differential risk of clostridium difficile infection with proton pump inhibitor use by level of...

ORIGINAL REPORT

Differential risk of Clostridium difficile infection with proton pumpinhibitor use by level of antibiotic exposure†

Vanessa Stevens1,2,4*, Ghinwa Dumyati3, Jack Brown1,2,4 and Edwin van Wijngaarden4

1Center for Health Outcomes, Pharmacoinformatics, and Epidemiology, SUNY Buffalo School of Pharmacy and Pharmaceutical Sciences,Buffalo, NY, USA2Department of Pharmacy, University of Rochester Medical Center, Rochester, NY, USA3Department of Medicine, University of Rochester Medical Center, Rochester, NY, USA4Department of Community and Preventive Medicine, University of Rochester School of Medicine and Dentistry, Rochester, NY, USA

ABSTRACTPurpose Clostridium difficile infection (CDI) is a major cause of hospital-acquired diarrhea worldwide. We examined the risk of CDI as-sociated with the use of acid-suppressive agents (proton pump inhibitors [PPI] and histamine-2 receptor blockers) and determined whetherthis risk varied by number or type of antibiotic (high or low CDI risk) received during hospitalization.Methods We conducted a retrospective cohort study of hospitalizations among adult patients at an academic teaching hospital in Rochester,New York, during which two or more days of antibiotics were prescribed. Multivariable marginal Cox proportional hazards models withtime-varying exposures were used to examine time to the development of CDI.Results A total of 10154 hospitalizations and 241 cases of CDI, defined as detection of C. difficile toxin in a diarrheal stool sample within60days of discharge, were identified. PPI use was independently associated with an increased risk of CDI (adjusted hazard ratio = 4.5; 95%confidence interval [CI]=2.3–9.0). Among hospitalizations during which one, two, three or four, and five or more antibiotics were pre-scribed, the adjusted hazard ratios for PPI use were 15.7 (CI=6.4–38.8), 4.9 (CI=2.2–11.2), 4.3 (CI=1.9–9.9), and 2.7 (CI=1.2–5.9), respec-tively (p for interaction=.002).Conclusions The use of PPI is common among patients receiving antibiotics during hospitalization. The greater risk of CDI in relation toPPI among hospitalizations during which fewer or low-risk antibiotics were prescribed suggests a potentially clinically relevant interactionbetween antibiotics and PPI. Further study is needed to elucidate possible mechanisms for the observed effect. Copyright © 2011 JohnWiley & Sons, Ltd.

key words—time-varying exposures; antibiotics; acid-suppressive agents; Clostridium difficile infection; effect modification

Received 1 March 2011; Revised 1 June 2011; Accepted 9 June 2011

INTRODUCTION

Clostridium difficile is an anaerobic, spore-formingbacteria that is part of the normal flora of up to 70%of neonates and approximately 5% of healthyadults.1–4 Symptoms of C. difficile infection (CDI)are mediated by the release of two toxins (A and B)and range from self-limiting diarrhea to intestinal

perforation, septic shock, and death.5 The spores thatare formed are heat and alcohol resistant and can per-sist for several months on inanimate surfaces, posinga serious problem for healthcare institutions.2 In thelast 10 years, the epidemiology of CDI has shifted,with the emergence of a new, highly fluoroquinoloneresistant strain that has been associated with increasedincidence and severity of disease, including increasesin intensive care unit admissions, colectomies, andmortality.6,7 These increases are thought to be relatedto dramatically higher levels of toxins A and B pro-duction, as well as the production of a previously un-recognized binary toxin.7

Antimicrobial exposure is the most common riskfactor for the development of CDI, although other

*Correspondence to: V. Stevens, Center for Health Outcomes, Pharmacoinfor-matics, and Epidemiology, State University of New York at Buffalo School ofPharmacy and Pharmaceutical Sciences, 315 Hochstetter Hall, Buffalo, NY14260, USA. E-mail: [email protected]†An abstract containing selected results from this article has been presented at the21st European Congress of Clinical Microbiology and Infectious Diseases andthe 27th International Congress of Chemotherapy. Milan, Italy, 7–10 May 2011.

Copyright © 2011 John Wiley & Sons, Ltd.

pharmacoepidemiology and drug safety 2011; 20: 1035–1042Published online 10 August 2011 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/pds.2198

medications that have similar inhibitory effects on thenormal intestinal flora, such as antineoplastic agents,have also been implicated.8 More recently, studieshave suggested a role for acid-suppressive agents, par-ticularly proton pump inhibitors (PPI), in the etiologyof CDI.9–16 In addition, two previous epidemiologicstudies have reported that the risk of CDI with PPIuse may be modified by concurrent treatment withantibiotics; however, these results have not been con-firmed in multivariable analyses.9,13 Results of studiesto evaluate the association between histamine-2 recep-tor (H2) blockers and CDI are conflicting,9,14 and thepotential for effect modification by antibiotic use hasnot been evaluated. In 2009, PPI were the third largestclass of drugs in US sales,17 and there is evidence tosuggest that much of this utilization is without appro-priate medical indication.18,19 Acid-suppressive agentuse and concurrent administration with antibiotics maytherefore represent modifiable risk factors for the devel-opment of CDI. Accordingly, we examined the relation-ship between use of PPI and H2 blockers and the risk ofCDI, and the potential for modification of the effectsby concurrent antibiotic exposure.

METHODS

Study design and participants

We conducted a retrospective cohort study amongpatients hospitalized at Strong Memorial Hospital(SMH), a tertiary care, academic teaching hospital inRochester, New York, to assess the relationship be-tween the class, total cumulative dose, number, anddays of antibiotic therapy and the risk of CDI. Rele-vant aspects of the study design and population aredescribed in detail elsewhere20 and are summarizedhere. Hospitalizations were included if they were con-tributed by a patient who was (i) 18years of age orolder; (ii) admitted to an adult, non-psychiatric inpa-tient unit; and (iii) prescribed two or more consecutivedays of antibiotics during their stay. Hospitalizationsfor patients with a toxin-positive assay for C. difficilein the 60days prior to admission or within 2 days fol-lowing admission were excluded. Patients were fol-lowed until 60days after discharge, the developmentof CDI, or the end of the study period, whichever camefirst. During the period from 1 January 2005 to 31December 2005, 10154 qualifying hospitalizationswere contributed by 7792 unique patients. One or moreacid-suppressive agents (PPI or H2 blockers or both)were prescribed during 6965 (68.6%) of hospitalizations.Data were obtained electronically from the SMH

pharmacy, clinical microbiology, and administrativedatabases. Information was collected on demographics,

admission and discharge dates, International Classifi-cation of Diseases, Ninth Revision (ICD-9) proceduraland diagnostic codes, medications received, and inpa-tient unit(s) of residence for each hospital stay. Approvalfor this study was obtained from the institutional reviewboards at SMH, Rochester General Hospital (RGH), andUnity Health System (for ACM laboratory).

OUTCOME ASSESSMENT

A case hospitalization was defined as the detection ofC. difficile toxin in the stool by enzyme immunoassaywithin 60days following hospital discharge. To maxi-mize the sensitivity of case ascertainment, we obtainedinformation on all positive assays for C. difficile duringthe study period from the SMH and the two other ma-jor laboratories that serve the community (RochesterGeneral Hospital and ACM laboratories). All threelaboratories routinely tested only unformed stools us-ing Premier Toxin A&B enzyme immunoassay kit(Meridian Biosciences, Cincinnati, OH, USA). Theprimary outcome of the study was time to the develop-ment of CDI and was defined as the time from admis-sion until 60days after discharge or end of study fornon-case hospitalizations and time from admission tothe detection of toxin in the stool for case hospitaliza-tions. Hospitalizations during which toxin was detectedwithin the first 48 hours of admission were excludedbased on presumed presence of CDI on hospitaladmission.

PHARMACOLOGIC EXPOSURES

Information was available on the timing, dose, and dura-tion of all medications prescribed and dispensed duringhospitalization. Chemotherapy, pharmacologic gastro-intestinal (GI) interventions (antacids, laxatives, andenemas), and immune suppressants (including steroids)prescribed prior to the development of CDI were catego-rized as dichotomous exposures.Hospitalizations were considered as exposed to

high-risk antibiotics if any of the following drugs wereprescribed based on previously observed associationswithin this cohort20: any cephalosporin or carbapenem,b-lactamase inhibitor combinations, fluoroquinolones,or intravenous vancomycin. All others were consideredlow risk. The total cumulative number of antibioticwas calculated as the number of unique antibiotic ordersseparated by one or more days and was divided intoquartiles because of lack of linearity with the log(hazard) of CDI. Total cumulative dose and daysof antibiotic therapy were also included in the orig-inal analysis but are omitted here because of the

v. stevens ET AL.1036

Copyright © 2011 John Wiley & Sons, Ltd. Pharmacoepidemiology and Drug Safety, 2011; 20: 1035–1042DOI: 10.1002/pds

strong correlation (.75 or above and p< .0001 foreach pair; data not shown) between the three quantita-tive measures of antibiotic consumption. Number ofantibiotics was chosen because it was the most predic-tive of outcome and would therefore most likely dem-onstrate the presence of interactions if they truly exist.Metronidazole orders written within 48 hours of a pos-itive assay for C. difficile and all oral vancomycinorders were considered treatment for CDI and werenot considered in the analysis. All antibiotic exposureswere allowed to vary over time.

Acid-suppressive agents

Exposures to PPI and H2 blockers were consideredseparately as dichotomous variables, the values ofwhich were allowed to vary according to the timingof exposures during hospitalization. Exposure status(exposed/unexposed) was assessed daily beginningon the day of admission until discharge, end of study,or the development of CDI, whichever came first.Because information was lacking regarding expo-sures after discharge, all patients were assumed tobe unexposed after discharge. PPI and H2 blockerswere treated as time-varying covariates using thecounting process formulation of the Cox proportionalhazards model.21

OTHER COVARIATES

The presence of the following comorbid conditions wasassessed based on medications prescribed within 48hours following admission: diabetes, respiratory illness,kidney disease, transplant, and cancer. These conditionswere included as part of the previously validated chronicdisease score for infectious diseases22 in the originalstudy20 but were considered as dichotomous variablesin the present analysis to avoid collinearity betweenthe score, which includes peptic ulcer disease, and ex-posure to the acid-suppressive agents, which are usedas treatment for these disorders.Admission and discharge dates, ICD-9 codes, and mi-

crobiologic data were used to assess previous hospitali-zations and positive assays for CDI at any of the threelaboratories within the study period, surgery within the2 months prior to admission, and GI interventions, in-cluding colonoscopies, endoscopies, and the placementor use of feeding tubes. HIV infection was defined asthe presence of an HIV-specific ICD-9 code or the pre-scription of medications used solely for the treatmentof HIV infection. Length of current hospital stay wascalculated as the time from admission to discharge fornon-case hospitalizations and time from admission until

the development of CDI for case hospitalizations. As aproxy for the Clostridium difficile- Associated Disease(CDAD) pressure,23,24 or exposure to C. difficilespores, inpatient units were ranked according to theirannual rate of CDI infection per 1000 patient daysand divided into approximate tertiles (high-risk,medium-risk, and low-risk units). This measure isreferred to as colonization risk and was allowed tochange as patients were transferred from one unit toanother during hospitalization.

DATA ANALYSIS

Demographic characteristics of the 7792 uniquepatients were compared between those who devel-oped CDI during any hospitalization and thosewho did not, using chi-square tests and Wilcoxonrank–sum tests, as appropriate. To account for theclustering of patients with multiple hospitalizationswithin the cohort, all hospitalization-specific bivariateand multivariable analyses were conducted using themarginal Cox proportional hazards models with robustcovariance estimation.25 Crude differences in time toCDI by exposure to acid-suppressive agents werecompared using stratified Kaplan–Meier survival anal-ysis and the log-rank test. Covariates were included inmultivariable analysis if they were associated with thetime to development of CDI at p< .20. All multivar-iable models were stratified on the receipt of antacids,laxatives, and enemas because of the nonproportion-ality of the hazards over time. A significant interac-tion between the antibiotic exposure variables andacid-suppressive (PPI or H2 blocker) use was definedas p< .10 on the likelihood ratio test comparing the fitof the reduced model (without the interaction term) withthat of the full model. A sensitivity analysis was con-ducted to determine whether the results varied accord-ing to whether patients were diagnosed with CDIduring hospitalization or after discharge. The resultswere nearly identical for both groups (data not shown).Statistical analyses were performed in SAS Version9.2 (Cary, NC, USA) using a two-sided alpha of .05.

RESULTS

Of the 7792 unique patients who contributed hospitali-zations to the study cohort, 6226 (80%) contributed only1, 1085 (14%) contributed 2, and 481 (6%) contributed3 or more hospitalizations. There were a total of 241CDI events from 240 patients. A description of the clin-ical characteristics of hospitalizations for patients withand without exposure to acid-suppressive agents isshown in Table 1. PPI and H2 blockers were prescribed

antibiotics, acid suppression, and c. difficile 1037


during 6322 (62%) and 1083 (11%) of hospitaliza-tions, respectively. Approximately 5% of patientswere prescribed both classes of acid-suppressive agentduring hospitalization.On univariate analysis, patients who were prescribed

PPI during hospitalization were 7.6 times (95% confi-dence interval [CI] = 5.7–10.2) more likely to developCDI than were patients without exposure (data notshown). Figure 1 shows unadjusted Kaplan–Meieranalysis of time to CDI for hospitalizations with andwithout PPI exposure. After controlling for the effectsof age, number and class of antibiotics received,comorbid conditions, and other procedures and medi-cations, this estimate was reduced to 4.5 (95%CI =2.3–9.0). H2 blockers were not associated with anincreased risk of CDI in either univariate or

multivariable analyses (Table 2). The number of anti-biotics received was associated with CDI in a dose-dependent manner, independent of acid-suppressiveagent use. Patients who received five or more antibioticsduring hospitalization were almost 6 times more likelyto develop CDI than were those who only receivedone (adjusted hazard ratio [AHR] = 5.9; 95%CI = 3.7–9.5). Other independent predictors of CDI include age,length of hospitalization at risk, kidney disease, and GIprocedures.Risks of CDI associated with the use of PPI varied

significantly by level of concurrent antibiotic expo-sures (p for interaction=.002). Among those withone, two, three or four, or five or more antibiotics,PPI users were 15.7 (95%CI = 6.4–38.8), 4.9 (95%CI = 2.2–11.2), 4.3 (95%CI = 1.9–9.9), and 2.7(95%CI = 1.2–5.9) times more likely to develop CDIthan were those without PPI exposures during hospi-talization, respectively. PPI use among patients with ex-posure to low-risk antibiotics was associated with a16.8-fold (95%CI = 8.1–35.1) increased risk of CDI rel-ative to patients with no PPI exposure, whereas PPIuse was not associated with CDI among patients whowere also receiving high-risk antibiotics (AHR=1.2;95%CI=0.7–1.9; p for interaction< .0001) (Table 3).No interactions between H2 blockers and antibioticuse were observed (Table 4).

DISCUSSION

Antibiotics remain the most significant contributors tothe risk of CDI. Acid-suppressive agents have alsoemerged as an important risk factor in recentyears.9,11–14 Despite the frequency with which thesemedication classes are given together, little empiric

Table 1. Distribution of demographic and clinical characteristics of hospi-talizations by acid-suppressive agent use

Characteristic

PPI H2 blockers

Any(n=6322)

None(n=3832)

Any(n=1083)

None(n=9071)

Development ofCDI

201 (3) 40 (1) 23 (2) 218 (2)

Median (IQR)no. of previoushospitalizations

0 (1) 0 (0) 0 (0) 0 (1)

Median (IQR)length of at-riskstay, days

8 (10) 5 (4) 9 (13) 7 (7)

Any proton pumpinhibitors

– – 516 (48) 5806 (64)

Any H2 blockers 516 (8) 567 (15) – –GI procedure 1252 (20) 187 (5) 222 (21) 1217 (13)HIV 120 (2) 62 (2) 13 (1) 169 (2)Previous CDI 53 (1) 3 (0) 1 (0) 55 (1)Patient location (colonization risk)Low-risk unit 1208 (19) 1575 (41) 270 (25) 2513 (28)Medium risk unit 3023 (48) 1337 (35) 521 (48) 3839 (42)High-risk unit 2091 (33) 920 (24) 292 (27) 2719 (30)

Antacids,laxatives, andenemas

4372 (69) 1813 (47) 849 (78) 5336 (59)

Immunesuppressants

2317 (37) 581 (15) 361 (33) 2537 (28)

Chemotherapy 465 (7) 81 (2) 98 (9) 448 (5)Cancer 253 (4) 51 (1) 50 (5) 254 (3)Diabetes 1707 (27) 523 (14) 217 (20) 2013 (22)Kidney disease 423 (7) 108 (3) 55 (5) 476 (5)Respiratoryillness

1531 (24) 541 (14) 178 (16) 1894 (21)

Transplant 896 (14) 117 (3) 57 (5) 476 (11)No. of antibioticsOne 2000 (32) 1775 (46) 301 (38) 3474 (38)Two 1598 (25) 963 (25) 288 (27) 2273 (25)Three or four 1699 (27) 876 (23) 273 (25) 2302 (25)Five or more 1025 (16) 218 (6) 221 (20) 1022 (11)

High-riskantibiotics

5889 (93) 3288 (86) 1000 (92) 8177 (90)

CDI, Clostridium difficile infection; PPI, proton pump inhibitors; H2,histamine-2; IQR, interquartile range; GI, gastrointestinal.

Figure 1. Stratified Kaplan–Meier analysis of time to the development ofCDI for hospitalizations during which PPI were and were not prescribed.Event-free survival was significantly greater for those without exposure toPPI by log–rank test (p < .0001). CDI, Clostridium difficile infection; PPI,proton pump inhibitors



information is available regarding concurrent use ofantibiotics and acid-suppressive therapy and the riskof CDI. We observed that PPI use was associatedwith a significant increase in the risk of CDI, evenafter controlling for the effects of potential confoun-ders. In addition, we noted that this risk was

significantly modified by the number and risk pro-file (high or low risk for inducing CDI) of antibio-tics concurrently administered. Few previousstudies have examined the association between PPIand CDI stratified by level or type of antibiotic expo-sure,9,13 and only one study evaluated for potentialdose–response by number or CDI risk profile.9 Dialand colleagues observed a twofold increase in therisk of CDI with PPI use (crude relative risk [RR]=2.1; 95%CI = 1.4–3.4).9 Among patients receivinghigh-risk antibiotics (defined a priori as ampicillinand analogues, clindamycin, and second-generationand third-generation cephalosporins based on litera-ture review9,26), patients with PPI were not at increasedrisk of CDI compared with those without PPI (RR = 1.1;95%CI = 0.5–2.3, calculated from published data).Although not presented, sufficient data were avail-able to calculate the risk estimate for patients takinglow-risk antibiotics. Among these patients, those whowere also receiving PPI were 3.4 times (95%CI = 1.9–7.0)more likely to develop CDI when compared withpatients without PPI exposure (calculated p for interac-tion=.02 by Breslow–Day test for homogeneity).9 Inour study, clindamycin was considered a low-risk anti-biotic, and intravenous vancomycin and fluoroquino-lones were classified as high-risk antibiotics.20 Thesedifferences could account for the disparity in the mag-nitude of observed risk estimates. The investigatorsalso noted decreasing risk associated with PPI useacross strata of increasing number of antibiotics, withthe strongest association between PPI and CDI beingobserved among patients with one antibiotic (RR =3.1; 95%CI = 1.5–6.3). Among patients who receivedtwo and three or more antibiotics, risk estimates were2.8 (95%CI = 0.8–10.3) and 1.1 (95%CI=0.6–2.1),

Table 3. Risk of CDI with proton pump inhibitor use stratified by concur-rent antibiotic therapy

Characteristic Adjusted hazard ratio * 95% confidence interval *

Number of antibioticsOne 15.7 6.4–38.8Two 4.9 2.2–11.2Three or four 4.3 1.9–9.9Five or more 2.7 1.2–5.9

Class of antibioticLow risk 16.8 8.1–35.1High risk† 1.2 0.7–1.9

CDI, Clostridium difficile infection.*Adjusted hazard ratio and 95% confidence interval from marginal Coxproportional hazards models with robust covariance estimation.

†Includes any cephalosporin, carbapenem, b-lactamase inhibitors, fluoro-quinolones, or intravenous vancomycin.

Table 4. Risk of CDI with H2 blocker use stratified by concurrent antibi-otic therapy

Characteristic Adjustedhazard ratio*

95% confidenceinterval*

p forinteraction*

Number of antibioticsOne† NA NATwo 1.1 0.1–7.7 NAThree or four 2.9 0.9–9.6Five or more 1.5 0.4–6.4

Class of antibioticLow risk† NA NA NAHigh risk{ 2.0 0.8–4.8

CDI, Clostridium difficile infection; H2, histamine-2; CI, confidenceinterval.*Adjusted hazard ratio, 95%CI, and p values from marginal Cox propor-tional hazards models with robust covariance estimation.

†Adjusted hazard ratio, 95%CI, and p values inestimable because of inade-quate cell sizes.

{Includes any cephalosporin, carbapenem, b-lactamase inhibitor, fluoroqui-nolone, or intravenous vancomycin.

Table 2. Pharmacologic and clinical characteristics associated with CDIin multivariable analysis

Characteristic Adjustedhazard ratio

95% confidenceinterval

p*

Proton pumpinhibitors

4.5 2.3–9.0 <.0001

H2 blockers 1.7 0.7–3.9 .25Number of antibiotics <.0001One RefTwo 2.2 1.4–3.5Three or four 2.6 1.7–4.0Five or more 5.9 3.7–9.5

Antibiotic class .001Low risk RefHigh risk† 2.7 1.5–4.9

Age, per 10years 1.2 1.1–1.3 <.0001Cancer 1.5 0.7–3.4 .30Diabetes 1.3 1.0–1.7 .09Kidney disease 1.8 1.2–2.8 .01Respiratory illness 1.0 0.8–1.4 .79GI procedures 1.4 1.0–1.9 .03HIV 2.1 1.0–4.4 .06Previous CDI 1.9 0.7–5.3 .23Chemotherapy 1.1 0.6–2.0 .70Immune suppressiveagents

0.8 0.6–1.1 .24

Length of hospitalstay at risk{

0.8 0.7–0.9 <.0001

Colonization risk .08Low-risk unit RefMedium risk unit 1.4 0.9–2.0High-risk unit 1.6 1.1–2.5

CDI, Clostridium difficile infection; H2, histamine-2; GI, gastrointestinal.*Adjusted hazard ratio, 95% confidence interval, and p values from mar-ginal Cox proportional hazards models with robust covariance estimation.

†Includes any cephalosporin, carbapenem, b-lactamase inhibitors, fluoro-quinolones, or intravenous vancomycin.

{Hazard ratio per 7-day increment in length of stay at risk.



respectively.9 The consistency of these patterns withour findings suggests the presence of a true interactionrather than a statistical artifact.The mechanism bywhich PPI may increase the risk of

CDI is not well understood. It has been suggested thatreduced acidity of the gastric contents resulting fromtreatment with acid-suppressive therapy allows vegeta-tive bacteria to pass into the small intestine.13 Althoughthere is some evidence to support this hypothesis,27,28

it has also been demonstrated that C. difficile sporesare unaffected by normal gastric pH,29 and the relativecontributions of spores and vegetative bacteria to thepathogenesis of disease are currently unclear. PPI havealso been shown to inhibit the phagocytic neutrophilresponse to Escherichia coli,30 leading to speculationthat altered host immune functions may increase therisk of incident or recurrent CDI.31 Alternatively, PPImetabolites may exhibit antibacterial effects in the GItract. The inhibitory properties of sulfide analoguesformed after proton-dependent activation of PPI againstHelicobacter spp. have been documented,32–34 but thesusceptibilities of bacterial strains common to the nor-mal flora were not evaluated, and the specific mecha-nism of bactericidal activity is unknown. Furtherresearch is needed to clarify whether PPI has sufficientactivity against the normal flora to have a meaningfulimpact on the risk of CDI.There is little information on potential mechanisms for

the interaction between PPI and antibiotics in the etiologyof CDI. It is possible that PPI-induced alterations to thepH, volume, and viscosity of gastric juices35,36 result inaltered absorption, distribution, and activity of antibio-tics. It has been shown that decreased gastric pHresults in decreased solubility of oral drug formula-tions and the fraction of drug that is systemicallyabsorbed.37 Decreased systemic availability may resultin an increase in the amount of antibiotics that arepassed into the intestine, which in turn may have agreater impact on the commensal flora. Studies con-ducted in the context of antibiotics and PPI giventogether as part of standard “triple-therapy” regi-mens for Helicobacter pylori infections have sug-gested that coadministration of PPI significantly altersthe pharmacokinetics of clarithromycin38 and the trans-fer of metronidazole to gastric juices.39 Similar obser-vations have been made for the oral cephalosporincefpodoxime proxetil.40 These results suggest that a bi-ologic interaction between PPI and antibiotics is at leastplausible, but more research is needed to clarify the issue.Length of stay was associated with a 20% reduction

in the risk of CDI for every 7 days. This is consistentwith at least one previous study12 of the associationsbetween PPI and antibiotics and the risk of CDI. It is

possible that censoring CDI patients at the time ofinfection results in an artificially lower length of staybecause it is likely that at least some portion of thelength of stay following CDI is attributable to the ad-mitting conditions, rather than CDI alone.This study is limited by the retrospective and adminis-

trative nature of the data. Information was only availableon exposures that occurred during hospitalization. How-ever, an estimated 70% of patients placed on acid-suppressive therapy during hospitalization are dischargedwith continuing therapy despite lack of medical indica-tion,19 suggesting that the intensity of acid-suppressivetherapy following discharge is likely to be similar tothat during hospitalization and will have little impacton the study results. We assumed that all patientswere unexposed to acid-suppressive agents afterdischarge. Any misclassification of acid-suppressiveagent exposure is expected to be similar for CDI casesand noncases, resulting in an underestimation of the trueeffect. Sensitivity analysis showed similar results amongpatients diagnosedwith CDIwhile hospitalized and thosediagnosed after discharge.We also lacked information onexposures that occurred prior to hospitalization. As a re-sult, our results likely represent the average of effectsamong long-term and short-term users of PPI and H2

blockers. Because cohort eligibility was conditionalon the receipt of antibiotics, we were unable to assessthe effect of acid-suppressive agents among patientswith no antibiotic exposure. In addition, we are unableto distinguish between biological and statistical expla-nations for the observed interactions. Our observationthat the risk of CDI associated with PPI use decreasesas level of antibiotic exposure increases could reflectdifferences in the relative importance of two biologi-cally distinct processes, where antibiotic exposure isthe primary driving risk factor. For instance, the riskof CDI is relatively low among patients who receiveonly one antibiotic, so the effect of PPI in this popula-tion is pronounced. In contrast, the risk of CDI amongpatients who receive five or more antibiotics is muchhigher, so the effect of PPI on the risk of CDI ismasked by the effects of administration of multipleantibiotics. As with any retrospective study, there isthe possibility that the observed associations are the re-sult of chance or bias. However, this is unlikely giventhe consistency of our results with those of Dial and col-leagues using different design and analytic methods.This study also has several strengths. Despite the fact

that hospitalized patients tend to have complex expo-sures, whichmay change considerably during the courseof admission, no previous studies have examined the as-sociation between acid-suppressive agents and CDIwhile accounting for these changes over time. The use



of a large cohort allowed detailed evaluation of the po-tential for effect modification while controlling for sev-eral important potential confounders.The prescription of acid-suppressive agents is common

among patients receiving antibiotics during hospitaliza-tion. Our study supports the observation that PPI are in-dependently associated with increased risk of CDI. Inthe context of low-risk antibiotics or minimal antibioticexposure, PPI may play a more important role in the riskof CDI. Additional pharmacokinetic and epidemiologicstudies are required to understand potential mechanismsfor an interaction.

AUTHOR CONTRIBUTIONS

The following are the contributions of the authors:

Dr. Stevens had full access to all the data in thestudy and takes full responsibility for the integrityof the data and the accuracy of the data analysis.Study concept and design: Stevens, van Wijngaarden,Brown, and DumyatiData acquisition: Stevens, van Wijngaarden, andDumyatiData cleaning and statistical analyses: StevensData interpretation: Stevens, van Wijngaarden, Brown,and DumyatiManuscript preparation: StevensManuscript review: Stevens, van Wijngaarden, Brown,and Dumyati.

STATEMENT OF ORIGINAL WORK

This article describes the analysis of secondary objec-tives of a previously conducted cohort study. None ofthe analyses presented here have been previously pub-lished in whole or in part.

CONFLICT OF INTEREST

There are no funding sources or relevant conflicts ofinterest to report.

KEY POINTS

• Proton pump inhibitors are independently associ-ated with CDI.

• The effect of PPI on the risk of CDI is signifi-cantly modified by antibiotic exposure.

• The risk of CDI in relation to PPI use is highestamong patients with lower risk antibiotic exposures.

REFERENCES

1. Bouza E, Munoz P, Alonso R. Clinical manifestations, treatment and controlof infections caused by Clostridium difficile. Clin Microbiol Infect 2005; 11(Suppl 4): 57–64.

2. Mitty RD, Lamont JT. Clostridium difficile colitis: epidemiology, pathophysiol-ogy, and treatment. Mt Sinai J Med 1994; 61(4): 329–335.

3. Reinke CM, Messick CR. Update on Clostridium difficile-induced colitis, Part 1.Am J Hosp Pharm 1994; 51(14): 1771–1781.

4. McClane BA, Lyerly DM, Moncrief JS, Wilkins TD. Enterotoxic Clostridia:Clostridium perfringens type A and Clostridium difficile. In Gram PositivePathogens (Vol. 2), Fischetti VA, Novick RP, Ferretti JJ, Portnoy DA, Rood JI(eds). ASM Press: Washington, D.C., 2006; 708-709-714.

5. Johnson S, Gerding DN. Clostridium difficile--associated diarrhea. Clin InfectDis 1998; 26(5): 1027–1034; quiz 1035–1026.

6. McDonald LC, Killgore GE, Thompson A, et al. An epidemic, toxin gene-variantstrain of Clostridium difficile. N Engl J Med 2005; 353(23): 2433–2441.

7. Rupnik M, Wilcox MH, Gerding DN. Clostridium difficile infection: new devel-opments in epidemiology and pathogenesis.Nat RevMicrobiol 2009; 7(7): 526–536.

8. Anand A, Glatt AE. Clostridium difficile infection associated with antineoplasticchemotherapy: a review. Clin Infect Dis 1993; 17(1): 109–113.

9. Dial S, Alrasadi K, Manoukian C, Huang A, Menzies D. Risk of Clostridiumdifficile diarrhea among hospital inpatients prescribed proton pump inhibitors:cohort and case-control studies. CMAJ 2004; 171(1): 33–38.

10. Dial S, Delaney JA, Barkun AN, Suissa S. Use of gastric acid-suppressive agentsand the risk of community-acquired Clostridium difficile-associated disease.JAMA 2005; 294(23): 2989–2995.

11. Dial S, Delaney JAC, Schneider V, Suissa S. Proton pump inhibitor use and riskof community-acquired Clostridium difficile-associated disease defined byprescription for oral vancomycin therapy. Can Med Assoc J 2006; 175(7):745–748.

12. Dalton BR, Lye-Maccannell T, Henderson EA, Maccannell DR, Louie TJ. Pro-ton pump inhibitors increase significantly the risk of Clostridium difficile infec-tion in a low-endemicity, non-outbreak hospital setting. Aliment PharmacolTher 2009; 29(6): 626–634.

13. Cunningham R, Dale B, Undy B, Gaunt N. Proton pump inhibitors as arisk factor for Clostridium difficile diarrhoea. J Hosp Infect 2003; 54(3):243–245.

14. Howell MD, Novack V, Grgurich P, et al. Iatrogenic Gastric Acid Suppressionand the Risk of Nosocomial Clostridium difficile Infection. Arch Intern Med2010; 170(9): 784–790.

15. Linsky A, Gupta K, Lawler EV, Fonda JR, Hermos JA. Proton Pump Inhibitorsand Risk for Recurrent Clostridium difficile Infection. Arch Intern Med 2010;170(9): 772–778.

16. Lowe DO, Mamdani MM, Kopp A, Low DE, Juurlink DN. Proton pump inhibi-tors and hospitalization for clostridium difficile-associated disease: a population-based study. Clin Infect Dis 2006; 43(10): 1272–1276.

17. IMS. Top Therapeutic Classes By US Sales, 2009. 2010. http://www.imshealth.com/deployedfiles/imshealth/Global/Content/StaticFile/Top_Line_Data/Top%20Therapy%20Classes%20by%20U.S.Sales.pdf[8 February 2011].

18. Craig DGN, Thimappa R, Anand V, Sebastian S. Inappropriate utilizationof intravenous proton pump inhibitors in hospital practice--a prospectivestudy of the extent of the problem and predictive factors. QJM 2010; 103(5):327–335.

19. Thomas L, Culley EJ, Gladowski P, Goff V, Fong J, Marche SM. Longitudinalanalysis of the costs associated with inpatient initiation and subsequent outpatientcontinuation of proton pump inhibitor therapy for stress ulcer prophylaxis in a largemanaged care organization. J Manag Care Pharm 2010; 16(2): 122–129.

20. Stevens V, Dumyati G, Fine LS, Fisher SG, van Wijngaarden E. CumulativeAntibiotic Exposures Over Time and the Risk of Clostridium difficile Infection.Clin Infect Dis 2011; 53(1): 42–48.

21. Therneau TM, Grambsch PM. Modeling Survival Data: Extending the CoxModel (Vol. 1). Springer: New York, 2000.

22. McGregor JC, Perencevich EN, Furuno JP, et al. Comorbidity risk-adjustmentmeasures were developed and validated for studies of antibiotic-resistant infec-tions. J Clin Epidemiol 2006; 59(12): 1266–1273.

23. Dubberke ER, Reske KA, Yan Y, Olsen MA, McDonald LC, Fraser VJ.Clostridium difficile--associated disease in a setting of endemicity: identificationof novel risk factors. Clin Infect Dis 2007; 45(12): 1543–1549.

24. Dubberke ER, Reske KA, Olsen MA, et al. Evaluation of Clostridium difficile-associated disease pressure as a risk factor for C difficile-associated disease. ArchIntern Med 2007; 167(10): 1092–1097.

25. Lin DY, Wei LJ. The Robust Inference for the Cox Proportional Hazards Model.J Am Stat Assoc 1989; 84(408): 1074–1078.

26. Kelly CP, LaMont JT. Clostridium difficile infection. Annu Rev Med 1998; 49:375–390.

27. Jump RL, Pultz MJ, Donskey CJ. Vegetative Clostridium difficile survives inroom air on moist surfaces and in gastric contents with reduced acidity: a



potential mechanism to explain the association between proton pump inhibitors andC. difficile-associated diarrhea? Antimicrob Agents Chemother 2007; 51(8):2883–2887.

28. Rao A, Jump RLP, Pultz NJ, Pultz MJ, Donskey CJ. In Vitro Killing of Nosoco-mial Pathogens by Acid and Acidified Nitrite. Antimicrob Agents Chemother2006; 50(11): 3901–3904.

29. Nerandzic MM, Pultz MJ, Donskey CJ. Examination of potential mechanismsto explain the association between proton pump inhibitors and Clostridiumdifficile infection. Antimicrob Agents Chemother 2009; 53(10): 4133–4137.

30. Zedtwitz-Liebenstein K, Wenisch C, Patruta S, Parschalk B, Daxbock F,Graninger W. Omeprazole treatment diminishes intra- and extracellular neutro-phil reactive oxygen production and bactericidal activity. Crit Care Med 2002;30(5): 1118–1122.

31. Kim JW, Lee KL, Jeong JB, et al. Proton pump inhibitors as a risk factor for re-currence of Clostridium-difficile-associated diarrhea. World J Gastroenterol2010; 16(28): 3573–3577.

32. Loo VG, Fallone CA, De Souza E, Lavallée J, Barkun AN. In-vitro susceptibilityof Helicobacter pylori to ampicillin, clarithromycin, metronidazole and omepra-zole. J Antimicrob Chemother 1997; 40(6): 881–883.

33. Sjostrom J, Kuhler T, Larsson H. Basis for the selective antibacterial activityin vitro of proton pump inhibitors against Helicobacter spp. Antimicrob AgentsChemother 1997; 41(8): 1797–1801.

34. Gatta L, Perna F, Figura N, et al. Antimicrobial activity of esomeprazole versusomeprazole against Helicobacter pylori. J Antimicrob Chemother 2003; 51(2):439–442.

35. Goddard AF, Spiller RC. The effect of omeprazole on gastric juice viscosity, pHand bacterial counts. Aliment Pharmacol Ther 1996; 10(1): 105–109.

36. Benini L, Castellani G, Bardelli E, et al. Omeprazole causes delay in gastricemptying of digestible meals. Dig Dis Sci 1996; 41(3): 469–474.

37. Ogawa R, Echizen H. Drug-Drug Interaction Profiles of Proton Pump Inhibitors.Clin Pharmacokinet 2010; 49(8): 509–533. 510.2165/11531320-000000000-000000000.

38. Ortiz RA, Calafatti SA, Moraes LA, et al. Effect of Helicobacter pylori infectionand acid blockade by lansoprazole on clarithromycin bioavailability. Braz J MedBiol Res 2007; 40(3): 383–389.

39. Calafatti SA, dos Santos A, da Silva CM, et al. Transfer of metronidazole togastric juice: impact of Helicobacter pylori infection and omeprazole. ScandJ Gastroenterol 2000; 35(7): 699–704.

40. Hughes GS, Heald DL, Barker KB, et al. The effects of gastric pH and food onthe pharmacokinetics of a new oral cephalosporin, cefpodoxime proxetil. ClinPharmacol Ther 1989; 46(6): 674–685.



differential risk of clostridium difficile infection with proton pump inhibitor use by level of...

Documents