eAppendix. Supplemental Appendix.
Customize your JAMA Network experience by selecting one or more topics from the list below.
McDonald EG, Milligan J, Frenette C, Lee TC. Continuous Proton Pump Inhibitor Therapy and the Associated Risk of Recurrent Clostridium difficile Infection. JAMA Intern Med. 2015;175(5):784–791. doi:10.1001/jamainternmed.2015.42
Copyright 2015 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.
Clostridium difficile infection (CDI) is associated with significant morbidity, mortality, and a high risk of recurrence. Proton pump inhibitor (PPI) use is associated with an initial episode of CDI, and PPIs are frequently overprescribed. For many, the use of PPIs could likely be discontinued before CDI recurrence.
To determine whether PPI use was associated with a risk of initial CDI recurrence, to assess what proportion of patients who developed CDI were taking a PPI for a non–evidence-based indication, and to evaluate whether physicians discontinued unnecessary PPIs in the context of CDI.
Design, Setting, and Participants
We conducted a retrospective cohort study of incident health care–associated CDI cases to determine the association between continuous PPI use and CDI recurrence within 90 days. The setting was 2 university-affiliated hospitals, the 417-bed Montreal General Hospital (Montreal, Quebec, Canada) and the 517-bed Royal Victoria Hospital (Montreal, Quebec, Canada). The cohort consisted of 754 patients who developed health care–associated CDI between January 1, 2010, and January 30, 2013, and who survived for a minimum of 15 days after their initial episode of nosocomial CDI.
Continuous PPI use.
Main Outcomes and Measures
Recurrence of CDI within 15 to 90 days of the initial episode.
Using a multivariable Cox proportional hazards model, the cause-specific hazard ratios for recurrence were 1.5 (95% CI, 1.1-2.0) for age older than 75 years, 1.5 (95% CI, 1.1-2.0) for continuous PPI use, 1.003 (95% CI, 1.002-1.004) per day for length of stay, and 1.3 (95% CI, 0.9-1.7) for antibiotic reexposure. The use of PPIs was common (60.7%), with only 47.1% of patients having an evidence-based indication. Proton pump inhibitors were discontinued in only 3 patients with CDI.
Conclusions and Relevance
After adjustment for other independent predictors of recurrence, patients with continuous PPI use remained at elevated risk of CDI recurrence. We suggest that the cessation of unnecessary PPI use should be considered at the time of CDI diagnosis.
Clostridium difficile is the most common cause of infectious diarrhea in hospitalized patients1 and is associated with significant attributable costs (US $11 000 per nosocomial case2) and morbidity and risk of mortality.3Quiz Ref IDEven when an initial episode of C difficile infection (CDI) is successfully treated, approximately 20% of patients will experience 1 or more recurrences.4 The management of patients with multiple recurrences is challenging, and therapeutic options are limited. Therefore, when faced with a patient with an initial episode of CDI, an important goal must be the prevention of subsequent episodes.
A reasonable approach to preventing recurrence is to minimize modifiable risk factors such as antibiotic exposure. Proton pump inhibitor (PPI) use is prevalent and has been identified as an important exposure before an initial episode of CDI.3,5,6 Consequently, Health Canada7 and the US Food and Drug Administration8 have released advisories that in all patient populations PPIs should be used at the lowest dosage possible for the shortest duration that is clinically indicated.
Despite associations with CDI and other recognized adverse events,8-11 PPIs are still overprescribed.12-14 Given that many patients receive these drugs without evidence-based indications, we hypothesized that unnecessary PPI use could be discontinued in patients with CDI to minimize recurrence risk. This would be an intervention of negligible cost with substantial potential benefit. Nevertheless, although the association between PPI use and incident CDI is well established, the evidence implicating PPIs with recurrence is limited to a few conflicting studies.15-18
The objectives of our study were 3-fold. First was to determine whether PPI use was associated with a risk of initial CDI recurrence. Second was to assess what proportion of patients who developed CDI were taking a PPI for a non–evidence-based indication. Third was to evaluate whether physicians discontinued unnecessary PPIs in the context of CDI.
Ethics approval was granted by the McGill University Health Centre Research Ethics Board. Informed consent was waived.
This was a retrospective cohort study consisting of all patients who developed an initial episode of health care–associated CDI between January 1, 2010, and January 30, 2013, and who survived for a minimum of 15 days after their initial episode of nosocomial CDI. The study was conducted at 2 university-affiliated hospitals, the 417-bed Montreal General Hospital (Montreal, Quebec, Canada) and the 517-bed Royal Victoria Hospital (Montreal, Quebec, Canada).
As per Quebec standards, health care–associated CDI cases were those in which (1) symptoms occurred more than 3 days after admission or (2) symptoms caused readmission in a patient who had been hospitalized within the previous 30 days and who was not a resident in a long-term care facility. All potential cases were adjudicated by an infection control nurse for symptoms and were diagnosed by polymerase chain reaction for toxin B. Patients who were polymerase chain reaction positive without 3 episodes of diarrhea were excluded from the study. Exceptions included patients with ileus, toxic megacolon, colitis determined by results from computed tomography or colonoscopy, or autopsy-proven disease.
Patients within the cohort were evaluated for a recurrence of CDI, defined as recurrent symptoms (without alternative explanation) in association with a second positive C difficile polymerase chain reaction assay occurring between 15 and 90 days after the initial episode. While the Society for Healthcare Epidemiology of America and the Infectious Diseases Society of America guidelines suggest that a recurrence should be defined as occurring within 56 days of the initial episode,19 they recognize that the definition is not evidence based. We chose to include episodes occurring between days 57 and 90 because we believed that they were also clinically important, as well as to ensure that our results would be directly comparable to those of the conflicting large observational studies15,16 that have previously addressed this issue. Censoring occurred at CDI recurrence or at death within 90 days. Those who did not die or experience a recurrence were censored at 90 days.
There were 809 patients identified, of whom 55 were excluded because they died before day 15 and therefore could never have had a recurrence. This left a cohort of 754 patients whose cases were analyzed.
The main exposure of interest was continuous PPI use. Exposure was defined as the receipt of a PPI at the time of the initial episode of C difficile with either (1) ongoing exposure for at least 75% of days in the hospital in those who remained hospitalized or (2) a discharge prescription for a PPI that was valid beyond 90 days after the initial episode. To ensure that patients with CDI recurrence remained exposed, they were also reevaluated at the time of recurrence for a valid PPI prescription from at least 48 hours before recurrence. More than 90% of exposed patients were receiving pantoprazole sodium (40 mg daily) because this is one of only 2 PPIs (along with dissolvable lansoprazole) on our hospital formularies.
We extracted information on the treatment of the initial episode of CDI, the unit that the infection was attributed to (medical, surgical, critical care, or other), the admitting hospital (to control for differences in local colonization pressure), and length of stay. In our hospitals, the duration of initial treatment of CDI is 14 days.
Age, sex, and date of death (if deceased) were obtained from the patient registration database. Patient comorbidities were obtained from International Statistical Classification of Diseases, 10th Revision, data manually abstracted by medical record personnel during the index admission (eAppendix in the Supplement). Vancomycin-resistant enterococci and methicillin-resistant Staphylococcus aureus colonization status were obtained from the hospitals’ respective microbiology laboratories.
Ninety-nine percent (752 of 754) of patients were exposed to antibiotics before the initial episode of C difficile. We obtained in-hospital antibiotic reexposure data from our hospital pharmacy database following the standard 14 days of therapy for index CDI (within 15-90 days of the initial episode).
A general internist (E.G.M.) reviewed the indications for PPI use through a medical record review of all consecutive patients admitted to a medical clinical teaching unit or to a critical care unit (191 of 493 PPI users [38.7%]). One hundred of these medical records were subsequently reviewed by a second physician (J.M.) for an independent verification of the indication. κ Agreement was 0.94. Consensus for 3 cases of disagreement was reached after discussion with a third physician (T.C.L.). Patients admitted to these units (and not to surgical units) were chosen because they have detailed admission notes, providing the most accurate history obtainable to determine the rationale for PPI therapy (eAppendix in the Supplement).
With statistical software (STATA, version 13; StataCorp LP), we used a Cox proportional hazards model to determine the cause-specific hazards ratios for both initial recurrence and death as a competing risk.20,21 We also performed competing risk regression22 to determine subdistribution hazard ratios for both end points, and we present both analyses for comparison.21
We first conducted a univariate analysis looking for associations between comorbid conditions and C difficile recurrence. Factors that were associated with P < .25 and those believed a priori to be important were considered for inclusion in the multivariable models. Backward elimination was then used to arrive at the final models, with antibiotic reexposure being an important potential confounder included in all models. All variables in the final model were checked for significant interactions.
The proportional hazards assumption was tested by including a factor × time interaction for all variables. If this led to violation of the proportional hazards assumption, the interaction with time was included in the model as the means of correction of that violation, and the results for such variables are presented for days 15, 28, and 56.
Data were missing for both initial antibiotic therapy and antibiotic reexposure. For our multivariable analysis, we imputed these variables with multiple imputation using chained equations.23 Imputation models included the variables to be included in the analysis models. Twenty imputations were performed using logit for binary variables. Convergence was verified graphically, and the residuals were plotted against the fitted values to evaluate model fit. A complete case analysis was also performed to confirm that the conclusions were not changed by the imputation technique.
Among 754 eligible patients, there were 52 deaths without recurrence (6.9%), 193 documented recurrences (25.6%), and 509 survivors without documented recurrence within 90 days (67.5%). The comorbidities and clinical characteristics of these patients are listed in Table 1. Among 193 patients who experienced a recurrence, 157 recurrences (81.3%) occurred between days 15 and 56. The distribution of the timing of all 193 recurrences was as follows: 15 to 21 days (37 patients [19.2%]), 22 to 28 days (43 patients [22.3%]), 29 to 35 days (28 patients [14.5%]), 36 to 42 days (24 patients [12.4%]), 43 to 49 days (12 patients [6.2%]), 50 to 56 days (13 patients [6.7%]), and 57 to 90 days (36 patients [18.7%]).
The PPI users differed from nonusers in multiple ways (Table 2). Most important, they were more likely to experience a CDI recurrence (28.8% vs 20.6%, P = .007) or to die (10.3% vs 4.7%, P = .007) within 15 to 90 days of the initial episode.
Quiz Ref IDIn the multivariable analysis, the following remained independently associated with the rate of recurrence: age older than 75 years, continuous PPI use, length of stay, and vancomycin treatment of the initial episode. The cause-specific hazards ratios and subdistribution hazard ratios are listed in Table 3.
Among 458 of 754 patients (60.7%) who were taking a PPI at CDI diagnosis, 191 (41.7%) underwent medical record review, of whom 30 of 191 (15.7%) were cases and 161 of 191 (84.3%) were controls. Ninety of these 191 PPI users (47.1%) had an evidence-based indication for therapy (Table 4). The most common evidence-based indication for PPI use was age older than 60 years with 2 other concurrent risk factors for peptic ulcer disease (39 patients [20.4%]). Quiz Ref IDProton pump inhibitor use was discontinued in only 3 of 458 patients (0.6%).
In this retrospective cohort study, 754 patients with an initial episode of health care–associated CDI were evaluated to determine if PPI use was associated with CDI recurrence. Quiz Ref IDRecurrence was common, occurring in 193 of 754 patients (25.6%) within 90 days. After adjustment using 2 different but complementary methods of accounting for the competing risk of death, patients who were continuous users of PPIs had a cause-specific hazard ratio of 1.5 (95% CI, 1.1-2.0) and a subdistribution hazard ratio of 1.4 (95% CI, 1.1-1.9) for CDI recurrence. Probable overuse of PPIs was seen in more than 50%. Despite this, PPI use was were almost never discontinued.
Although the association between PPI use and incident CDI is well established, few studies have looked at the association of PPI use with recurrent disease. Recurrences are associated with a higher attributable cost, increased length of stay, and greater morbidity and mortality than an initial episode of CDI.31 Recurrent CDI is difficult to treat, with patients who experience an initial episode of recurrence having a 40% chance of developing a second recurrence and a 60% chance of further recurrences.32,33
A PubMed search identified 4 prior studies15-18 examining the association between PPI use and recurrent CDI. Two of these studies17,18 involved fewer than 150 patients each and demonstrated an association between PPI use and recurrence.
The association was strengthened with the findings by Linsky et al16 in a mixed population of 1166 inpatients and outpatients diagnosed as having CDI by enzyme-linked immunoabsorbent assay. In their cohort, 45% of patients were taking a PPI. After adjusting for comorbidities, antibiotic reexposure, and censoring for death, they established that PPI use was associated with CDI recurrence (hazard ratio, 1.4; 95% CI, 1.1-1.8).
In contrast, Freedberg et al15 analyzed hospitalized patients with CDI and found no association among 894 patients with incident CDI diagnosed by polymerase chain reaction. Exposure was defined as PPI receipt for a minimum of 2 days during CDI therapy (62% were exposed). This definition differs from other studies that have looked at continuous exposure. Unlike in the study by Linsky et al16 and in the present study, Freedberg et al15 did not adjust for antibiotic reexposure, potentially confounding the analysis.
Quiz Ref IDOur observational study has the potential for unmeasured and residual confounding. Except for peak white blood cell count and the number of episodes of diarrhea, we believe that we adjusted for many of the confounders associated with recurrent CDI.34 Although we do not routinely type for Cdifficile strain, provincially this has not changed over the study period, and roughly 50% are North American pulsed-field gel electrophoresis type 1 (NAP1) in Quebec.
There is a possibility that patients who experienced CDI recurrence at another center would not have been captured in our study. However, patients in our jurisdiction are routinely routed back to their usual hospital, and we believe that we captured most recurrences.
Rehospitalization and outpatient visits may be important risk factors for the receipt of a PPI and for recurrent CDI. While we were unable to specifically adjust for these interactions, we have attempted to minimize this confounding through the adjustment for age, comorbid illnesses, admission to a medical unit, and length of stay.
Although we did not have a strict method to determine PPI adherence in the community, we attempted to minimize incorrectly labeling patients with recurrence as being exposed by reverifying ongoing PPI use at the time of CDI recurrence. While this may have introduced differential misclassification bias in the patients who did not experience a recurrence, we believe that this would only have served to bias our study toward the null and therefore does not invalidate our conclusions.
Finally, patients whose initial CDI therapy was extended beyond 14 days were not examined separately. This represented less than 5% of patients, and we believe that this is unlikely to have nullified the association we observed between PPI use and recurrent CDI.
The strengths of our study are the large number of participants, our use of both the Cox proportional hazards model and competing risk regression model for dealing with the competing risk of death, the adjustment for potential confounders (including antibiotic reexposure), and a reasonable definition of continuous PPI exposure. To date, this is also the largest study to enumerate whether PPIs prescribed in the population with CDI are for evidence-based indications, as well as to demonstrate that their use is almost never discontinued in the context of infection.
Currently available options for the prevention of recurrent CDI are expensive, are difficult to access, and have limited evidence. Fidaxomicin is potentially equivalent to oral vancomycin for the initial treatment of disease and decreased the rates of recurrence (15.4% vs 25.3%, P = .005) when evaluated at days 36 to 40 for non-NAP1 strains.35 In our hospitals, fidaxomicin is not widely accessible because the Quebec formulary cites limited cost-effectiveness vs that of vancomycin. Another emerging viable treatment option, fecal transplantation,36 is also not widely available and is subject to governmental restrictions in many countries.
By contrast, PPI use is common, it is associated with CDI recurrence in 2 large observational studies,15,16 and the association is biologically plausible given the effect of PPI therapy on the gut microbiome.37 For more than half of the patients with incident CDI, we have shown that the indication for PPI use is not evidence based and have demonstrated that their PPI could be safely discontinued. The cost of PPI discontinuation is negligible, and tapering the PPI dosage before discontinuation may help to curb any complications such as symptoms of rebound hyperacidity. In the absence of a randomized clinical trial clearly demonstrating that ongoing nonindicated PPI use is safe, we suggest that the use of non–evidence-based PPIs should almost certainly be discontinued in the context of CDI.
New drugs and fecal transplantation may represent future treatments of CDI. In today’s era of choosing wisely, we propose that the prevention of CDI recurrence should begin with the cessation of unnecessary PPIs, a potentially unnecessary and frequently overused class of medications.
Accepted for Publication: January 7, 2015.
Corresponding Author: Emily G. McDonald, MD, Division of General Internal Medicine, Department of Medicine, McGill University Health Centre, 687 Pine Ave W, Room A421, Montreal, QC H3A 1A1, Canada (firstname.lastname@example.org).
Published Online: March 2, 2015. doi:10.1001/jamainternmed.2015.42.
Author Contributions: Drs McDonald and Lee had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: McDonald, Frenette, Lee.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: All authors.
Critical revision of the manuscript for important intellectual content: McDonald, Frenette, Lee.
Statistical analysis: McDonald, Lee.
Administrative, technical, or material support: Milligan.
Study supervision: Frenette, Lee.
Conflict of Interest Disclosures: None reported.
Create a personal account or sign in to: