Association of Household Pets, Common Dietary Factors, and Lifestyle Factors with Clostridium difficile Infection

Abstract

Background

Clostridium difficile infection (CDI) is one of the most common hospital-acquired infections and is associated with significant morbidity and mortality. Since owning a cat or dog could enrich the gut microbiome, we hypothesized that it would be protective against CDI.

Aims

We conducted a survey study on patients tested for CDI in order to assess whether living in the presence of a pet is associated with a decreased risk of CDI.

Methods

We surveyed subjects aged 18–90 over a 14-month period using a retrospective case–control design. Subjects with CDI were matched by gender and age to patients who tested negative and had no prior history of CDI. A web-based survey was provided to subjects by mail or assisted by phone. Conditional logistic regression was used to assess for associations between CDI and the various risk factors.

Results

205 CDI positive and 205 CDI negative subjects (response rate of 50.2%) were included. After matching for age and sex, living with a cat or dog was not associated with negative CDI testing. Exploratory multivariable modeling identified an unexpected association between positive CDI testing and high meat intake (OR 2.13, 95% CI 1.21–3.77) as well as between positive CDI testing and cat allergies (OR 1.88, 95% CI 1.02–3.46).

Conclusion

Living with a cat or dog was not associated with negative CDI testing. Several novel risk factors for CDI have been identified including high meat intake and cat allergies.

Introduction

Clostridium difficile infection (CDI) is one of the most common hospital-acquired infections and is associated with significant morbidity and mortality [1, 2]. It is estimated that there are 450,000 cases of CDI in the USA each year accounting for 30,000 deaths [3]. Annual economic burden of CDI is estimated to be over 6 billion dollars in the US alone [4].

Several risk factors have been identified for CDI including advanced age, multi-morbidity, antibiotic exposure, use of acid suppression therapy, and frequent healthcare exposure [5–7]. Disturbances to gut microbiota are believed to play a central role in CDI susceptibility and pathogenesis as well [8]. Numerous risk factors have been identified in preclinical studies; however, few have been shown in humans to have meaningful clinical implications [9–11].

A recent landmark study suggested that living in close proximity with dogs and other farm animals may be protective against asthma and other allergic disease through sharing of microbes between animals and family members [12]. Recent veterinary studies have demonstrated that the presence of Clostridium difficile isolates in household pets; however, the clinical significance of this finding has yet been determined [13–15]. We recently discontinued a study of CDI in cats and dogs of patients tested for CDI because it was very difficult to recruit patients who were CDI positive and owned cats or dogs. This observation led to the hypothesis that owning a cat or dog could enrich the gut microbiome and be protective against CDI despite veterinary studies demonstrating colonization of this pathogen.

We conducted a case-control study surveying patients tested for CDI in order to assess whether living in the presence of a cat or dog is associated with a decreased risk of CDI. In addition, we assessed various other environmental and dietary risk factors to control for potential confounders.

Materials and Methods

Study Population

We retrospectively identified all subjects aged 18–90 tested for CDI at Michigan Medicine (Ann Arbor, Michigan) between September 2017 and November 2018. Subjects were identified by electronic medical record system using the institutions unique laboratory code. CDI testing results were verified by manual chart review (JAB and KR). CDI cases were identified using a multistep process: Initially, a combination of glutamate dehydrogenase (GDH) antigen enzyme immunoassay (EIA) and toxin A/B EIA (C. Diff Quik Chek Complete^®, Alere Inc., Waltham, MA) was performed. In instances of concordance (i.e., both negative or both positive), the results are reported out as negative or positive. In instances of discordance, reflex testing was performed by highly sensitive polymerase chain reaction (PCR) for the presence of toxin B gene (tcdB) using a commercial assay (Simplexa™, Focus Diagnostics, Cypress, CA).

Subjects with negative CDI testing were initially excluded if CDI testing had been positive previously or the patient carried a diagnosis of CDI in their problem list. CDI negative subjects were subsequently excluded if they reported having any prior history of CDI in the survey (question 13, Supplemental Table 1). CDI positive subjects were included if they had a positive CDI testing at our institution during the 14-month study period. Subjects with discordance between survey results and electronic health record were contacted for verification of treatment with appropriate therapy (i.e., if a subject had a positive CDI test at our institution, however, they reported that they had never had CDI in their life on the survey). CDI positive patients who received appropriate therapy were included, and survey results for question 13 were manually changed. CDI negative subjects and CDI positive subjects were matched according to age and sex. Matched CDI negative subjects had undergone CDI testing within the same week, if available, as their corresponding CDI positive pair and were selected after CDI positive subject had successfully completed the survey. We calculated that a sample size of 300 (150 per group) was needed to detect an expected proportional difference of 0.25 between two categorical values at an α of 0.05. This study was approved by the University of Michigan Institutional Review Board (HUM00133198).

Survey Instrument

Survey development was informed by previously established and hypothesis driven observed risk factors reported in the literature in order to control for potential confounders. The survey was administered through a web-based link (SurveyMonkey^®; Palo Alto, CA, USA). The survey consisted of 23 questions taking approximately 5 min on average to complete (Supplemental Table 1) and was administered within 2–4 weeks of CDI testing. The survey was distributed initially via letter followed by direct phone call to non-respondents. Letters included a link to the web-based survey, a unique study ID number, and a 5 US dollar honorarium. Due to poor response rate, the initial protocol was adjusted to include direct contact by phone call with manual answer entry into online survey by caller.

Study Variables

We abstracted baseline information on prior CDI and demographics, including age, sex, and race/ethnicity from the electronic medical health records. Prior CDI was determined by manual chart review of available notes, laboratory testing, or medical problem list (JAB). CDI negative subjects were subsequently excluded if they reported having any prior history of CDI in the survey (question 13, Supplemental Table 1).

The patients were asked questions about risk factor exposure in the 3 months immediately prior to CDI testing including questions about diet patterns, medication exposure (antibiotics, acid suppression, vitamins, and probiotics), healthcare interactions, disability level, and pet and pet allergy status. Disability level was assessed by Katz index of independence in ADL where 6 indicates complete independence and 18 indicates full assistance required in each of the six ADL categories [16]. Based on the results of the survey, post hoc electronic medical record extraction of decongestant medication use was obtained in to test decongestant medication as a potential confounder in the association between CDI status and cat allergy (Supplemental Table 2).

Statistical Analysis

We performed bivariable conditional logistic regression and multivariable conditional logistic regression to asses for associations between CDI status and collected predictors. The primary outcome was CDI status (dependent variable). The relationship between variables was evaluated visually using Pearson correlations. Collinear independent variables were not included in the final regression model. Conditional logistic regression was performed by the simultaneous entry of all predictors and sequential elimination of predictors not significantly associated with the dependent variable. Multiple models were evaluated including a model with pets, our initial predictor of interest, followed by exploratory analysis of other predictors collected. We conducted a series of analyses in which pets, prior CDI, and probiotics were tested in separate models to evaluate for potential confounding variables. Various interactions were tested in our final model. Odds ratios (OR) were used to represent the association between the predictors and CDI status. Statistical significance was considered if p < 0.05 (two-tailed). All analyses were performed using R version 3.5.0 (R Foundation for Statistical Computing, Vienna, Austria). The corresponding data and analysis can be downloaded at https://github.com/jberinstl1/final_cdi_pets.

We also calculated the E value [17], which estimates how strong unmeasured confounders would need to be to explain away the association between CDI status and probiotics.

Results

896 subjects (461 CDI negative and 435 CDI positive) received either a mailed survey or phone call requesting participation in this study. 446 subjects did not complete the survey (response rate 50.2%). Of the subjects who completed the survey, 57 completed the survey using the mailed link and 387 completed the survey via telephone assistance. Of the remaining 444 subjects, 30 presumed CDI negative subjects without a documented prior history of CDI in our electronic health records were removed due to self-reporting a prior history of CDI in the survey. Of the 414 remaining (206 CDI negative and 208 CDI positive), four were removed due to being outside of the study age range. 205 CDI negative and 205 CDI positive respondents were included in the final analysis. Respondent characteristics are detailed in Table 1. Subjects were matched according to gender and age (mean difference 1.1, range 0 to 5.2 years).

Table 1.

Respondentcharacteristics

	CDI negative (n = 205)	CDI positive (n = 205)	p value
Female, n (%)	109 (53.2%)	109 (53.2%)	1.00
Age, mean ± SD	56.55 ± 17.03	55.95 ± 16.74	0.72
Race, n (%)			0.42
White	180 (87.8%)	173 (84.4%)
Black	19 (9.3%)	20 (9.8%)
Asian	1 (0.5%)	5 (2.4%)
American Indian and Alaska Native	0 (0.0%)	1 (0.5%)
Other	1 (0.5%)	3 (1.5%)
Unknown	4 (2.0%)	3 (1.5%)
Prior CDI	0 (0.0%)	50 (24.4%)	< 0.01
Antibiotic use, n (%)	117 (57.1%)	141 (68.8%)	0.01
PPI use, n (%)	81 (39.5%)	84 (41.0%)	0.76
H2RA use, n (%)	28 (13.7%)	36 (17.6%)	0.28
Decongestant use, n (%)	107 (52.2%)	104 (50.7%)	0.77
Hospitalization, n (%)	89 (43.4%)	119 (58.0%)	< 0.01
SAR/SNF, n (%)	12 (5.9%)	33 (16.1%)	< 0.01
ADL score, Mean ± SD	6.6 ± 2.20	7.3 ± 3.20	< 0.01
Dog exposure, n (%)	95 (46.3%)	91 (44.4%)	0.69
Cat exposure, n (%)	46 (22.4%)	43 (21.0%)	0.72

n number of participants in the group, CDI Clostridium difficile infection, SD standard deviation, SAR subacute rehabilitation facility, SNF skilled nursing facility, H2RA histamine 2 receptor antagonists, PPI proton pump inhibitor, ADL activity of daily living

ADL score ranges from 6 indicating complete independence to 18 indicating complete dependence for all ADLs

Predictors of CDI Infection

Correlations between CDI status and predictors were evaluated in order to assess for collinearity and to inform our final prediction model (Supplemental Fig. 1). Several predictors were significantly associated with positive CDI testing in conditional bivariable logistic regression (Fig. 1). These included a combined recent hospitalization and recent healthcare facility exposure (Hospitalization/SAR/SNF) (OR 1.80 95% CI 1.25–2.68, p = 0.002), ADL score (OR 1.12 – 95% CI 1.03–1.22, p=0.007), recent antibiotic use (OR 1.65, 95% CI 1.10–2.48, p = 0.015), and high meat intake (OR 1.80, 95% CI 1.06–3.05, p = 0.029).

Fig. 1.

Predictors of CDI using bivariable conditional logistic regression. CDI = Clostridium difficile infection, SAR = subacute rehabilitation facility, SNF = skilled nursing facility, H2RA = histamine 2 receptor antagonists, PPI = proton pump inhibitor, ADL = activity of daily living. “Hospitalization/SAR/SNF” represent a combined variable of recent hospitalization, subacute rehabilitation facility exposure, or skilled nursing facility. ADL score ranges from scores of 6 indicating complete independence and 18 indicating complete dependence for all ADLs

A model including pet exposure as the main predictor of interest did not yield any significant association with CDI status, even after adjusting for a variety of covariables (Supplemental Table 3). As a result, pet status was not included in our final model. Exploratory conditional multivariable logistic regression modeling matched for age and sex demonstrated several statically significant predictors of CDI status (Fig. 2). In the multivariable model (Table 2), positive CDI testing was statically significantly associated with high meat intake (OR 2.13, 95% CI 1.21–3.77, p = 0.009), cat allergies (OR 1.88, 95% CI 1.02–3.46, p = 0.044), the combination of recent hospitalization and recent healthcare facility exposure (OR 1.78, 95% CI 1.19–2.65, p = 0.005), ADL score which has been scaled into three groups (OR 1.37, 95% CI 1.06–1.78, p = 0.017). The model also included high salad intake due its effects on the other predictors however high salad intake did not demonstrate statistical significance (OR 0.70, 95% CI 0.46–1.08, p = 0.103).

Fig. 2.

Predictors of CDI using conditional multiple logistic regression. CDI = Clostridium difficile infection, SAR = subacute rehabilitation facility, SNF = skilled nursing facility, ADL = activity of daily living. “Hospitalization/SAR/SNF” represent a combined variable of recent hospitalization, subacute rehabilitation facility exposure, or skilled nursing facility. ADL = activity of daily living. ADL score based on Katz index of independence in ADL which has been rescaled in this model into three groups where scores of 6 indicates complete independence, 12 indicates significant dependence, and 18 indicates complete dependence for all ADLs

Table 2.

Predictors of CDI using conditional multiple logistic regression

Predictor	Odds ratio	Lower CI	Upper CI	p value
High meat intake	2.13	1.21	3.77	0.009
Cat allergy	1.88	1.02	3.46	0.044
Hospitalization/SAR/SNF	1.78	1.19	2.65	0.005
ADL score	1.37	1.06	1.78	0.017
High salad intake	0.70	0.46	1.08	0.103

“Hospitalization/SAR/SNF” represent a combined variable of recent hospitalization, subacute rehabilitation facility exposure, or skilled nursing facility. ADL score ranges from 6 indicating complete independence to 18 indicating complete dependence for all ADLs

CDI Clostridium difficile infection, CI confidence interval, SAR subacute rehabilitation facility, SNF skilled nursing facility, ADL activity of daily living

In order to assess for potential confounding variables between CDI status and cat allergies, we performed a post hoc analysis of the impact of common nasal decongestants on CDI status and cat allergies. Correlations were seen between cat and dog allergies and these nasal decongestants (Supplemental Fig. 1); however, no association was seen between CDI status and nasal decongestants in either bivariable or multivariable models.

Probiotics, which demonstrated a statistically significant association with CDI status, were not included in our final bivariable and multivariable models, as probiotics were believed to be a marker of prior CDI infection rather than a true risk factor. This was confirmed by modeling the association of prior CDI on probiotics (OR 9.06, 95% CI 2.98–27.58, p < 0.001). The E value for the association between probiotics and CDI status was 2.42. This means that unmeasured confounders correlated with both probiotics and CDI status would need to have an odds ratio of 2.42 with respect to CDI status to explain away the observed association between probiotics and CDI status.

Discussion

After matching for age and sex and controlling for exposures to known CDI risk factors, living with a cat or dog does not appear to be associated with CDI status. This observation provides clinical correlation to veterinary studies demonstrating the presence and rare shedding of Clostridium difficile isolates in household pets [13]. While multiple studies have isolated Clostridium difficile from cats and dogs, it is unclear if this finding has any meaningful clinical implications on humans in contact with these animals. Our study, which is the first to evaluate this in a real-world clinical setting, suggests that owning pets has no meaningful clinical effects on the population studied. While this hypothesis was motivated by previously observed unreported data, the original study was performed in a smaller unmatched sample. As a result, pet ownership was not included in our final model even when pet status was adjusted for potential confounding covariables.

While no protective effect of pet ownership was found, exploratory multivariable modeling identified several unexpected but interesting associations between CDI status and common dietary and lifestyle factors. Our results demonstrate a positive association between CDI testing and high meat intake, recent hospitalization and healthcare facility exposure, high disability score, and cat allergies, identifying these common dietary and lifestyle factors as potential risk factors for CDI acquisition.

Our results confirm previously published literature that living in a nursing home, recent hospitalization, and greater disability is associated with positive CDI testing [18–20]. However, high meat intake and the presence of allergies to cats are novel risk factors not previously identified in the literature. In addition, while high salad intake did not reach statistical significance, it is likely playing a role in predicting CDI status based on its effect on the other variables in our multivariable model.

The association of meat with positive CDI test could be directly related to the presence of clostridium spores in commercial meat in both the USA and Canada [21, 22] or could be related to changes in gut microbiota composition related to meat protein intake [23, 24]. Similarly, salad could be contributing through the positive effects of short-chain fatty acids on microbiota diversity [11], which has been shown to be protective of other luminal gastrointestinal disorders such as colorectal adenoma [25].

The association of cat allergies with positive CDI testing could suggest that a T helper (Th) 2 predominant immune system, as seen in atopic/allergic-mediated diseases, could predispose subjects to CDI. It is well established that excessive Th2 cytokine production inhibits Th1 cell activation and proliferation and could blunt the host’s antibacterial defenses against CDI [26]. To assess possible unmeasured confounders in this pathway, common anti-allergic medication use was assessed. Although the exact nature and mechanism of these findings require further investigations, these finding should be hypothesis generating for future studies.

While antibiotics were associated with CDI in our bivariable model, gastric acid suppression use in the previous 30 days was not significant in either the bivariable or multivariable model. This is likely due to the fact that the previously reported association between CDI risk and acid suppression therapy was seen in larger samples sizes compared to our current study which was likely underpowered to evaluate for this weak association [27].

The association between probiotic use and positive CDI status likely represents a marker for prior CDI infection rather than a true risk factor for CDI. Similar observations were noted in a recent large retrospective study, which demonstrated an increased risk of CDI recurrence in FMT patients taking prophylactic probiotics prior to systemic antibiotic administration [28]. This was confirmed in our study by evaluating the association between prior CDI status and probiotic use. For this reason, probiotics were not included in either our bivariable or multivariable model.

The strength of our case-control survey is the relatively large sample size. We had adequate power to detect a potential effect between pets and CDI. We were able to evaluate for several predictors while accounting for potential confounders. We confirmed the validity of our approach by showing that our risk estimates are generally in line with those reported previously. As with all retrospective case-control studies, recall bias remains a pertinent concern. This was minimized by our attempts to contact subjects within 2–4 weeks of initial CDI testing. Recall bias and differential misclassification were further reduced through blinding of subjects to the risk factors, as the risk factors under investigation are not currently well established or well publicized. With a 50% non-response rate, there is a potential for non-response and selection bias among the remaining participants. In addition, there could be differences in patient characteristics between those who responded to our initial mailed survey compared to patients who were contacted by phone by a study team member with assistant survey completion. The potential difference between these two populations was not evaluated due to lack of sufficient power between these two groups. The inability to account for unmeasured and residual confounding variables remains a concern with this type of study design; however, a wide variety of confounding variables were assessed through the diversity of questions included in our survey design. In addition, one could reasonably expect the effects of pet ownership on CDI status to be confounded by age and disability level as these factors are likely to contribute to a subject’s inability to care for a pet while placing them at increased risk for CDI. However, the effect of age and disability level was accounted for in our study design. We performed pair-wise matching according to age, in order to minimize the contribution of this variable on any association between our predictors and outcome of interest. In addition, we accounted for subject disability level (and therefore their inability to care for household pets) by including disability level in our adjusted regression analysis. This study cannot make any claims about causality, as it identifies significant associations, and it should be considered hypothesis generating for future studies.

This study represents the first reported association of several common dietary and lifestyle factors with CDI in a real-world uncontrolled clinical study. The identification of additional risk factors for CDI is critical to the prevention of this common and costly infection. Several of the dietary and environmental risk factors had been shown in preclinical studies to have effects on microbiome composition with theoretical susceptibility to disease in preclinical animal models; however, to our knowledge, these risk factors have not been associated with CDI in humans. In addition, while clostridial shedding by pets poses a theoretical threat to humans, our study demonstrates no association between pets and CDI status, providing relief to pet owners. Nevertheless, this study remains exploratory in nature, and we acknowledge the limitations associated with our approach and the need to validate our findings with future studies.

Abbreviations

ADL

Activity of daily living

CDI

Clostridium difficile infection

EIA

Enzyme immunoassay

GI

Gastrointestinal

H2RA

Histamine 2 receptor antagonists

PPI

Proton pump inhibitor

SAR

Subacute rehabilitation facility

SNF

Skilled nursing facility

Th

T helper