Non-Steroidal Anti-Inflammatory Drugs Are Not Associated with Increased Risk of Clostridioides difficile Infection: A Propensity-Score-Matched Case-Control Study

Abstract

Prior research identified an increased risk for Clostridioides difficile infection (CDI) following exposure to certain non-steroidal anti-inflammatory drugs (NSAIDs). We conducted a retrospective case-control study to evaluate the risk for CDI associated with NSAID use. NSAID use was not associated with an increased risk of CDI.

Many risk factors for acquisition of Clostridioides difficile infection (CDI) are known, including advanced age, healthcare exposure, comorbid medical conditions, and, most notably, exposure to antibiotics. Exposure to other medications has also been demonstrated to increase risk, such as proton pump inhibitors¹.

Non-steroidal anti-inflammatory drugs (NSAIDs) are among the most commonly utilized medications. They are frequently taken over-the-counter and generally considered safe, but they are associated with significant gastrointestinal side effects ². Prior studies have suggested an association between NSAID use and risk of CDI ^1,3, possibly modulating CDI risk by altering the composition of the gut microbiome ². One hypothesis for these changes is that they result from the modification of inflammatory markers and subsequent reduction in gut blood flow ². However, prior studies had two major limitations: a) inadequate assessment of over-the-counter NSAID use ³; and b) failure to account for treatment assignment bias ³. Although these studies adjusted for confounding via matched controls, there is a potential for misattribution of risk, which could be related to the comorbid condition leading to NSAID use rather than to the NSAID use itself.

We report here the results of a retrospective case control study designed to evaluate NSAID exposure as a risk factor for C. difficile infection, with inclusion of over-the-counter use of NSAIDs and matching of cases to controls on the basis of their propensity for NSAID exposure.

The study was approved by the University of Michigan Institutional Review Board. This was a retrospective case-control study with subjects drawn from a cohort previously published by Menon et al. ⁴, which includes subjects consecutively enrolled after being tested for CDI by the clinical microbiology laboratory at the University of Michigan in 2016. The cohort was previously used to evaluate for correlation between stool toxin levels and disease severity⁴.Those with diarrhea and positive CDI testing either by enzyme immunoassay (EIA) for toxins A/B or by polymerase chain reaction (PCR) for the tcdB gene were identified as cases, while those whose testing was negative were identified as controls. Repeat positives occurring within 2 weeks were excluded as this was considered to be the same CDI episode.

NSAID use data were abstracted from the electronic medical record (EMR) using EMERSE™ 5, an EMR search engine custom-built at the University of Michigan, which permitted the text of clinical notes and other documentation to be queried within 30 days of C. difficile testing. Keywords included a comprehensive list of terms for NSAIDs, including trade names. Results from the query were manually reviewed by one investigator (AR) to confirm NSAID exposure within 30 days preceding C. difficile testing. Additional data regarding comorbid conditions and laboratory values were queried electronically from the EMR. A random forest model implemented via the R package missForest version 1.4 ⁶ was used to impute the missing data; to avoid bias, outcomes were excluded from the imputation matrix. Additional details are in the supplement.

The primary analysis focused on non-aspirin NSAID use (hereafter referred to as NSAID use). A logistic regression model was used to create a propensity score for NSAID use selecting variables on the basis of biologic or clinical plausibility for a relationship with NSAID use. Various models were fit and ultimately the collection of variables which provided the best fit to NSAID use by area under the receiver operator characteristic curve were included in the final propensity score: gender, back pain, baseline serum creatinine, osteoarthritis, rheumatoid arthritis, serum albumin, and use of anticoagulant or antiplatelet medications. Cases were matched 1:1 with controls using a matching caliper equal to 0.2 times the standard deviation of the logit of the propensity score ⁷.

Conditional logistic regression via the R package survival version 3.2–7 ⁸ was used to assess for bi-variable relationships between predictors such as NSAID use and the outcome of CDI. Subsequently, using purposeful selection ⁹, a multivariable conditional logit model was built to test whether NSAIDs increase the risk of CDI, while adjusting for comorbid disease burden, age and prior history of CDI. All analyses were conducted in R version 4.0.3 (R Foundation for Statistical Computing, Vienna, Austria).

The cohort included 1378 individuals of which 634 were CDI cases. Data from all patients within the cohort were used to build the propensity score. There were 6 cases that were unable to be matched to controls within our pre-specified caliper, and there were 116 unmatched controls, resulting in a total of 628 matched pairs included in the final analysis.

NSAID use was similar between the two groups on unadjusted analysis (P =.155). The adjusted, multivariable model demonstrated that NSAID use is not a significant risk factor for CDI (P =.816) after adjustment for comorbid disease burden, age and history of prior CDI (Table 1). Prior history of CDI was found to be an independent risk factor for current CDI regardless of NSAID use (Supplemental Figure 1). Unexpectedly, burden of comorbid disease demonstrated a protective effect in our model. However, box plots of these data show considerable overlap in weighted Elixhauser score across the four groups (Supplemental Figure 3).

Table 1:

Study Population and Modeling Results

			Unadjusted	Multivariable Model
Variable, n (%) or mean ±SD	Cases (N=628)	Controls (N=628)	P-value	OR [95% CI]	P
Non-aspirin NSAID a	140 (22.3)	161 (25.6)	.15	0.97 [0.72,1.29]	.816
Age (years)	58 ± 18.2	56.2 ± 17.4	.05	1.09 [1.01, 1.17]b	.02
Weighted Elixhauser score	11.6 ± 10.8	13.99 ± 11.8	<.001	0.98 [0.97,0.99]	<.001
Prior CDI	185 (29.5)	82 (13.1)	<.001	2.64 [1.96, 3.56]	<.001
White race	542 (86.3)	522 (83.1)	.13
BMI* (kg/m2)	27.4 ± 6.8	28.4 ± 8.6	.02
Antibiotic Usec	246 (57.7)	345 (71.9)	.62
Proton pump inhibitor use	234 (37.3)	282 (44.9)	.006
Baseline hemoglobin (g/dl)	10.3 ± 2.3	10 ± 2.4	.02
Serum 25-OH vitamin D (ng/ml)	25.9 ± 6.9	25.1 ± 6.5	.05
Inflammatory bowel disease	108 (17.2)	96 (15.3)	.36
# Hospitalizations in year prior to CDI	0.13 ± 0.4	0.16 ± 0.4	.19
# Positive prior CDI episodes	0.6 ± 1.3	0.3 ±1	<.001

^aIn prior 30 days

^bper 10-year increase

^cActive Antibiotic Use (percent based on 480 controls and 426 cases due to missingness)

Abbreviations: BMI, body mass index; CDI, C. difficile infection; NSAID, non-steroidal anti-inflammatory drug

A sensitivity analysis was conducted replacing weighted Elixhauser score with a modified Elixhauser that excluded renal disease, a known risk factor for CDI, and NSAID use was still not associated with CDI (P =.9; Table 2). A second sensitivity analysis was conducted comparing scheduled, intermittent, and no NSAID use, and neither scheduled (P =.345) nor intermittent (P =.708) NSAID use was associated with CDI (Supplementary Table 1). Secondary analyses of aspirin use showed no associations with CDI when modeled in place of non-aspirin NSAID use (P = .21), alongside it as a covariate (P = .21), or as an interacting term (P = .59; Table 2). NSAID use remained without association with CDI when antibiotic use (P = .93) or PPI use (P = .76; Table 2) were added to the model.

Table 2:

Sensitivity & Secondary Analysis

	Multivariable Model (Original)		Modified Weighted Elixhauserc		Aspirin Substitutione		Aspirin Covariatef		Aspirin Interactiong		Antibiotic Usei		PPI Usek
	OR [95% CI]	P	OR [95% CI]	P	OR [95% CI]	P	OR [95% CI]	P	OR [95% CI] or ß-estimateh ±standard error	P	OR [95% CI]	P	OR [95% CI]	P
Non-Aspirin NSAID a	0.97 [0.72,1.29]	.816	0.98 [0.74,1.31]	.90	-	-	0.97 [0.73,1.29]	.83	0.07 ±0.17	.66	1.02 [0.68, 1.52]	.93	0.96 [0.72, 1.27]	.76
Aspirin	-	-	-	-	0.84 [0.64,1.10]	.21	0.84 [0.64,1.10]	.21	−0.08 ±0.15	.59	-	-	-	-
Non-Aspirin NSAID: Aspirin Interaction	-	-	-	-	-	-	-	-	−0.41 ±0.31	.19	-	-	-	-
Age b	1.09 [1.01,1.17]	.02	1.08 [1.01,1.16]	.03	1.10 [1.02,1.18]	.009	1.10 [1.02,1.18]	.01	1.10 [1.02,1.18]	.01	1.23 [1.11, 1.37]	<.001	1.09 [1.01, 1.16]	.02
Weighted Elixhauser Score	0.98 [0.97,0.99]	<.001	-	-	0.98 [0.97,0.99]	<.001	0.98 [0.97,0.99]	<.001	0.98 [0.97,0.99]	<.001	0.98 [0.96, 0.99]	.002	0.98 [0.97, 0.99]	<.001
Modifiedd Weighted Elixhauser Score	-	-	0.98 [ 0.96,0.99]	<.001	-	-	-	-	-	-	-	-	-	-
Prior CDI	2.64 [1.96,3.56]	<.001	2.52 [1.86,3.40]	<.001	2.66 [1.97,3.58]	<.001	2.65 [1.96,3.58]	<.001	2.66 [1.97,3.59]	<.001	2.16 [1.38,3.37]	<.001	2.63 [1.95, 3.55]	<.001
Antibiotic Use	-	-	-	-	-	-	-	-	-	-	0.61 [0.43, 0.89]	.009	-	-
PPI Use	-	-	-	-	-	-	-	-	-	-	-	-	0.84 [0.65, 1.08]	.18

^aIn prior 30 days

^bper 10-year increase

^cModified weighted Elixhauser score substituted for weighted Elixhauser score in the model

^drenal disease removed from weighted Elixhauser score

^eAspirin use substituted for Non-Aspirin NSAID use in the model

^fAspirin use added to the model as a covariate

^gAspirin added as an interacting term in the model

^hCoefficients and standard errors provided for terms constituting the interaction, since odds ratios for each depend on the status of the other

ⁱAntibiotic use added to the model as a covariate

^kPPI use added to the model as a covariate

Finally, secondary analysis stratified by CDI diagnostic method (EIA or PCR) showed no significant changes compared to the primary analysis among those who were EIA positive. For the PCR positive cases, the age at diagnosis no longer demonstrated a significant association with risk for CDI (Supplementary Table 2).

In this retrospective case-control study, NSAID use was not associated with an increased risk of Clostridioides difficile infection. To our knowledge this is the first study of NSAID use as a risk factor for CDI to account for treatment assignment bias utilizing propensity score matching. This is significant as it increases our confidence that our modeled risk of CDI is causally related to NSAID use itself, rather than the underlying indication for the NSAID use. Prior observational studies have suggested an association between NSAID use and increased risk of CDI ³, but did not account for possible treatment assignment bias. Both on unadjusted and adjusted modeling, our findings do not support an association between NSAID use and an increased risk for CDI. Notable strengths of this study include accounting for over-the-counter NSAID use through text based query, which increased our sensitivity in detecting NSAID use, and the multiple sensitivity and secondary analyses we conducted, which all support the primary conclusion above.

There are some limitations to recognize. Our control subjects had diarrhea suspicious for CDI and tested negative. Thus, this control group may not be representative of the general population of patients who use NSAIDs. The ideal comparator group would be asymptomatic individuals with negative testing. However, such asymptomatic control subjects’ C. difficile colonization status is not able to be determined retrospectively, whereas we can be sure our control patients were not colonized. Since knowing the CDI status was central to our study, we chose a control group where we could be certain of that status. They also represent the same referral network and come from the same source population as the cases. As these were individuals who were tested for CDI, they also likely had a similar pre-test probability of CDI as our cases; therefore, it is unlikely we were comparing a group with CDI to a group with low risk for CDI. With the retrospective data collection, misclassification is possible. A patient with as-needed use of NSAIDs may have been incorrectly identified, as an infrequently used as-needed medication may not have been administered within 30 days of C. difficile testing. Alternatively, over-the-counter use may not have been recorded. Furthermore, our study does not address any potential dose-response effect as data regarding specific doses and durations were not reliably available. To address this limitation, we conducted a sensitivity analysis comparing scheduled, PRN, and no NSAID use, and saw no evidence of any such dose-response.

Finally, this study does not provide information regarding outcomes of CDI and any association with NSAID use. Prior studies have demonstrated increased severity of infection and worse outcomes ¹⁰ from CDI in mice after exposure to NSAIDs, although a recent retrospective observational study in humans showed no difference in outcomes ¹¹. Misoprostol, a drug which protects against the effects of NSAIDs on the GI tract, has also been demonstrated to decrease severity of CDI in mice ¹², and these murine studies strongly suggest a role for NSAIDs in mediating the severity and outcome of CDI. Future research should focus on the severity of illness and outcomes of patients with CDI who receive NSAID medications.

Highlights:

• Retrospective case control study of NSAIDs and risk for Clostridioides difficile

• Propensity score matching used to account for treatment assignment bias

• Included assessment of over the counter NSAID medication use with text based query

• No association was found between NSAIDs and Clostridioides difficile infection

• Multiple sensitivity and secondary analyses support the primary conclusion