Risk factors for recurrent Clostridium difficile infection (CDI) hospitalization among hospitalized patients with an initial CDI episode: a retrospective cohort study

Over the past decade Clostridium difficile infection (CDI) has increased in both frequency and severity in the US and abroad. A study from Quebec identified a 5-fold rise in the incidence of hospitalizations with CDI over 13 years, accompanied by a doubling in the risk of complicated disease [1]. Similarly, multiple US-based studies have reported a more-than-doubling of hospitalizations with a CDI diagnosis between 2000 and 2005 [2, 3]. These numbers have continued to rise through 2009, albeit less rapidly [4]. Much of this growth is thought to be due to the recent emergence of the hypervirulent epidemic strain of C. difficile, BI/NAP1/027. A fluoroquinolone-resistant toxin overproducer, this strain has now been detected in most of the states in the US, in North America, Europe and beyond [5].

One of the most challenging aspects of CDI is its propensity to recur. Both metronidazole and vancomycin, first-line therapies recommended in the joint evidence-based practice guideline from the Society of Healthcare Epidemiology of America (SHEA) and Infectious Diseases Society of America (IDSA), have exhibited unacceptably high rates of recurrence [6]. Indeed, a recent meta-analysis has found that CDI recurs in 13% – 50% of all patients after an initial episode, and in the setting of a randomized controlled trial, the recurrence rate was 25% [7–9].

Recurrent CDI (rCDI) is a cause of much morbidity, and its economic impact is likely substantial. Several studies have identified important risk factors for rCDI, including advanced age, chronic renal insufficiency, elevated white blood cell count, low serum albumin, use of proton pump inhibitors (PPI), and continued use of systemic antimicrobials during the initial CDI episode (iCDI) [7, 10–13]. However, a meta-analysis identified major gaps in our understanding of the risk factors for CDI recurrence [7]. Although the authors found concomitant antimicrobials, gastric acid suppressants and older age to be strongly predictive of rCDI, other factors, including iCDI treatment and specific non-CDI antimicrobials, could not be evaluated adequately due to the lack of robust data. Additionally, most studies have focused on the factors immediately preceding rCDI onset, ignoring the possibility that factors present at or near the onset of the iCDI episode may also impact this risk. In fact, recent data suggest that the burden of community-onset healthcare-facility associated (CO-HCFA) CDI is much higher than previously appreciated, and poses an additional risk pool for inpatient exposure [14, 15]. Since CO-HCFA implies an ongoing exposure to the healthcare system, it may itself be a marker for a recurrence.

A precise understanding of who is likely to recur is an important clinical question for two reasons. First, if there are modifiable exposures that increase this risk, knowing what they are may aid clinicians in avoiding them. Second, if patient characteristics not subject to modification predispose to rCDI, recognizing them may help target preventive measures more effectively. In order to define more fully the risk factors for rCDI, we conducted a single center retrospective cohort analysis among patients hospitalized with an iCDI episode.

This study was approved by the Washington University Institutional Review Board, and its conduct was in compliance with the Helsinki Declaration.

We have identified eight discrete independent risk factors for recurrent CDI. Although some characteristics, such as age, cannot be altered, several of them constitute modifiable exposures. New gastric acid suppression and concomitant antimicrobial exposures were associated with increased hazards of developing recurrent CDI. Reducing these exposures could potentially decrease the risk of recurrent CDI. This may serve as yet another reason for institutions to engage in aggressive antimicrobial stewardship programs.

Prior investigations have reported advanced age, chronic renal insufficiency, elevated white blood cell count, low serum albumin, use of PPI and H2RB, as well as continued use of systemic antimicrobials to be important risk factors for rCDI [7, 10–13, 22]. Our results are in general agreement with these prior data. Gastric acid suppressors have garnered a particular interest with respect to their impact on iCDI and rCDI incidence. Specific to recurrent disease, a recent meta-analysis substantiated this concern, finding a more-than doubling of the risk of rCDI in the setting of these drugs [7]. At the same time, it is unclear whether both PPIs and H2RBs are associated with the risk of rCDI, or whether one is a more likely culprit than the other. For example a meta-analysis by Kwok and colleagues implicated PPIs but not H2RBs in a 2-fold rise of rCDI incidence [23]. Similarly, Tleyjeh et al. in a meta-analysis of 33 studies focusing specifically on H2RB exposure reported a smaller, albeit still significant, association between receiving H2RBs and development of CDI [24]. Both meta-analyses suggested that gastric protection in conjunction with antibiotic administration carries a higher risk of CDI development than exposure to PPIs or H2RBs alone [23, 24]. In our study, we examined gastric acid suppressors as a single category because our prior work, including preliminary analyses for this study (data not shown), has consistently found no difference in the associations between these two classes of medications and CDI [17, 18]. Whether gastric acid suppression is truly an independent risk factor for CDI or a marker for patients at risk for CDI remains unknown [6].

A large body of evidence also ties concomitant use of non-CDI antimicrobials to an increased risk of a recurrence [7, 17, 18, 25]. We found that high-risk antimicrobials raise the risk for rCDI, particularly when administered after the completion of iCDI treatment. We have also confirmed previous findings that link exposure to such specific antimicrobials as IV vancomycin and fluoroquinolones to the risk for CDI incidence [17, 18, 26]. The BI/NAP1/027 strain has been associated with fluoroquinolone exposures, and may be more likely to cause rCDI than other strains of C. difficile [27]. Consequently, it is possible that fluoroquinolone exposure is a marker for CDI specifically due to this strain. For IV vancomycin, however, this association may represent not a causal relationship, but rather a marker for higher illness severity and, thus, confounding by indication.

We were also able to demonstrate the importance of timing of antimicrobial exposure after the end of CDI treatment. When modeled as time dependent variables, high-risk antimicrobials, fluoroquinolones, and IV vancomycin were all associated with rCDI. When modeled as time independent variables, the hazards of rCDI associated with high-risk antimicrobials dropped from 2.95 (2.25-3.86) to 1.86 (1.42-2.42), and fluoroquinolones and IV vancomycin were no longer associated with rCDI. Intuitively, this makes sense. An antimicrobial should not increase the risk of rCDI after CDI treatment has ended until the patient is exposed to the antimicrobial. Not modeling antimicrobials as time dependent variables after CDI treatment has ended dilutes the association with rCDI, since the days not on these drugs are included in the model.

A direct relationship between CO-HCFA status and iCDI and rCDI development is a newer finding [15]. Namely, the CO-HCFA designation of the iCDI episode is associated with at least a 25% and as much as a 2-fold increase in the risk of rCDI. A likely mechanism relates to the fact that CO-HCFA defines a population of patients who is likely sicker as evident by recent hospitalizations, and more likely to be exposed to antimicrobials. However, CO-HCFA CDI remained an independent risk factor when controlling for recent hospitalizations.

It is worth noting that the recurrence rate we observed in the current study is at the lower end of what has been reported previously. For example, a recent meta-analysis by Garey et al. examined the literature on risk factors for rCDI [7]. In the 12 studies meeting the inclusion criteria, the rates of recurrence ranged from 13% to 50%. More current data from randomized controlled trials suggest that CDI is likely to recur in approximately 25% of the patients treated for iCDI with vancomycin [8, 9]. A potential explanation for the lower rCDI rate in our study compared to others is how cases of CDI were identified. Most stools submitted for C. difficile testing at the BJH microbiology laboratory come from inpatients and the emergency room. A minority of specimens come from outpatients or affiliated skilled nursing facilities. It is likely that milder cases of rCDI were missed because the patient did not require care in an emergency room or need to be admitted. Therefore, the rCDI in this study may consist of more clinically important episodes, occurring in sicker patients, many of whom required an admission or evaluation in the emergency department.

It is possible that patients who resided outside the St. Louis metropolitan area would not be likely to return to BJH for testing for a recurrence. To examine the impact of this potential loss to follow up, we performed a sensitivity analysis of rCDI risk factors by excluding all patients who resided beyond the greater St. Louis postal code. After excluding the 1230 (31.1%) patients with iCDI who met this criterion, the rCDI risk factors and their hazard ratios did not change appreciably (data not shown). This suggested that our results were not biased by including these patients.

Conversely, rCDI in randomized trials may be subject to a detection bias. Patients in trials are prospectively monitored for recurrent diarrhea and instructed to seek testing if it occurs. Even if the patient’s symptoms are not from CDI, the person may test positive for CDI as many patients continue to shed C. difficile in stool after cessation of CDI treatment [28].

Our study has some limitations. As a retrospective observational study it is prone to several forms of bias, most notably a selection bias. To mitigate this, we enrolled all consecutive patients meeting our enrollment criteria. To avoid misclassification of the main outcome variable, we applied a stringent case definition to CDI, which included a positive toxin assay. Although confounding is an issue with observational data, we adjusted for all the available relevant potential confounders in the regression model. However, the possibility of residual confounding remains. The biggest limitation, however, is its generalizability, since the data reflected patients and treatment patterns at an urban academic medical center with a large referral base, and may not have mirrored those of institutions with different characteristics or patients with iCDI diagnosed and managed completely in the outpatient setting. Additionally, many of the patients who resided outside the St. Louis metropolitan area may not have had their specimens retested at the BJH laboratory. After excluding these patients from the analysis as part of a sensitivity analysis, neither the rCDI hazard ratios nor the rCDI risk factors were majorly impacted in the overall cohort.

In summary, we have demonstrated that a number of modifiable factors exist whose presence raises the risk for developing rCDI. Avoiding such exposures as non-CDI antimicrobial treatment and gastric acid suppressors may go a long way toward attenuating the burden of rCDI. On the other hand recognizing CO-HCFA and advanced age as predispositions to rCDI should serve patients and clinicians well by highlighting the importance of targeting these populations for more aggressive prevention efforts.