NUR 410

American Journal of Infection Control 47 (2019) 2�8

Contents lists available at ScienceDirect

American Journal of Infection Control

journal homepage: www.aj ic journal .org

Major Article

Predictors of Clostridium difficile infection and predictive impact of probiotic use in a diverse hospital-wide cohort

D1X XMartha L. Carvour D2X XMD, PhD a,b,*, D3X XShane L. Wilder D4X XBS c, D5X XKeenan L. Ryan D6X XPharmD, PhC d, D7X XCarla Walraven D8X XPharmD, BCPS-AQ ID d, D9X XFares Qeadan D10X XPhD a, D11X XMeghan Brett D12X XMDb, D13X XKimberly Page D14X XPhD, MPH, MS a

a Division of Epidemiology, Biostatistics, and Preventive Medicine, Department of Internal Medicine, University of New Mexico, Albuquerque, NM b Division of Infectious Diseases, Department of Internal Medicine, University of New Mexico, Albuquerque, NM c University of New Mexico School of Medicine, Albuquerque, NM d Department of Pharmacy, University of New Mexico Hospital, Albuquerque, NM

Key Words:

* Address correspondence to Martha L. Carvour, MD, Medicine, University of New Mexico Health Sciences Cen ico, MSC 10-5550, Albuquerque, NM 87131.

E-mail address:[email protected] (M.L. Carv Funding/support: Supported by the Infectious Diseas

tion Medical Scholars Program and the University of New tional Sciences Center (National Institutes of Health gran

Conflicts of interest: None to report.

https://doi.org/10.1016/j.ajic.2018.07.014 0196-6553/© 2018 Association for Professionals in Infect

Background: Hospital-based predictive models for Clostridium difficile infection (CDI) may aid with surveil- lance efforts. Methods: A retrospective cohort of adult hospitalized patients who were tested for CDI between May 1, 2011, and August 31, 2016, was formed. Proposed clinical and sociodemographic predictors of CDI were evaluated using multivariable predictive logistic regression modeling. Results: In a cohort of 5,209 patients, including 1,092 CDI cases, emergency department location (adjusted odds ratio [aOR], 1.91; 95% confidence interval [CI], 1.51, 2.41; compared with an intensive care unit reference category, which had the lowest observed odds in the study) and prior exposure to a statin (aOR, 1.26, 95% CI, 1.06, 1.51), probiotic (aOR, 1.39; 95% CI, 1.08, 1.80), or high-risk antibiotic (aOR, 1.54; 95% CI, 1.29, 1.84), such as a cephalosporin, a quinolone, or clindamycin, were independent predic- tors of CDI. Probiotic use did not appear to attenuate the odds of CDI in patients exposed to high-risk antibiotics, but moderate-risk antibiotics appeared to significantly attenuate the odds of CDI in patients who received probiotics. Conclusions: Emergency department location, high-risk antibiotics, probiotics, and statins were indepen- dently predictive of CDI. Further exploration of the relationship between probiotics and CDI, especially in diverse patient populations, is warranted. © 2018 Association for Professionals in Infection Control and Epidemiology, Inc. Published by Elsevier Inc.. All

rights reserved.

Clostridium difficile Health care�associated infection Hospital epidemiology Probiotics

PhD, Department of Internal ter, 1 University of New Mex-

our). es Society of America Founda- Mexico Clinical and Transla-

t UL1TR001449).

ion Control and Epidemiology, Inc. Published by Elsevier Inc.. All rights reserved.

Clostridium difficile (recently renamed Clostridioides difficile1) is the most common cause of health care�associated infection in the United States, affecting nearly half a million patients per year and requiring an estimated $4.8 billion in direct acute care costs.2-4 Although mor- tality rates after C difficile infection (CDI) have improved,5 recurrence after treatment occurs in as many as 20% of cases.6 New antimicrobial therapies for CDI—as well as alternative methods to prevent or treat

CDI, such as prebiotic and probiotic agents and fecal microbiota trans- plantation—have been developed.7-12

CDI prevention and treatment have become high priorities in the health care system. Hospital-level CDI data are compared with national benchmarks, and in January 2015, the Centers for Medicare and Medicaid Services began to withhold funding for hospitals in the lowest quartile. Hospital-onset CDI data are publicly reported on Medicare’s Hospital Compare website.13 Meanwhile, the association of CDI and antimicrobial exposure14,15 has prompted increased sup- port of antimicrobial stewardship programs in acute care hospitals.

Recently, we reviewed our hospital’s experience with CDI over an approximately 5-year period at the University of New Mexico (UNM) Hospital, where CDI rates have been higher than expected compared with national benchmarks. We sought to identify CDI predictors that might be monitored or modified at the hospital level, with the long- term goal of reducing CDI rates at UNM.

Table 1 Antibacterial risk strata used for logistic regression modeling, with corresponding unadjusted and adjusted odds ratios for Clostridium difficile infection for each stratum

High-risk antibacterial agents Moderate-risk antibacterial agents Low-risk antibacterial agents

Cefaclor Cefadroxil Cefazolin Cefdinir Cefepime Cefixime Cefotaxime Cefoxitin Cefpodoxime Cefprozil Ceftaroline Ceftazidime Ceftriaxone Cefuroxime Cephalexin Ciprofloxacin Clindamycin Levofloxacin Moxifloxacin Norfloxacin Ofloxacin

Amoxicillin Amoxicillin-clavulanate Ampicillin Ampicillin-sulbactam Avibactam-ceftazidime Dicloxacillin Ertapenem Imipenem-cilastatin Meropenem Nafcillin Oxacillin Penicillin G benzathine Penicillin G potassium Penicillin G sodium Penicillin V potassium Piperacillin Piperacillin-tazobactam Vancomycin*

Amikacin Azithromycin Aztreonam Clarithromycin Colistimethate Dapsone Daptomycin Doxycycline Erythromycin Fosfomycin Gentamicin Linezolid Minocycline Nitrofurantoin Rifabutin Rifampin Rifapentine Rifaximin Streptomycin Sulfadiazine Sulfamethoxazole Sulfamethoxazole-trimethoprim Tetracycline Tigecycline Tobramycin Trimethoprim

Unadjusted OR = 1.46 (95% CI, 1.24, 1.72) Unadjusted OR = 1.07 (95% CI, 0.93, 1.22) Unadjusted OR = 0.96 (95% CI, 0.83, 1.11) Adjusted OR = 1.60 (95% CI, 1.33, 1.92)y Adjusted OR = 1.03 (95% CI, 0.89, 1.20)y Adjusted OR = 0.83 (95% CI, 0.71, 0.98)y

CI, confidence interval; OR, odds ratio. *Vancomycin administered by any systemic route (eg, intravenous infusion) was classified as a moderate risk antibacterial agent. Vancomycin administered by mouth, by feeding tube, or per rectum was classified as a possible Clostridium difficile treatment agent. Other treatment agents included metronidazole and fidaxomicin. Treatment agents were not included in the low-, moderate-, or high-risk strata. yAdjusted ORs are from a multivariable model containing location type, age (�65 vs <65 years), all 3 antibacterial risk strata, antifungal agents, probiotics, and statins.

M.L. Carvour et al. / American Journal of Infection Control 47 (2019) 2�8 3

METHODS

Hospital setting

The UNM Hospital is a >500-bed academic medical center in Albuquerque, New Mexico, which serves as a safety net hospital for a geographically expansive, “majority-minority” state and offers care for medically underserved populations throughout the state. It is also the only level 1 trauma center in NewMexico.

Data source

Data were retrospectively obtained from the UNM Clinical and Translational Science Center Clinical Data Warehouse, which extracts data from the UNM electronic medical records for research use. A unique study number was assigned to each patient in the cohort to permit linkage across analytic files, and original identifiers were removed before transmission of the data to the research team. The UNM institutional review board reviewed and exempted the study.

Cohort selection

All hospitalized adult patients (�18 years of age) with �1 CDI assay recorded between May 1, 2011, and August 31, 2016, were eli- gible. A new CDI assay system was implemented at the UNM Hospital in April 2011, so data collection for our study began the month after this change. CDI tests using any assay (eg, enzyme immunoassay or nucleic acid amplification/polymerase chain reaction, which were in combined use during the study period) and any diagnostic result (eg, positive or negative) were included. In keeping with our hospital’s

laboratory protocol, only specimens conforming to the shape of the container were eligible.

Outcome definition

Patients with any positive CDI assay results at any time during the study period were classified as CDI cases. If �1 positive result was recorded for the patient, the first positive result during the study period was used as the index record. Patients with �1 CDI assay with no recorded positive result during the study period were classified as not having CDI, and the first negative result in the study period was used as the index record.

Sociodemographic predictors

Sex and race/ethnicity were based on the electronic medical record. Age was defined as the patient’s age at the time of diagnosis and was analyzed as both a continuous and categorical variable (eg, �65 vs <65 years, based on findings elsewhere in the CDI literature16).

Spatiotemporal predictors

Season was defined using the month during which the CDI assay was performed (eg, December-February, March-May, June-August, and September-November). For modeling purposes, season was used instead of year of diagnosis. Seasonality may be associated with other important patterns (eg, other seasonal outbreaks and antimicrobial prescribing patterns)17 and may be carried forward to future years as a meaningful temporal unit.

Table 3 Multivariable predictive logistic regression model for Clostridium difficile infection (n = 4,278)

Unadjusted OR (95% confidence interval) Adjusted OR (95% confidence interval)*

Location type Emergency department 1.72 (1.37, 2.15) 1.91 (1.51, 2.41) General inpatient unit 1.13 (0.94, 1.36) 1.11 (0.92, 1.34) Other inpatient unit 1.19 (0.80, 1.75) 1.26 (0.84, 1.90) Intensive care unit 1.00 (reference category) 1.00 (reference category)

High-risk antibacterial agent within preceding 180 days 1.46 (1.24, 1.72) 1.54 (1.29, 1.84) Probiotics within preceding 180 days 1.55 (1.24, 1.95) 1.39 (1.08, 1.80) Statin within preceding 180 days 1.24 (1.05, 1.46) 1.26 (1.06, 1.51)

OR, odds ratio. *Adjusted ORs are from a model containing all of the variables shown in this table. The C-statistic for the adjusted model is 0.59.

Table 2 Characteristics of patients with and without CDI (n = 5,209 unless otherwise specified)

CDI (n = 1,092) No CDI (n = 4,117) P value*

Male sex 559 (51.2%) 2,105 (51.1%) .97 Race/ethnicity (n = 5,073) .61

White non-Hispanic 425 (39.9%) 1,551 (38.7%) Hispanic 379 (35.6%) 1,507 (37.6%) American Indian/Alaskan Native 189 (17.7%) 653 (16.3%) Black non-Hispanic 22 (2.1%) 85 (2.1%) Other 51 (4.8%) 211 (5.3%)

Age Mean (median, SD) in years 56.6 (57.0, 17.1) 57.5 (59.0, 17.1) .12 N (%) �65 365 (33.4%) 1,497 (36.4%) .07

Season .56 December-February 291 (26.7%) 1,066 (25.9%) March-May 273 (25.0%) 1,057 (25.7%) June-August 268 (24.5%) 1,076 (26.1%) September-November 260 (23.8%) 918 (22.3%)

Location type (n = 4,599) <.0001 General inpatient 519 (54.3%) 2,078 (57.0%) Emergency department 204 (21.3%) 538 (14.8%) Intensive care unit 195 (20.4%) 882 (24.2%) Other 38 (4.0%) 145 (4.0%)

Proton pump inhibitor within preceding 180 days (n = 4,822) 477 (45.7%) 1,699 (45.0%) .70 Immunosuppressant within preceding 180 days

Any (n = 4,851) 453 (43.2%) 1,645 (43.3%) .99 Steroid (n = 4,822) 365 (34.9%) 1,255 (33.2%) .30

Statin within preceding 180 days (n = 4,822) 236 (22.6%) 720 (19.1%) .01 Probiotic within preceding 180 days (n = 4,822) 117 (11.2%) 284 (7.5%) .0002 Antibacterial agent within preceding 180 days (n = 4,822)

High risk 817 (78.2%) 2,685 (71.1%) <.0001 Moderate risk 508 (48.6%) 1,775 (47.0%) .35 Low risk 349 (33.4%) 1,295 (34.3%) .59

Antifungal agent within preceding 180 days (n = 4,822) 126 (12.1%) 482 (12.8%) .54 Diabetes, �1 study criterion (n = 4,989) 384 (36.0%) 1,335 (34.1%) .24

CDI, Clostridium difficile infection. *All P values are from an unadjusted logistic regression model in which the modeled outcome is CDI and the variable listed in the table is the single predictor in the model. The P values shown in the table were used to determine eligibility for the multivariable selection procedure (eligible if P< .10).

4 M.L. Carvour et al. / American Journal of Infection Control 47 (2019) 2�8

Location was defined as the last recorded hospital unit prior to the CDI assay (eg, the presumed location of the patient at the time of diag- nosis), and locations were classified into 4 broad categories—emer- gency department (ED); general inpatient units, including medical and surgical wards; intensive care units; and other inpatient units, includ- ing obstetrics/gynecology, rehabilitation, and preadmission units.

Clinical predictors

Antibacterial and antifungal agents, bacterial and fungal probiot- ics, steroids and other immunosuppressants, statins,18,19 proton pump inhibitors (PPIs),20 and antidiabetes medications21 recorded within the 180-day window prior to the CDI diagnosis were included.

Antibacterial agents were classified into low, moderate, and high CDI risk strata (Table 1).14,15,22

Diabetes was defined as having either a prescription for �1 antidia- betes medications (eg, metformin or insulin) within the 180-day win- dow prior to the CDI assay or any hemoglobin A1c �6.5% during the study period, using the hemoglobin A1c value nearest to the date of the CDI assay.

Predictive modeling

Sociodemographic and clinical predictors of CDI were evaluated with predictive logistic regression modeling. Variables with P < .10 in an unadjusted model were eligible for inclusion in a multivariable

M.L. Carvour et al. / American Journal of Infection Control 47 (2019) 2�8 5

model. Variables with P < .05 in the multivariable model were retained. Manual forward and backward selection procedures were applied, and the resulting models were compared.

Although stringent P value cutoffs (as presented earlier) were applied to achieve a parsimonious model, a sensitivity analysis was performed with <.20 used for entry into the model and <.10 for retention. The classification of antibacterial agents into low-, moder- ate-, and high-risk strata was internally evaluated in our dataset by comparing the odds ratios (ORs) across strata. Analyses were con- ducted in SAS version 9.4 (SAS Institute, Cary, NC). P values <.05 were deemed statistically significant.

Power calculations

A multivariable model with �10 predictors was anticipated. Thus, a minimum of 100 cases was desired (eg, 10 CDI cases per predictor variable).23 Hospital epidemiologic surveillance data available before this study suggested that an average of 200-300 cases of CDI occurred each year. Because this annual estimate can include recurrent cases, a conservative minimum of 100 cases per year was expected. To permit stratified analyses,>5 years of data were included.

Post hoc analyses

During the planned analysis, probiotics were identified as a positive predictor of CDI. Because probiotics are a proposed pre- ventive therapy for CDI, their role as a surrogate marker of CDI risk (ie, as a clinical predictor but not necessarily a causal factor) was considered. A series of exploratory post hoc analyses was performed to better understand the context of probiotic use within the dataset, assess for potential evidence of bias, and gen- erate future hypotheses.

First, to determine whether this finding reflected a diagnostic lag—that is, whether probiotics were ordered in the days or weeks before a CDI diagnosis in the setting of concurrent or unapparent CDI—probiotic orders recorded in the 0-60 days before the assay were compared with those recorded between 61 and 120 days and between 121 and 180 days.

Next, to determine whether the apparent relationship of probiot- ics and positive CDI assays was modified by any other variable in the model, antibacterial or antifungal use, or the type of assay used to diagnose CDI, interaction terms were tested for each of these varia- bles in a multivariable model containing location type, age (�65 vs <65 years), antibacterial use, and antifungal use.

Last, to evaluate whether the observed association was driven by a particular subtype of probiotics, probiotic orders were stratified into bacterial (including Lactobacillus or Bifidobacterium species) and fun- gal (including Saccharomyces species) subtypes. Adjusted ORs were compared for these subtypes.

RESULTS

Cohort characteristics

The cohort consisted of 5,209 patients who were tested for CDI during the study period, including 1,092 cases with �1 positive CDI assay during the study period. The characteristics of patients with and without CDI are summarized in Table 2. CDI cases were more likely to be <65 years of age (P = .07); to be located in the ED at the time of the diagnosis (P < .0001); and to receive a statin (P = .01), pro- biotic (P = .0002), or high-risk antibacterial agent (P < .0001) in the 180-day window before the CDI diagnosis. All of these variables were eligible for multivariable modeling.

There were no significant differences between groups with respect to sex, race/ethnicity, seasonality, diabetes, or other

medication types (Table 2). The organization of antibacterial agents into high-, moderate-, and low-risk strata corresponded to an expected gradient in the ORs across the 3 strata (Table 1), although the low- and moderate-risk strata had similar odds and overlapping confidence intervals (CIs).

Multivariable model results

Inpatient location type, statins, probiotics, and high-risk anti- bacterial agents were significant independent predictors of CDI in the multivariable model (Table 3). Patients in the ED had the highest odds of a positive CDI assay (adjusted odds ratio [aOR], 1.91; 95% CI, 1.51, 2.41; compared with the intensive care unit reference category, which had the lowest odds). Receipt of high- risk antibacterial agents in the 180 days preceding CDI was asso- ciated with a >50% increase in the odds of a positive assay (aOR, 1.54; 95% CI, 1.29, 1.84).

The dichotomized age variable (�65 vs <65 years) was not retained in the multivariable model (P = .07). In the sensitivity analy- sis, using P < .20 for entry into the model and P < .10 for retention, the dichotomized age variable was retained (aOR, 1.15; 95% CI, 0.98, 1.34; for age <65 vs �65 years). However, this model produced simi- lar characteristics (Akaike information criterion and C-statistic) and similar beta estimates for the other variables in the model compared with the primary model. Similarly, forcing PPI or steroid use into the model shown in Table 3 revealed comparable beta estimates for all other variables.

Probiotics analysis

Probiotics recorded between 0 and 60 days and between 61 and 120 days were associated with significantly increased odds of CDI, with the highest odds observed between 61 and 120 days (Appendix Table A1). This pattern differed for another prescription-related pre- dictor (ie, statins; Appendix Table A1), which was examined for com- parison. Probiotics did not significantly interact with the type of diagnostic assay (P = .40) or the year of diagnosis (P = .20).

A significant interaction was observed between probiotics and moderate-risk antibacterial agents (P = .01; Appendix Fig A1), in which coadministration of probiotics and moderate-risk antibacterial agents in the 180 days preceding the CDI diagnosis attenuated the odds associated with probiotics alone. A similar overall pattern was observed for both low- and high-risk antibacterial agents, although these interactions were not statistically significant (P = .08 for each test for interaction; Appendix Fig. A1). This pattern was not observed with antifungal agents (P = .19; Appendix Fig. A1).

Bacterial probiotics, including Lactobacillus and Bifidobacterium species, were associated with the highest independent odds of a posi- tive CDI assay (aOR, 1.49; 95% CI, 1.11, 2.01; compared with no probi- otics and adjusted for location type, age �65 vs <65 years, antibacterial and antifungal agents, and statins). Fungal probiotics, including Saccharomyces species, were associated with a weaker increase in the odds of CDI (aOR, 1.22; 95% CI, 0.75, 1.98; compared with no probiotics and adjusted as shown earlier).

DISCUSSION

In this diverse cohort of >5,200 hospitalized patients, including >1,000 CDI cases, several factors independently predicted the occur- rence of CDI. Important contextual factors about our cohort should be noted. As a majority-minority state, New Mexico represents a unique study population with respect to race and ethnicity. Approximately one-third of the patients in our study identified as Hispanic, and >16% identified as American Indian/Alaskan Native.

6 M.L. Carvour et al. / American Journal of Infection Control 47 (2019) 2�8

A distinctive set of geographic and socioeconomic factors also influences health care in New Mexico. More than 40% of New Mexi- cans live in an area with a primary care health professional shortage, and about 20% of the state’s population lives at or below the poverty line.24 The UNM Hospital provides care for many medically under- served patients throughout the state.

Our model should be interpreted with 2 important methodologi- cal provisos in mind. First, it was constructed for the purposes of pre- dicting CDI among those tested, not for demonstrating causal relationships between any 1 variable and the outcome of CDI. Second, patients with and without CDI were all tested for CDI and therefore may have shared more clinical factors compared with others in the hospital population. Thus, our results cannot be extrapolated directly to the risk of CDI resulting from the predictors that were significant— or not significant—in our model.

As an example, we did not observe significantly increased odds of CDI among patients who received PPIs or steroids. However, the cohort consisted of hospitalized patients in whom CDI was already suspected, with high overall rates of PPI (45.1%) and steroid (33.6%) use. Although PPIs and steroids did not predict CDI in our study, this does not exclude the possibility that either PPIs or steroids could increase CDI risk.

Similarly, we did not find age �65 years to be a significant predic- tor of CDI. In fact, older patients in our cohort were less likely to have CDI. This must also be interpreted in the context of the study popula- tion—adult hospitalized patients—for which our model predicts the odds of diagnosis and not necessarily the incidence of infection. Patients �65 years old may have been diagnosed at home, at nursing facilities, or at other hospitals where Medicare-eligible patients may be seen.

Variables that independently predicted CDI in our study (Table 3) were inpatient location type and use of high-risk antibacterial agents, statins, and probiotics. Patients were most likely to have a recorded location in the ED at or immediately preceding their CDI diagnosis. This finding could represent a number of underlying factors, such as a high frequency of ED visits, perhaps comprising a primary health care access point for many patients in the com- munity; high frequencies of antibiotic prescribing or CDI testing in the ED; or potential delays in admission to other inpatient units owing to precautionary isolation practices. Similarly, statin use may constitute a measure of increased health care access in our study. The relationship of statin use and CDI remains a sub- ject of interest in the literature,18,19 although the nature and direction of this relationship is not yet clear.

The positive association of probiotic use with subsequent CDI was unexpected. Prior studies have suggested that probiotics may prevent CDI, although results have varied depending on the type, timing, and setting of CDI, as well as the type of probiotic.8-12,25- 27 To date, there is neither scientific consensus nor Food and Drug Administration approval for the uniform use of probiotics to prevent CDI.

As described above, this result should be interpreted with care. Probiotics independently predicted the odds of CDI in our cohort, but this does not demonstrate that probiotics caused or contributed to the causes of CDI. Even so, if probiotics had exerted a strong protec- tive effect in the cohort, we might have expected probiotics to impose a negative (or perhaps a null) predictive impact. Recognizing that the direction of the association in our study was unexpected, we under- took a series of post hoc analyses to better understand the context of this result.

First, we anticipated that diagnostic lags between the onset of CDI symptoms (at which time probiotics might have been ordered) and CDI diagnoses may have created an inaccurate impression that probiotic use actually preceded the infection. As shown in the Appendix (Table A1), however, the odds of CDI

after a probiotic prescription remain elevated for up to 4 months after recorded probiotic use; in fact, the highest odds were observed for probiotics recorded 2-4 months before the CDI diag- nosis. Thus, it is not likely that short-term diagnostic lags can fully explain our observation.

Similarly, if the observed impact of probiotics differed significantly over time—that is, if this was concentrated early in the study period— we might have concluded that probiotics were markers of existing CDI and that most of the cases driving the association were preexist- ing or recurrent infections. However, the relationship of probiotics and CDI diagnosis did not change significantly over time, as evi- denced by the absence of a significant statistical interaction between probiotics and the year of diagnosis.

Next, we considered the possibility that patients were more likely to be treated with probiotics during periods of increased exposure to the health care environment. The observed temporal patterns in the Appendix Table A1 do not support this. The highest odds of probiotic use occurred in a time window distinct from that in which the CDI diagnosis was made, and this pattern differed from the association of CDI and statin prescriptions—another possible surrogate for health care exposures.

Finally, we anticipated that probiotics might be a surrogate marker for another correlated variable or set of variables. To obtain preliminary, hypothesis-generating information, we assessed whether the predictive impact of probiotics differed according to other clinical factors, including antimicrobial therapies. This analysis revealed several further, unexpected findings (Appendix Fig. A1).

Coadministration of probiotics with high-risk antibacterial agents in the 180 days preceding the CDI diagnosis did not significantly attenuate the odds of CDI associated with high-risk antibacterial ther- apies (Appendix Fig A1C). Instead, coadministration of moderate-risk antibacterial agents with probiotics in the 180 days preceding the CDI diagnosis actually appeared to attenuate the odds of CDI associ- ated with probiotics (Appendix Fig A1B). This general pattern was observed for all antibacterial strata but not for antifungal agents (Appendix Fig A1). Meanwhile, bacterial probiotics were also stronger predictors of CDI than fungal probiotics.

Prior evidence suggests that race, ethnicity, and socioeconomic position may all impact CDI risk28 and microbiomic composition at various anatomic sites.29,30 If so, specific probiotic therapies for CDI may only be useful insofar as we understand the underlying microbiomic environments across which these are applied. Fur- ther attention may need to be directed to understanding the CDI epidemic and its microbiomic drivers in diverse and medically underserved populations to distinguish between the causes of CDI on a population level—a worldwide problem and one clearly still observed in our hospital—and not just the causes of individual cases.31,32

In this diverse cohort, patients with CDI were most commonly diagnosed while in the ED and were likely to have prior exposures to high-risk antibacterial agents, probiotics, and statins. Our study is limited by the retrospective and observational nature of data collec- tion. Specific information about clinical impressions, adherence with prescribed therapies, and exposure to therapies other than those recorded in the electronic medical record was not available for this study. Future prospective CDI research should consider potential dif- ferences in microbiomic composition, CDI prevention, and CDI treat- ment in diverse and medically underserved populations.

References

1. CLSI. Clinical and Laboratory Standards Institute supplement M100, 28th edition. Available from: http://em100.edaptivedocs.net/Login.aspx. Accessed July 16, 2018.

2. Magill SS, Edwards JR, Bamberg W, Beldavs ZG, Dumyati G, Kainer MA, et al. Multi- state point-prevalence survey of health care-associated infections. N Engl J Med 2014;370:1198-1208.

M.L. Carvour et al. / American Journal of Infection Control 47 (2019) 2�8 7

3. Lessa FC, Mu Y, Bamberg WM, Beldavs ZG, Dumyati GK, Dunn JR, et al. Burden of Clostridium difficile infection in the United States. N Engl J Med 2015;372:825-34.

4. Dubberke ER, Olsen MA. Burden of Clostridium difficile on the healthcare system. Clin Infect Dis 2012;55(Suppl 2):88-92.

5. Shrestha MP, Bime C, Taleban S. Decreasing Clostridium difficile-associated fatality rates among hospitalized patients in the United States: 2004-2014. Am J Med 2018;131:90-6.

6. Eyre DW, Walker AS, Wyllie D, Dingle KE, Griffiths D, Finney J, et al. Predictors of first recurrence of Clostridium difficile infection: implications for initial manage- ment. Clin Infect Dis 2012;55(Suppl 2):77-87.

7. Galp�erine T, Guery B. Exploring ways to improve CDI outcomes. M�edecine Mal Infect 2018;48:10-17.

8. Allen SJ. The potential of probiotics to prevent Clostridium difficile infection. Infect Dis Clin North Am 2015;29:135-44.

9. Hell M, Bernhofer C, Stalzer P, Kern JM, Claassen E. Probiotics in Clostridium difficile infection: reviewing the need for a multistrain probiotic. Benef Microbes 2013;4:39-51.

10. Goldenberg JZ, Ma SS, Saxton JD, Martzen MR, Vandvik PO, Thorlund K, et al. Probi- otics for the prevention of Clostridium difficile-associated diarrhea in adults and children. Cochrane Database Syst Rev 2013(5):CD006095.

11. Evans CT, Johnson S. Prevention of Clostridium difficile infection with probiotics. Clin Infect Dis 2015;60(Suppl 2):122-8.

12. Pattani R, Palda VA, Hwang SW, Shah PS. Probiotics for the prevention of antibi- otic-associated diarrhea and Clostridium difficile infection among hospitalized patients: systematic review and meta-analysis. Open Med 2013;7:e56-67.

13. Medicare.gov. Hospital Compare. Available from: https://www.medicare.gov/hos- pitalcompare/search.html?. Accessed July 16, 2018.

14. Slimings C, Riley TV. Antibiotics and hospital-acquired Clostridium difficile infec- tion: update of systematic review and meta-analysis. J Antimicrob Chemother 2014;69:881-91.

15. Crew PE, Rhodes NJ, O’Donnell JN, Miglis C, Gilbert EM, Zembower TR, et al. Corre- lation between hospital-level antibiotic consumption and incident health care facility-onset Clostridium difficile infection. Am J Infect Control 2018;46:270-5.

16. van Werkhoven CH, van der Tempel J, Jajou R, Thijsen SF, Diepersloot RJ, Bonten MJ, et al. Identification of patients at high risk for Clostridium difficile infection: development and validation of a risk prediction model in hospitalized patients treated with antibiotics. Clin Microbiol Infect 2015;21, 786.e1-8.

17. Brown KA, Daneman N, Arora P, Moineddin R, Fisman DN. The co-seasonality of pneumonia and influenza with Clostridium difficile infection in the United States, 1993-2008. Am J Epidemiol 2013;178:118-25.

18. Motzkus-Feagans CA, Pakyz A, Polk R, Gambassi G, Lapane KL. Statin use and the risk of Clostridium difficile in academic medical centres. Gut 2012;61:1538-42.

19. Nseir W, Bishara J, Mograbi J, Mahamid M, Khalaila W, Taha M, et al. Do statins protect against the development of Clostridium difficile-associated diarrhoea? J Antimicrob Chemother 2013;68:1889-93.

20. Kwok CS, Arthur AK, Anibueze CI, Singh S, Cavallazzi R, Loke YK. Risk of Clostrid- ium difficile infection with acid suppressing drugs and antibiotics: meta-analysis. Am J Gastroenterol 2012;107:1011-19.

21. Piper MS, Saad RJ. Diabetes mellitus and the colon. Curr Treat Options Gastroen- terol 2017;15:460-74.

22. Brown KA, Khanafer N, Daneman N, Fisman DN. Meta-analysis of antibiotics and the risk of community-associated Clostridium difficile infection. Antimicrob Agents Chemother 2013;57:2326-32.

23. Peduzzi P, Concato J, Kemper E, Holford TR, Feinstein AR. A simulation study of the number of events per variable in logistic regression analysis. J Clin Epidemiol 1996;49:1373-9.

24. University of New Mexico Health Sciences Center. 2017 NM health data summary. Available from: https://hsc.unm.edu/research/ctsc/assets/doc/CERC/nm-health- data-summary.pdf. Accessed July 16, 2018.

25. Goldstein EJC, Citron DM, Claros MC, Tyrrell KL. Bacterial counts from five over-the-counter probiotics: are you getting what you paid for? Anaerobe 2014;25:1-4.

26. Maziade P-J, Andriessen JA, Pereira P, Currie B, Goldstein EJC. Impact of adding prophylactic probiotics to a bundle of standard preventative measures for Clostrid- ium difficile infections: enhanced and sustained decrease in the incidence and severity of infection at a community hospital. Curr Med Res Opin 2013;29:1341-7.

27. Chopra T, Goldstein EJC. Clostridium difficile infection in long-term care facilities: a call to action for antimicrobial stewardship. Clin Infect Dis 2015;60(Suppl 2):72-6.

28. Mao EJ, Kelly CR, Machan JT. Racial differences in Clostridium difficile infection rates are attributable to disparities in health care access. Antimicrob Agents Che- mother 2015;59:6283-7.

29. Mello CS, Carmo-Rodrigues MS, Filho HBA, Melli LC, Tahan S, Pignatari AC, et al. Gut microbiota differences in children from distinct socioeconomic levels living in the same urban area in Brazil. J Pediatr Gastroenterol Nutr 2016;63:460-5.

30. Ravel J, Gajer P, Abdo Z, Schneider GM, Koenig SS, McCulle SL, et al. Vaginal micro- biome of reproductive-age women. Proc Natl Acad Sci U S A 2011;108(Suppl 1):4680-7.

31. Rose G. Sick individuals and sick populations. Int J Epidemiol 1985;14:32-8. 32. Schwartz S, Carpenter KM. The right answer for the wrong question: consequences

of type III error for public health research. Am J Public Health 1999;89:1175-80.

APPENDIX

This document contains 2 supplemental exhibits from the post hoc analysis of probiotic use and Clostridium difficile infection as described in the main article text.

8 M.L. Carvour et al. / American Journal of Infection Control 47 (2019) 2�8

Fig. A1. Association of probiotics and Clostridium difficile infection stratified by coexposure to low-risk antibacterial agents (A), moderate-risk antibacterial agents (B), high-risk antibacterial agents (C), or antifungal agents (D) (n = 4,278). A statistically significant interaction was observed between probiotics and moderate-risk antibacterial agents (P = .01). Adjusted ORs and 95% CIs are shown for patients who received probiotics with or without an antimicrobial agent in the same 180-day period. Each panel (A-D) repre- sents a separate model. ORs are adjusted for location type, age (�65 vs <65 years), statins, and all other antibacterial and antifungal strata not already included in the interaction term. For instance, the moderate-risk panel (B) shows the ORs for a hybrid variable combining moderate-risk antibacterial agents with or without probiotics, and the ORs in that figure are adjusted for low- and high-risk antibacterial agents, antifungal agents, statins, age, and location. CI, confidence interval; OR, odds ratio.

Table A1 Comparative odds of Clostridium difficile infection for patients exposed to probiotics or statins in different time windows within the 180-day period before diagnosis

Time window Probiotics OR (95% confidence interval)*

Statins OR (95% confidence interval)*

0-60 days (n = 4,209) 1.32 (0.99, 1.77) 1.24 (1.03, 1.51) 61-120 days (n = 1,199) 1.84 (1.02, 3.32) 0.84 (0.58, 1.20) 121-180 days (n = 948) 0.95 (0.41, 2.18) 1.31 (0.90, 1.91)

OR, odds ratio. *ORs correspond to a multivariable model containing probiotics, statins, location type, age (�65 vs<65 years), all 3 antibacterial risk strata, and antifungal agents. ORs represent the odds of C difficile infection for patients exposed to the medication vs those not exposed (eg, probiotic vs no probiotic, statin vs no statin).