[Skip to Navigation]
Sign In
Figure 1. 
Kaplan-Meier rates of nosocomial Clostridium difficile infection during the hospitalization

Kaplan-Meier rates of nosocomial Clostridium difficile infection during the hospitalization

Figure 2. 
Rates of Clostridium difficile infection stratified by the type of antibiotics and acid-suppressive therapy. High-risk antibiotics included fluoroquinolones, cephalosporins, intravenous β-lactam/β-lactamase inhibitors, macrolides, clindamycin, and carbapenems. All other antibiotics were classified as low risk. H2RA indicates H2-receptor antagonist; and PPI, proton pump inhibitor.

Rates of Clostridium difficile infection stratified by the type of antibiotics and acid-suppressive therapy. High-risk antibiotics included fluoroquinolones, cephalosporins, intravenous β-lactam/β-lactamase inhibitors, macrolides, clindamycin, and carbapenems.13,22,23 All other antibiotics were classified as low risk. H2RA indicates H2-receptor antagonist; and PPI, proton pump inhibitor.

Figure 3. 
Multivariable models assessing the relationship of acid-suppressive therapy and nosocomial Clostridium difficile infection. A, Cohort analysis adjusted for propensity score. The generalized estimating equation (GEE) model included antibiotics, age, sex, length of hospital stay, and baseline comorbidity status. Results were adjusted for the propensity to receive more-intense acid-suppressive therapy. B, Pair-wise cohort analysis matched for propensity score. Three pairs of cohorts matched by propensity score were analyzed (no acid suppression/H2RA only; no acid suppression/daily PPI; and no acid suppression/PPI more frequently than daily). Results were also adjusted for antibiotic therapy, age, sex, and baseline comorbidity status. C, Nested matched case-control analysis. Patients with C difficile infection were matched at a 1:2 ratio by a nearest-neighbor–matching algorithm with a “greedy” heuristic to patients without infection. Matching criteria included diagnosis related group, Charlson index with caliper of 1 point, age with caliper of 5 years, and propensity score. Conditional logistic regression with adjustment for comorbidities and sex was fitted. H2RA indicates H2-receptor antagonist; PPI, proton pump inhibitor. Error bars represent 95% confidence intervals.

Multivariable models assessing the relationship of acid-suppressive therapy and nosocomial Clostridium difficile infection. A, Cohort analysis adjusted for propensity score. The generalized estimating equation (GEE) model included antibiotics, age, sex, length of hospital stay, and baseline comorbidity status. Results were adjusted for the propensity to receive more-intense acid-suppressive therapy. B, Pair-wise cohort analysis matched for propensity score. Three pairs of cohorts matched by propensity score were analyzed (no acid suppression/H2RA only; no acid suppression/daily PPI; and no acid suppression/PPI more frequently than daily). Results were also adjusted for antibiotic therapy, age, sex, and baseline comorbidity status. C, Nested matched case-control analysis. Patients with C difficile infection were matched at a 1:2 ratio by a nearest-neighbor–matching algorithm with a “greedy” heuristic to patients without infection. Matching criteria included diagnosis related group, Charlson index with caliper of 1 point, age with caliper of 5 years, and propensity score. Conditional logistic regression with adjustment for comorbidities and sex was fitted. H2RA indicates H2-receptor antagonist; PPI, proton pump inhibitor. Error bars represent 95% confidence intervals.

Table 1. 
Clinical and Demographic Characteristics Stratified by the Diagnosis of Nosocomial Clostridium difficilea
Clinical and Demographic Characteristics Stratified by the Diagnosis of Nosocomial Clostridium difficilea
Table 2. 
Multivariable Analysis for Factors Associated With Nosocomial Clostridium difficile Infectiona
Multivariable Analysis for Factors Associated With Nosocomial Clostridium difficile Infectiona
1.
Blossom  DBMcDonald  LC The challenges posed by reemerging Clostridium difficile infection.  Clin Infect Dis 2007;45 (2) 222- 227PubMedGoogle Scholar
2.
Kelly  CPLaMont  JT Clostridium difficile—more difficult than ever.  N Engl J Med 2008;359 (18) 1932- 1940PubMedGoogle Scholar
3.
Kyne  LHamel  MBPolavaram  RKelly  CP Health care costs and mortality associated with nosocomial diarrhea due to Clostridium difficile Clin Infect Dis 2002;34 (3) 346- 353PubMedGoogle Scholar
4.
Centers for Disease Control and Prevention (CDC), Severe Clostridium difficile–associated disease in populations previously at low risk—four states, 2005.  MMWR Morb Mortal Wkly Rep 2005;54 (47) 1201- 1205PubMedGoogle Scholar
5.
Vonberg  RPKuijper  EJWilcox  MH  et al. European C difficile–Infection Control Group; European Centre for Disease Prevention and Control (ECDC), Infection control measures to limit the spread of Clostridium difficile Clin Microbiol Infect 2008;14 ((suppl 5)) 2- 20PubMedGoogle Scholar
6.
Gurian  LWard  TTKaton  RM Possible foodborne transmission in a case of pseudomembranous colitis due to Clostridium difficile: influence of gastrointestinal secretions on Clostridium difficile infection.  Gastroenterology 1982;83 (2) 465- 469PubMedGoogle Scholar
7.
Kaur  SVaishnavi  CPrasad  KKRay  PKochhar  R Comparative role of antibiotic and proton pump inhibitor in experimental Clostridium difficile infection in mice.  Microbiol Immunol 2007;51 (12) 1209- 1214PubMedGoogle Scholar
8.
Aseeri  MSchroeder  TKramer  JZackula  R Gastric acid suppression by proton pump inhibitors as a risk factor for Clostridium difficile–associated diarrhea in hospitalized patients.  Am J Gastroenterol 2008;103 (9) 2308- 2313PubMedGoogle Scholar
9.
Choudhry  MNSoran  HZiglam  HM Overuse and inappropriate prescribing of proton pump inhibitors in patients with Clostridium difficile–associated disease.  QJM 2008;101 (6) 445- 448PubMedGoogle Scholar
10.
Dial  SAlrasadi  KManoukian  CHuang  AMenzies  D Risk of Clostridium difficile diarrhea among hospital inpatients prescribed proton pump inhibitors: cohort and case-control studies.  CMAJ 2004;171 (1) 33- 38PubMedGoogle Scholar
11.
Lowe  DOMamdani  MMKopp  ALow  DEJuurlink  DN Proton pump inhibitors and hospitalization for Clostridium difficile–associated disease: a population-based study.  Clin Infect Dis 2006;43 (10) 1272- 1276PubMedGoogle Scholar
12.
Beaulieu  MWilliamson  DPichette  GLachaine  J Risk of Clostridium difficile–associated disease among patients receiving proton-pump inhibitors in a Quebec medical intensive care unit.  Infect Control Hosp Epidemiol 2007;28 (11) 1305- 1307PubMedGoogle Scholar
13.
Pépin  JSaheb  NCoulombe  MA  et al.  Emergence of fluoroquinolones as the predominant risk factor for Clostridium difficile–associated diarrhea: a cohort study during an epidemic in Quebec.  Clin Infect Dis 2005;41 (9) 1254- 1260PubMedGoogle Scholar
14.
Shah  SLewis  ALeopold  DDunstan  FWoodhouse  K Gastric acid suppression does not promote clostridial diarrhoea in the elderly.  QJM 2000;93 (3) 175- 181PubMedGoogle Scholar
15.
Wilcox  MHMooney  LBendall  RSettle  CDFawley  WN A case-control study of community-associated Clostridium difficile infection.  J Antimicrob Chemother 2008;62 (2) 388- 396PubMedGoogle Scholar
16.
Metz  DC Clostridium difficile colitis: wash your hands before stopping the proton pump inhibitor.  Am J Gastroenterol 2008;103 (9) 2314- 2316PubMedGoogle Scholar
17.
Hill  AB The environment and disease: association or causation?  Proc R Soc Med 1965;58295- 300PubMedGoogle Scholar
18.
Rothman  KJGreenland  S Causation and causal inference in epidemiology.  Am J Public Health 2005;95 ((suppl 1)) S144- S150PubMedGoogle Scholar
19.
Swaen  Gvan Amelsvoort  L A weight of evidence approach to causal inference.  J Clin Epidemiol 2009;62 (3) 270- 277PubMedGoogle Scholar
20.
Huang  JQHunt  RH Pharmacological and pharmacodynamic essentials of H(2)-receptor antagonists and proton pump inhibitors for the practising physician.  Best Pract Res Clin Gastroenterol 2001;15 (3) 355- 370PubMedGoogle Scholar
21.
World Health Organization, International language for drug utilization research. 2009;http://www.whocc.no/. Accessed January 30, 2010
22.
Baxter  RRay  GTFireman  BH Case-control study of antibiotic use and subsequent Clostridium difficile–associated diarrhea in hospitalized patients.  Infect Control Hosp Epidemiol 2008;29 (1) 44- 50PubMedGoogle Scholar
23.
Winston  DJLazarus  HMBeveridge  RA  et al.  Randomized, double-blind, multicenter trial comparing clinafloxacin with imipenem as empirical monotherapy for febrile granulocytopenic patients.  Clin Infect Dis 2001;32 (3) 381- 390PubMedGoogle Scholar
24.
Elixhauser  ASteiner  CHarris  DRCoffey  RM Comorbidity measures for use with administrative data.  Med Care 1998;36 (1) 8- 27PubMedGoogle Scholar
25.
Charlson  MEPompei  PAles  KLMacKenzie  CR A new method of classifying prognostic comorbidity in longitudinal studies: development and validation.  J Chronic Dis 1987;40 (5) 373- 383PubMedGoogle Scholar
26.
Quan  HSundararajan  VHalfon  P  et al.  Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data.  Med Care 2005;43 (11) 1130- 1139PubMedGoogle Scholar
27.
Moss  RRHumphries  KHGao  M  et al.  Outcome of mitral valve repair or replacement: a comparison by propensity score analysis.  Circulation 2003;108 ((suppl 1)) II90- II97PubMedGoogle Scholar
28.
Austin  PC Some methods of propensity-score matching had superior performance to others: results of an empirical investigation and Monte Carlo simulations.  Biom J 2009;51 (1) 171- 184PubMedGoogle Scholar
29.
Zhang  JYu  KF What's the relative risk? a method of correcting the odds ratio in cohort studies of common outcomes.  JAMA 1998;280 (19) 1690- 1691PubMedGoogle Scholar
30.
Leffler  DCloud  JWKelly  CP Gastric acid–suppressive agents and risk of Clostridium difficile–associated disease.  JAMA 2006;295 (22) 2599- 2600PubMedGoogle Scholar
31.
Dial  SDelaney  JACBarkun  ANSuissa  S Gastric acid–suppressive agents and risk of Clostridium difficile–associated disease [in reply].  JAMA 2006;295 (22) 2600- 2591Google Scholar
32.
Perkins  SMTu  WUnderhill  MGZhou  XHMurray  MD The use of propensity scores in pharmacoepidemiologic research.  Pharmacoepidemiol Drug Saf 2000;9 (2) 93- 101PubMedGoogle Scholar
33.
Pham  CQRegal  REBostwick  TRKnauf  KS Acid suppressive therapy use on an inpatient internal medicine service.  Ann Pharmacother 2006;40 (7-8) 1261- 1266PubMedGoogle Scholar
34.
Parente  FCucino  CGallus  S  et al.  Hospital use of acid-suppressive medications and its fall-out on prescribing in general practice: a 1-month survey.  Aliment Pharmacol Ther 2003;17 (12) 1503- 1506PubMedGoogle Scholar
35.
Nardino  RJVender  RJHerbert  PN Overuse of acid-suppressive therapy in hospitalized patients.  Am J Gastroenterol 2000;95 (11) 3118- 3122PubMedGoogle Scholar
36.
Health Care Utilization Project, Introduction to the HCUP Nationwide Inpatient Sample (NIS). 2007;http://www.hcup-us.ahrq.gov/db/nation/nis/NIS_2007_INTRODUCTION.pdf. Accessed August 15, 2009
37.
Janicki  TStewart  S Stress-ulcer prophylaxis for general medical patients: a review of the evidence.  J Hosp Med 2007;2 (2) 86- 92PubMedGoogle Scholar
38.
Cook  DJFuller  HDGuyatt  GH  et al. Canadian Critical Care Trials Group, Risk factors for gastrointestinal bleeding in critically ill patients.  N Engl J Med 1994;330 (6) 377- 381PubMedGoogle Scholar
39.
ASHP Therapeutic Guidelines on Stress Ulcer Prophylaxis, ASHP Therapeutic Guidelines on Stress Ulcer Prophylaxis: ASHP Commission on Therapeutics and approved by the ASHP Board of Directors on November 14, 1998.  Am J Health Syst Pharm 1999;56 (4) 347- 379PubMedGoogle Scholar
40.
Farrell  CPMercogliano  GKuntz  CL Overuse of stress ulcer prophylaxis in the critical care setting and beyond [published online ahead of print August 14, 2009].  J Crit Care 10.1016/j.jcre.2009.05.014PubMedGoogle Scholar
41.
Murphy  CEStevens  AMFerrentino  N  et al.  Frequency of inappropriate continuation of acid suppressive therapy after discharge in patients who began therapy in the surgical intensive care unit.  Pharmacotherapy 2008;28 (8) 968- 976PubMedGoogle Scholar
42.
Wohlt  PDHansen  LAFish  JT Inappropriate continuation of stress ulcer prophylactic therapy after discharge.  Ann Pharmacother 2007;41 (10) 1611- 1616PubMedGoogle Scholar
Original Investigation
Less Is More
May 10, 2010

Iatrogenic Gastric Acid Suppression and the Risk of Nosocomial Clostridium difficile Infection

Arch Intern Med. 2010;170(9):784-790. doi:10.1001/archinternmed.2010.89
Abstract

Background  The incidence and severity of Clostridium difficile infections are increasing. Acid-suppressive therapy has been suggested as a risk factor for C difficile, but this remains controversial.

Methods  We conducted a pharmacoepidemiologic cohort study, performing a secondary analysis of data collected prospectively on 101 796 discharges from a tertiary care medical center during a 5-year period. The primary exposure of interest was acid suppression therapy, classified by the most intense acid suppression therapy received (no acid suppression, histamine2-receptor antagonist [H2RA] therapy, daily proton pump inhibitor [PPI], and PPI more frequently than daily).

Results  As the level of acid suppression increased, the risk of nosocomial C difficile infection increased, from 0.3% (95% confidence interval [CI], 0.21%-0.31%) in patients not receiving acid suppressive therapy to 0.6% (95% CI, 0.49%-0.79%) in those receiving H2RA therapy, to 0.9% (95% CI, 0.80%-0.98%) in those receiving daily PPI treatment, and to 1.4% (1.15%-1.71%) in those receiving more frequent PPI therapy. After adjustment for comorbid conditions, age, antibiotics, and propensity score–based likelihood of receipt of acid-suppression therapy, the association persisted, increasing from an odds ratio of 1 (no acid suppression [reference]) to 1.53 (95% CI, 1.12-2.10) (H2RA), to 1.74 (95% CI, 1.39-2.18) (daily PPI), and to 2.36 (95% CI, 1.79-3.11) (more frequent PPI). Similar estimates were found with a matched cohort analysis and with nested case-control techniques.

Conclusions  Increasing levels of pharmacologic acid suppression are associated with increased risks of nosocomial C difficile infection. This evidence of a dose-response effect provides further support for the potentially causal nature of iatrogenic acid suppression in the development of nosocomial C difficile infection.

Clostridium difficile infection poses a clear and escalating threat to public health. Incidence and severity of disease are increasing,1,2 and costs total more than $1 billion in the United States annually.3 Although other populations are at risk,4 most cases remain both nosocomial and iatrogenic.2 Prevention is therefore paramount, but preventive measures have generally focused on isolation of infected patients, hand hygiene, and antibiotic stewardship.5

Recently, attention has turned to the possible contribution of proton pump inhibitor (PPI) use to the acquisition of C difficile. This association is biologically plausible. More than 2 decades ago, Gurian et al6 showed that more-acidic gastric juices were more effective than less-acidic gastric juices in killing C difficile and neutralizing its toxin. Mice receiving PPIs have similar susceptibility to C difficile infection as mice receiving antibiotics.7 More recently, some,8-11 but not all,11-15 epidemiologic studies have suggested an association between C difficile infection and PPI use, both in the inpatient and outpatient setting. However, a recent review summarized the state of the literature regarding PPI and C difficile infection as remaining “controversial1”; another recent prominent C difficile review did not mention PPIs at all.2

Moreover, little work has addressed whether a dose-response effect exists between acid suppression and C difficile infection,16 in spite of the fact that this is often viewed as one of the epidemiologic pillars of causal inference.17-19 Therefore, we sought to examine the relationship between increasing levels of pharmacologic acid suppression and nosocomial acquisition of C difficile in a large pharmacoepidemiologic cohort.

Methods
Setting and design

This was a pharmacoepidemiologic cohort study, in which we performed a secondary analysis of data prospectively collected for other reasons on patients discharged between January 1, 2004, and January 31, 2008, at the Beth Israel Deaconess Medical Center, a large, urban, tertiary care center in Boston, Massachusetts. The hospital's institutional review board approved the study with a waiver of informed consent.

Data sources

Data were obtained from electronic medical databases created as part of usual care. These databases contain information from each admission, such as demographics, discharge diagnosis codes, medication orders, microbiologic results, length of stay, and inpatient mortality.

Patients and definitions

All patients who were at least 18 years old and had a length of stay of 3 or more days were included. The outcome of interest, nosocomial C difficile infection, was defined as a newly positive C difficile toxin assay result on or after the third hospital day, a definition used by others.8 Only a first diagnosis of C difficile was included; subsequent admissions of patients with index C difficile infection were excluded.

The primary exposure of interest was receipt of acid suppression therapy. Exposure was classified by the most intense acid suppression therapy received before a positive C difficile test result or hospital discharge, whichever was earlier. The 4 a priori acid suppression groups were no acid suppression therapy, histamine2-receptor antagonist (H2RA) therapy, daily PPI, and PPI more frequently than daily. We used these classifications because they result in stepwise increases in gastric acid suppression20 and because they represent common clinical therapeutic approaches. Because of formulary selections in our hospital, clinical dosing of PPIs closely parallels the defined daily dose21; more than 98% of daily PPIs were equivalent to the defined daily dose.

We collected other potentially important predictors of C difficile infection. We assessed antibiotics received during the hospitalization before a diagnosis of C difficile infection or discharge, categorizing patients as having received no antibiotics, low-risk antibiotics, or high-risk antibiotics. High-risk antibiotics were identified based on the medical literature and included fluoroquinolones, cephalosporins, intravenous β-lactam/β-lactamase inhibitors, macrolides, clindamycin, and carbapenems.13,22,23 Other antibiotics were classified as low risk. To classify patients' comorbidity status, we used the method of Elixhauser et al24 for classification of specific comorbid conditions; we used the Charlson Comorbidity Index25,26 to represent the cumulative burden of comorbid illness.

Statistical methods

We performed unadjusted comparisons using the t test, the Mann-Whitney test, the χ2 test, or the Fisher exact test, as appropriate. We used the Kaplan-Meier method to assess time to the C difficile diagnosis. Statistical analyses were performed with SAS software, version 9.1 (SAS Institute Inc, Cary, North Carolina) and SPSS, version 16.0.1 (SPSS Inc, Chicago, Illinois).

Propensity score

Patients for whom physicians prescribe acid-suppressive medications are likely to differ systematically from those who do not receive the medications, resulting in potential selection bias. Thus, in addition to multivariable adjustment for comorbidities, we applied a propensity score adjustment technique. Since the method for propensity score construction for multilevel variables is less well established, as a first step we created a binary propensity score to reflect the probability of receiving the strongest level of acid suppression (PPI more often than daily) compared with no acid suppression therapy. The propensity score was derived from a logistic generalized estimating equation (GEE) model, which included 19 clinical and demographic variables. We then applied the propensity score equation to the rest of the study cohort (the H2RA only and daily PPI groups). Finally, we verified that the mean propensity scores of the 4 exposure groups were correctly ordered (ie, that there was a linear increase from the lowest level of acid-suppressive therapy to the highest). The propensity score was then included in the final logistic GEE regression model, with C difficile infection as the dependent variable.

Because each patient could have been admitted to the hospital more than once during the study period, we used a GEE model to address patient-level clustering, with a dependent variable of nosocomial C difficile infection. To avoid model overfitting, forward stepwise selection was used to identify potentially significant predictors (P < .10) for retention in the model. Model discrimination was assessed using the area under the receiver operating characteristics curve; model fit was tested using the Hosmer-Lemeshow statistic (calculated under the assumption of independent correlation structure of the model). The model included exposure to acid suppression, exposure to prior antibiotics, age, sex, comorbid illness, propensity score, and log-transformed length of stay as an offset variable. Since acquiring C difficile infection results in a longer length of stay, for C difficile–positive patients, length of stay was substituted with time to the diagnosis.

Additional analytic approaches

Because true findings should be resilient to the analytic technique applied, we also analyzed the relationship between acid suppression and C difficile infection using 2 additional approaches.

Cohort Propensity Score Matching

We defined patients with no acid suppression therapy as a reference group. Then all other patients (those with varying acid suppression therapy levels) were matched 1:1 with the reference cohort using a propensity score with caliper of 0.10.27,28 A nearest-neighbor–matching algorithm with a “greedy” heuristic (one that always implements the best immediate or local solution) was applied. In greedy nearest-neighbor matching, a patient from the treatment cohort was randomly selected, and matching was attempted with the “nearest” patient from the reference group. This process was repeated until matches had been attempted for all patients in each acid suppression group. Each matched pair was unique, and data for unmatched patients were not used in subsequent analyses. By this approach, we created 3 matched cohorts: no acid suppression/H2RA; no acid suppression/daily PPI; and no acid suppression/PPI more frequently than daily. We applied a logistic GEE regression in each of the matched cohort pairs.

Nested Matched Case Control

Patients with C difficile infection were matched at a 1:2 ratio by a nearest-neighbor–matching algorithm with a “greedy” heuristic to patients without C difficile infection on the basis of their diagnosis related group, Charlson index with caliper of 1 point, age with caliper of 5 years, and propensity score with caliper of 0.10.27,28 Only cases with 2 controls matched were included in the analysis. Only the first admission for each patient was included. Multivariable conditional logistic regression analysis was then performed with inclusion of 4 groups of acid suppression therapy and adjustment for sex and number of comorbidities.

Lastly, because some hospital services are often believed to be very low risk for C difficile infection, we performed a sensitivity analysis that excluded patients admitted to the obstetrics and psychiatric services.

Results

The study period included 174 224 admissions and 1344 cases of toxin-positive C difficile infection. Of these admissions, 72 428 did not meet inclusion criteria: 72 013 had a length of stay less than 3 days, 447 involved patients who were younger than 18 years, and 177 involved patients who had been diagnosed as having C difficile during prior admissions. Some patients had more than 1 exclusion criterion. Thus, 101 796 admissions were eligible for analysis. Nosocomial C difficile infection occurred in 665 cases (0.7% of admissions). Patients with nosocomial C difficile infection were older, more likely to be male, and more likely to have comorbid conditions such congestive heart failure, renal failure, and cancer. Exposure to antibiotics was strongly associated with development of C difficile, and nosocomial C difficile infection was associated with higher mortality rates (Table 1). Length of stay was strongly associated with the rate of nosocomial C difficile infection (Figure 1): patients hospitalized for less than 7 days had a lower rate of C difficile infection than patients with longer stays (0.4% vs 1.1%; P < .001).

Propensity score

The final propensity score included 19 variables that differentiated between the probability of receiving maximal acid suppression therapy and no acid suppression: age, sex, history of heart failure, chronic lung disease, diabetes mellitus, hypothyroidism, neurologic disorders, chronic liver diseases, gastric ulcer, malignancy, connective tissue diseases, coagulation disorders, AIDS/human immunodeficiency virus, chronic renal failure, obesity, history of weight loss, anemia, alcohol abuse, and history of blood loss. Applying the propensity score formula to all 4 acid suppression exposure groups, from lowest to highest, yielded mean propensity scores of 0.15, 0.19, 0.26, and 0.32. Simple regression demonstrated a strong linear trend based on propensity score analysis, further underscoring the adequacy of our approach.

RISK OF NOSOCOMIAL C DIFFICILE INFECTION

In unadjusted analyses, increasing levels of acid-suppressive therapy were associated with increasing rates of nosocomial C difficile infection. As the level of acid suppression increased, the risk of developing nosocomial C difficile infection increased, from 0.3% (95% confidence interval [CI], 0.21%-0.31%) to 0.6% (95% CI, 0.49%-0.79%), to 0.9% (0.80%-0.98%), to 1.4% (1.15%-1.71%). Antibiotic therapy strongly predicted C difficile infection: 0.2% in patients who did not receive antibiotics, 0.4% in patients treated with low-risk antibiotics, and 1.1% in patients who received high-risk antibiotics (P < .001). Figure 2 shows the risk of C difficile infection, simultaneously stratified by acid suppression therapy and antibiotics.

After adjustment for comorbid conditions, age, receipt of antibiotics, and propensity score–based likelihood of receipt of acid suppression therapy, the association persisted. In our main analysis, which adjusted for propensity score as a continuous variable, we found that as the level of acid suppression increased, the adjusted odds of developing C difficile infection also increased, from an odds ratio (OR) of 1 (reference) to 1.53 (95% CI, 1.12-2.10) for H2RA only, to 1.74 (95% CI, 1.39-2.18) for daily PPI, and to 2.36 (95% CI, 1.79-3.11) for PPI more frequently than daily (Table 2). Because the incidence of C difficile infection in our unexposed cohort is low (0.25%), adjusted ORs are close approximations of adjusted relative risks.29

In addition, receipt of prior antibiotics, age, and comorbid conditions were important risk factors for nosocomial C difficile. Low-risk antibiotics were associated with a near doubling of the odds of C difficile infection, and high-risk antibiotics with a more than 3-fold increase in the odds of disease. The model had good discrimination (area under the receiver operating characteristics curve, 0.77 [95% CI, 0.76-0.79]) and good calibration (P = .64, Hosmer-Lemeshow test).

To ascertain whether our results might be an artifact of the analytic method chosen, we used additional analytic approaches to verify our findings. In our main analysis, we used the entire cohort and adjusted for propensity to receive high-level acid suppression, treating this as a continuous variable. In our first additional approach, we conducted a cohort analysis matched by propensity score. In our second additional approach, we conducted a nested, matched case-control analysis. The results of these different approaches yielded results very similar to the main analysis (Figure 3).

We performed a sensitivity analysis, excluding patients admitted to the psychiatric and obstetrics services, which have low risks of nosocomial C difficile infection. The C difficile infection rate in the remaining cohort of 80 906 admissions was 0.8%. Multivariable GEE analysis demonstrated that the association between acid suppression and C difficile persisted. Compared with no acid suppression, H2RA treatment was associated with an OR of 1.29 (95% CI, 0.94-1.67); daily PPI, with an OR of 1.47 (95% CI, 1.18-1.84); and PPI more frequently than daily, with an OR of 1.98 (95% CI, 1.51-2.59).

Comment

This study demonstrates that increasing levels of pharmacologic acid suppression are independently associated with an increased risk of nosocomial C difficile infection. The strength of the association is both clinically and statistically meaningful. Compared with no acid suppression, receipt of a daily PPI was associated with a more than 70% increase in the odds of developing C difficile; patients who received more frequent PPIs had a more than doubling of risk.

We attempted to design a study that would contribute meaningfully to the debate about whether acid suppression contributes causally to nosocomial C difficile infection or is simply a confounder. Although a substantial number of studies have assessed the relationship between PPI use and C difficile infection, with findings on both sides of the issue,8-15 we found none that addressed the question of dose-response effect (biological gradient). In fact, a recent editorial applied all of the commonly used causal criteria first laid out by Sir Austin Bradford Hill17 to the question and found that “none of the studies published to date . . . have yet demonstrated a dose-response curve, whereby a stronger association between PPI exposure and the development of colitis was noted in individuals receiving higher doses of PPIs (biological gradient).”16(p2315) We therefore focused on a key element of the epidemiologic ascertainment of causality—biological gradient.19 Our study provides clear evidence of just such a dose-response effect: more intense acid suppression is associated with increased risks of C difficile infection. In addition to addressing the issue of biological gradient, our results also provide further evidence in support of the temporality, consistency, and strength of the association.

Our study has several additional strengths. First, our cohort is large and includes all consecutive adults meeting inclusion criteria, limiting the likelihood of selection bias. Second, we ensured that acid suppression and antibiotic exposure occurred before development of C difficile, addressing the temporality criterion of Bradford Hill.17 Third, we adjusted for the burden of comorbid disease, a criticism of some prior studies.30,31 Fourth, we controlled for the likelihood of administration of more-intense acid suppression therapy using techniques commonly recommended in pharmacoepidemiology.32 Finally, our results were resilient to the specific analytic technique applied, with remarkably similar estimates of the dose-response relationship.

Our study also has several key limitations. First, it remains an observational study. A randomized design would provide greater confidence in our findings. However, conducting a randomized controlled trial to assess whether a therapy causes C difficile infection would ethically be problematic. In addition, because the acquisition of C difficile is a relatively rare event, such a trial would be very large and therefore costly. Second, there may be residual confounding. We attempted to control for selection bias using careful application of propensity score–based techniques and further adjusted for significant confounders, but unmeasured variables may still cause residual confounding. Effect estimates, as expected, moved closer to the null as we adjusted for covariates, but the magnitude of effect still remains plausibly large. Third, we were unable to collect information about use of acid-suppressive medications or antibiotics before admission, which could be an important effect modifier.

These results may have important public health implications. Our results suggest that, compared with no acid suppression, we should expect at least 1 additional case of nosocomial C difficile infection for every 533 patients who receive a daily PPI, after controlling for other risk factors. (This is based on a nosocomial C difficile infection incidence of 0.3% in patients receiving no acid suppression and an adjusted increase of 70% in odds of disease in patients receiving daily PPIs.) Although this seems like a relatively large number-needed-to-harm, the magnitude of exposure is large. We found that 60% of patients received acid-suppressive therapy, similar to others' estimates.33-35 Since there are about 32.7 million annual adult discharges in the United States annually,36 the number of potentially attributable nosocomial C difficile cases in the United States numbers in the tens of thousands per year. This is particularly important, since prior work has found that more than two-thirds of inpatient acid-suppressive prescriptions are not strictly indicated.33-35

In conclusion, we found that the intensity of acid-suppressive therapy was strongly and independently associated with the development of nosocomial C difficile infection. Although determining causality is an extremely complex process,18 it seems unlikely that a randomized controlled trial of acid suppressive therapy, with a primary outcome of nosocomial C difficile infection, will ever be conducted. Our study therefore adds new key information about the dose-response relationship between acid-suppressive medications and nosocomial C difficile infection. Given the importance of nosocomial C difficile infection to individual patient outcomes, hospitals, health systems, and physicians can take several reasonable, important steps even in the absence of randomized controlled trials. First, physicians should ensure that each patient receives the least-intense acid-suppressive therapy that is appropriate for his or her clinical condition. In particular, unless and until there is clear evidence that low-risk, noncritically ill patients receive benefit from stress ulcer prophylaxis,37 we should strive to minimize exposure to acid-suppressive medications in this group. Many intensive care unit patients are also at low risk for clinically significant gastrointestinal tract bleeding,38 and guidelines have not recommend prophylaxis for these patients.39 Hospitals can also take an important step in this regard by re-examining any standing protocols for low-risk patients that include stress ulcer prophylaxis by default and by ensuring that prophylactic medications are not continued beyond discharge.40-42 Finally, future interventional trials of acid-suppressive medications in inpatients should actively collect information about incident C difficile infection and other adverse events. Researchers can also further help reduce uncertainty in this field by using risk-benefit analysis to create tools that help select patients in whom stress ulcer prophylaxis may be clearly beneficial or clearly harmful.

Correspondence: Michael D. Howell, MD, MPH, Critical Care Quality, Silverman Institute for Healthcare Quality and Safety, Beth Israel Deaconess Medical Center, 330 Brookline Ave, Boston, MA 02215 (mhowell@bidmc.harvard.edu).

Accepted for Publication: February 10, 2010.

Author Contributions:Study concept and design: Howell, V. Novack, Grgurich, Soulliard, L. Novack, and Talmor. Acquisition of data: Howell. Analysis and interpretation of data: Howell, V. Novack, L. Novack, Pencina, and Talmor. Drafting of the manuscript: Howell, V. Novack, Grgurich, Soulliard, and Talmor. Critical revision of the manuscript for important intellectual content: V. Novack, Grgurich, Pencina, and Talmor. Statistical analysis: Howell, V. Novack, L. Novack, and Pencina. Administrative, technical, and material support: Grgurich, Soulliard, and Talmor. Study supervision: Howell.

References
1.
Blossom  DBMcDonald  LC The challenges posed by reemerging Clostridium difficile infection.  Clin Infect Dis 2007;45 (2) 222- 227PubMedGoogle Scholar
2.
Kelly  CPLaMont  JT Clostridium difficile—more difficult than ever.  N Engl J Med 2008;359 (18) 1932- 1940PubMedGoogle Scholar
3.
Kyne  LHamel  MBPolavaram  RKelly  CP Health care costs and mortality associated with nosocomial diarrhea due to Clostridium difficile Clin Infect Dis 2002;34 (3) 346- 353PubMedGoogle Scholar
4.
Centers for Disease Control and Prevention (CDC), Severe Clostridium difficile–associated disease in populations previously at low risk—four states, 2005.  MMWR Morb Mortal Wkly Rep 2005;54 (47) 1201- 1205PubMedGoogle Scholar
5.
Vonberg  RPKuijper  EJWilcox  MH  et al. European C difficile–Infection Control Group; European Centre for Disease Prevention and Control (ECDC), Infection control measures to limit the spread of Clostridium difficile Clin Microbiol Infect 2008;14 ((suppl 5)) 2- 20PubMedGoogle Scholar
6.
Gurian  LWard  TTKaton  RM Possible foodborne transmission in a case of pseudomembranous colitis due to Clostridium difficile: influence of gastrointestinal secretions on Clostridium difficile infection.  Gastroenterology 1982;83 (2) 465- 469PubMedGoogle Scholar
7.
Kaur  SVaishnavi  CPrasad  KKRay  PKochhar  R Comparative role of antibiotic and proton pump inhibitor in experimental Clostridium difficile infection in mice.  Microbiol Immunol 2007;51 (12) 1209- 1214PubMedGoogle Scholar
8.
Aseeri  MSchroeder  TKramer  JZackula  R Gastric acid suppression by proton pump inhibitors as a risk factor for Clostridium difficile–associated diarrhea in hospitalized patients.  Am J Gastroenterol 2008;103 (9) 2308- 2313PubMedGoogle Scholar
9.
Choudhry  MNSoran  HZiglam  HM Overuse and inappropriate prescribing of proton pump inhibitors in patients with Clostridium difficile–associated disease.  QJM 2008;101 (6) 445- 448PubMedGoogle Scholar
10.
Dial  SAlrasadi  KManoukian  CHuang  AMenzies  D Risk of Clostridium difficile diarrhea among hospital inpatients prescribed proton pump inhibitors: cohort and case-control studies.  CMAJ 2004;171 (1) 33- 38PubMedGoogle Scholar
11.
Lowe  DOMamdani  MMKopp  ALow  DEJuurlink  DN Proton pump inhibitors and hospitalization for Clostridium difficile–associated disease: a population-based study.  Clin Infect Dis 2006;43 (10) 1272- 1276PubMedGoogle Scholar
12.
Beaulieu  MWilliamson  DPichette  GLachaine  J Risk of Clostridium difficile–associated disease among patients receiving proton-pump inhibitors in a Quebec medical intensive care unit.  Infect Control Hosp Epidemiol 2007;28 (11) 1305- 1307PubMedGoogle Scholar
13.
Pépin  JSaheb  NCoulombe  MA  et al.  Emergence of fluoroquinolones as the predominant risk factor for Clostridium difficile–associated diarrhea: a cohort study during an epidemic in Quebec.  Clin Infect Dis 2005;41 (9) 1254- 1260PubMedGoogle Scholar
14.
Shah  SLewis  ALeopold  DDunstan  FWoodhouse  K Gastric acid suppression does not promote clostridial diarrhoea in the elderly.  QJM 2000;93 (3) 175- 181PubMedGoogle Scholar
15.
Wilcox  MHMooney  LBendall  RSettle  CDFawley  WN A case-control study of community-associated Clostridium difficile infection.  J Antimicrob Chemother 2008;62 (2) 388- 396PubMedGoogle Scholar
16.
Metz  DC Clostridium difficile colitis: wash your hands before stopping the proton pump inhibitor.  Am J Gastroenterol 2008;103 (9) 2314- 2316PubMedGoogle Scholar
17.
Hill  AB The environment and disease: association or causation?  Proc R Soc Med 1965;58295- 300PubMedGoogle Scholar
18.
Rothman  KJGreenland  S Causation and causal inference in epidemiology.  Am J Public Health 2005;95 ((suppl 1)) S144- S150PubMedGoogle Scholar
19.
Swaen  Gvan Amelsvoort  L A weight of evidence approach to causal inference.  J Clin Epidemiol 2009;62 (3) 270- 277PubMedGoogle Scholar
20.
Huang  JQHunt  RH Pharmacological and pharmacodynamic essentials of H(2)-receptor antagonists and proton pump inhibitors for the practising physician.  Best Pract Res Clin Gastroenterol 2001;15 (3) 355- 370PubMedGoogle Scholar
21.
World Health Organization, International language for drug utilization research. 2009;http://www.whocc.no/. Accessed January 30, 2010
22.
Baxter  RRay  GTFireman  BH Case-control study of antibiotic use and subsequent Clostridium difficile–associated diarrhea in hospitalized patients.  Infect Control Hosp Epidemiol 2008;29 (1) 44- 50PubMedGoogle Scholar
23.
Winston  DJLazarus  HMBeveridge  RA  et al.  Randomized, double-blind, multicenter trial comparing clinafloxacin with imipenem as empirical monotherapy for febrile granulocytopenic patients.  Clin Infect Dis 2001;32 (3) 381- 390PubMedGoogle Scholar
24.
Elixhauser  ASteiner  CHarris  DRCoffey  RM Comorbidity measures for use with administrative data.  Med Care 1998;36 (1) 8- 27PubMedGoogle Scholar
25.
Charlson  MEPompei  PAles  KLMacKenzie  CR A new method of classifying prognostic comorbidity in longitudinal studies: development and validation.  J Chronic Dis 1987;40 (5) 373- 383PubMedGoogle Scholar
26.
Quan  HSundararajan  VHalfon  P  et al.  Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data.  Med Care 2005;43 (11) 1130- 1139PubMedGoogle Scholar
27.
Moss  RRHumphries  KHGao  M  et al.  Outcome of mitral valve repair or replacement: a comparison by propensity score analysis.  Circulation 2003;108 ((suppl 1)) II90- II97PubMedGoogle Scholar
28.
Austin  PC Some methods of propensity-score matching had superior performance to others: results of an empirical investigation and Monte Carlo simulations.  Biom J 2009;51 (1) 171- 184PubMedGoogle Scholar
29.
Zhang  JYu  KF What's the relative risk? a method of correcting the odds ratio in cohort studies of common outcomes.  JAMA 1998;280 (19) 1690- 1691PubMedGoogle Scholar
30.
Leffler  DCloud  JWKelly  CP Gastric acid–suppressive agents and risk of Clostridium difficile–associated disease.  JAMA 2006;295 (22) 2599- 2600PubMedGoogle Scholar
31.
Dial  SDelaney  JACBarkun  ANSuissa  S Gastric acid–suppressive agents and risk of Clostridium difficile–associated disease [in reply].  JAMA 2006;295 (22) 2600- 2591Google Scholar
32.
Perkins  SMTu  WUnderhill  MGZhou  XHMurray  MD The use of propensity scores in pharmacoepidemiologic research.  Pharmacoepidemiol Drug Saf 2000;9 (2) 93- 101PubMedGoogle Scholar
33.
Pham  CQRegal  REBostwick  TRKnauf  KS Acid suppressive therapy use on an inpatient internal medicine service.  Ann Pharmacother 2006;40 (7-8) 1261- 1266PubMedGoogle Scholar
34.
Parente  FCucino  CGallus  S  et al.  Hospital use of acid-suppressive medications and its fall-out on prescribing in general practice: a 1-month survey.  Aliment Pharmacol Ther 2003;17 (12) 1503- 1506PubMedGoogle Scholar
35.
Nardino  RJVender  RJHerbert  PN Overuse of acid-suppressive therapy in hospitalized patients.  Am J Gastroenterol 2000;95 (11) 3118- 3122PubMedGoogle Scholar
36.
Health Care Utilization Project, Introduction to the HCUP Nationwide Inpatient Sample (NIS). 2007;http://www.hcup-us.ahrq.gov/db/nation/nis/NIS_2007_INTRODUCTION.pdf. Accessed August 15, 2009
37.
Janicki  TStewart  S Stress-ulcer prophylaxis for general medical patients: a review of the evidence.  J Hosp Med 2007;2 (2) 86- 92PubMedGoogle Scholar
38.
Cook  DJFuller  HDGuyatt  GH  et al. Canadian Critical Care Trials Group, Risk factors for gastrointestinal bleeding in critically ill patients.  N Engl J Med 1994;330 (6) 377- 381PubMedGoogle Scholar
39.
ASHP Therapeutic Guidelines on Stress Ulcer Prophylaxis, ASHP Therapeutic Guidelines on Stress Ulcer Prophylaxis: ASHP Commission on Therapeutics and approved by the ASHP Board of Directors on November 14, 1998.  Am J Health Syst Pharm 1999;56 (4) 347- 379PubMedGoogle Scholar
40.
Farrell  CPMercogliano  GKuntz  CL Overuse of stress ulcer prophylaxis in the critical care setting and beyond [published online ahead of print August 14, 2009].  J Crit Care 10.1016/j.jcre.2009.05.014PubMedGoogle Scholar
41.
Murphy  CEStevens  AMFerrentino  N  et al.  Frequency of inappropriate continuation of acid suppressive therapy after discharge in patients who began therapy in the surgical intensive care unit.  Pharmacotherapy 2008;28 (8) 968- 976PubMedGoogle Scholar
42.
Wohlt  PDHansen  LAFish  JT Inappropriate continuation of stress ulcer prophylactic therapy after discharge.  Ann Pharmacother 2007;41 (10) 1611- 1616PubMedGoogle Scholar
×