with AIN and an understanding of their possible association with PPIs. Pharmacists in particular are in a strategic position to link changes in a patient’s health status with recent PPI utilization.

Like all case–control studies, this analysis has limitations. Misclassification bias can occur if subjects are inaccurately classified regarding the outcome of interest (i.e., renal disease) or the exposure being investigated (i.e., PPI use). By more broadly defining PPI-associated renal disease, we were able to capture more cases, but we were also likely to have included instances of renal injury not associated with PPI exposure. This misclassification of cases could overestimate or underestimate the true relationship between PPI use and renal disease, depending on their distribution between exposure categories. Because our primary and secondary analysis controlled for or removed potential confounders from the analysis, we believe that the misclassified cases are not likely to be associated with PPI use, which is expected to make our estimate conservative.

Lack of OTC PPI utilization can lead to misclassification of exposure. Because OTC use was not captured in a claims database, it is possible that some subjects who used OTC PPIs were misclassified as non-users. Table 3 shows that cases, patients with a renal disease diagnosis, were more likely to be prescribed PPIs than controls. It is likely, given the differences in underlying comorbidities, that cases also used more OTC PPIs. This misclassification of cases would underestimate the effect of PPI injury and make a positive finding more difficult, leading to a more conservative estimate of association.

Surveillance bias could overestimate the impact of PPI exposure. Surveillance bias can occur, as Gordis described, “[i]f a population is monitored over a period of time, disease ascertainment may be better in the monitored population than in the general population…which leads to an erroneous estimate of the relative risk or odds ratio” [30]. In our study, individuals with renal disease were more likely to have an underlying chronic disease (Table 3). If subjects with chronic disease were more likely to see a physician, it is also more likely that a diagnosis of renal disease would have been made for those subjects. In an effort to control for the effect of surveillance bias, we created a second model where all persons with chronic diseases were removed. Evidence of an association between renal disease and PPI exposure remained (OR = 2.25) in the sample population without chronic diseases (Table 5).

Appropriate control of confounding variables is always important in the analysis of observational studies. In order to produce groups with similar important baseline characteristics, we matched cases to multiple controls based on age, gender, county of residence, and initial date of coverage. Additionally, we controlled for other causes of renal disease and health status, both directly and with propensity scores. It is possible that even after controlling for these known confounders, some residual confounding variables remain. It is also possible that additional confounding variables exist, such as OTC NSAID use, which could not be controlled for in the model, which is a limitation inherent to all observational studies.

The results of this nested case-controlled study affirm an association between renal disease and PPI use. Our study revealed that patients with a renal disease diagnosis were twice as likely to have a previous prescription for a PPI. Therefore, it is necessary that pharmacist and physician awareness and recognition of patient complaints or clinical manifestations of this potentially harmful event are increased. Furthermore, it is important that future research seek to establish a definitive causal relationship.

This paper was presented as a poster presentation at the 2006 Annual Meeting of the Americal College of Clinical Pharmacy, St. Louis, MO, October 26–29, 2006. GLC holds a Pharm.D. and Master of Science in Epidemiology and is an Assistant Professor in the Department of Pharmacy Practice at the University of Nebraska Medical Center. DSC is a Pharm.D. with additional credentials as a Board Certified Pharmacotherapy Specialist (BCPS) who is an Assistant Professor in the Department of Pharmacy Practice at the University of Nebraska Medical Center. DGK is a Ph.D. and M.B.A. who is an Assistant Professor in the Department of Pharmacy Practice at the University of Nebraska Medical Center.