Introduction

Since 2002, outbreaks caused by Clostridium difficile infection (CDI) have been reported in Canada, the USA and Europe, associated with the emergence of a new hypervirulent type. This type has been characterised as North American pulsed-field type 1, restriction endonuclease analysis group type BI, toxinotype III and polymerase chain reaction (PCR) ribotype 027 (type 027) [1–5]. During outbreaks in the USA and Canada, the reported incidences of CDI varied between 155 and 225 per 10,000 hospital admissions [3, 6]. Peak incidences of CDI due to type 027 during outbreaks in the Netherlands were remarkably lower, around 50 per 10,000 hospital admissions [7, 8].

Most recent studies on the risk factors of CDI focussed on outbreaks, whereas less is known about CDI in settings with a low incidence. Well described risk factors for CDI in outbreak situations are prior use of antibiotics, increased disease severity and, in case of outbreaks caused by type 027, advanced age and prior use of fluoroquinolones [9–11].

The aim of our study was to identify risk factors for CDI in a true endemic setting. A second aim was to establish risk factors specific for CDI, in comparison with factors for diarrhoea in general. To answer these questions, we performed a prospective case–control study at the Leiden University Medical Center, the Netherlands, during a period of 34 months.

Methods

Patients

From July 2006 through April 2009, all hospitalised patients with CDI were included in the study. Tests for CDI were performed daily upon request and on all unformed faecal samples from patients admitted for two days or more, regardless of the physicians’ requests. For each hospitalised CDI patient, two controls were included, matched for the ward at which CDI was diagnosed and the time of admission. The controls included one control patient without diarrhoea (control patient) and one control patient with diarrhoea and a negative C. difficile toxin test (non-CDI patient). Controls were consecutive patients on the alphabetical ward list.

Definitions

Definitions as proposed by the European and American Centres of Disease Control were used [2, 12]. Diarrhoea was defined as ≥3 unformed stools per 24-h period. CDI was defined as the presence of diarrhoea in combination with a positive toxin test for C. difficile. A community association was defined as the development of CDI outside the hospital or within 48 h after admission, without a history of admission in the previous 3 months. We defined diarrhoea as severe when it occurred with one or more of the following: bloody stools, hypovolaemia, fever (T > 38.0°C) and leukocytosis (>12.0 × 109/l), hypo-albuminaemia (<20 g/l) or pseudomembranous colitis. A complicated course of CDI was defined as: admission to the intensive care unit (ICU), a surgical intervention in association with CDI or death within one month. Mortality was considered to be contributable to CDI when a patient died during admission, partly due to the consequences of CDI.

Isolation and characterisation of Clostridium difficile

C. difficile toxins in stools were detected by VIDAS C. difficile toxin A during the first 12 months of the study and VIDAS toxin A/B assay during the ensuing 22 months (bioMérieux, France). Each positive sample was cultured. Available isolates were identified as C. difficile using a PCR assay to detect the presence of gluD and were PCR ribotyped as previously described [8, 13].

Data collection

Approval was obtained from the Medical Review Ethics Committee to collect demographical and clinical patient data. Information was collected on patients’ age, sex, co-morbidity, ward of acquisition, disease severity, clinical course and mortality. Furthermore, data were collected on surgery, invasive procedures, admissions, use of antibiotics and other medications in the 3 months prior to CDI. We gathered this information through consultation of the physician in charge, as well as by using patient records and the hospital electronic medical information system. The period of 3 months prior to CDI was determined by calculating backwards from a reference date. For CDI and non-CDI patients, this reference date was defined as the day on which the diarrhoea started. The reference date for control patients was determined by adding the hospitalised period of the matched CDI patient (time between admission and start of diarrhoea) to the admission date of the control patient. Co-morbidity was assessed by both the Charlson co-morbidity index and the ICD-10 classification in ten disease groups, as mentioned in Table 1 [14].

Table 1 Baseline characteristics of patients with Clostridium difficile infection (CDI), patients with non-CDI diarrhoea and control patients
Full size table

Statistical analysis

Continuous data were compared between groups using the t-test. Pearson’s Chi-square test and Fisher’s exact test were used for the analysis of proportions. Factors that were associated in the univariate analysis (UVA) with a p-value <0.10, as well as putative risk factors from earlier studies, were analysed in a multi-variable model. Here, associations were always adjusted for age, sex, ward and Charlson co-morbidity index. To evaluate the effect of medications and interventions on (CDI) diarrhoea, we performed additional adjustments for co-medication and other interventions. When comparing non-CDI patients with control patients, we also corrected for the time between admission and the reference date. Relative risks were estimated as odds ratios (ORs) and presented with a 95% confidence interval (95% CI). Statistical significance was reached with a two-sided p-value <0.05; trends were defined by a p-value <0.10. All analyses were performed using the SPSS for Windows software package, version 17.0.

Results

During the 34-month study period, 93 patients were diagnosed with CDI. The incidence varied from 0 to 43 per 10,000 hospital admissions, with an average of 17.5. During this period, no outbreaks were observed. CDI was community-associated in four patients (4.3%). Most patients (n = 30; 32.3%) were hospitalised at the Department of Internal Medicine, followed by the general surgery ward (n = 15; 16.1%). Eighty-nine CDI patients were positive on both toxin testing and culture (95.7%). Isolates from 63 (67.7%) patients were available for PCR ribotyping: type 014 was the most frequently found type (n = 10; 15.9%), followed by types 078 (n = 8; 12.7%) and 015 (n = 5; 7.9%). Type 027 was not present. Three patients with CDI had a co-infection with an enterovirus, norovirus and Cryptosporidium, respectively.

The 93 CDI patients were compared to 76 control patients and 64 patients with non-CDI diarrhoea. Of all patients, physicians responded and records were available; however, in some cases (the exact number is depicted in the footnotes of Table 1), no information about the use of nasogastric intubation, surgery or endoscopy was noted.

In the group of non-CDI patients, two patients were diagnosed with a rotavirus and Giardia lamblia, respectively. Among the other 62 patients, no causal agent was found. CDI patients had a median age of 56 years; non-CDI diarrhoea and control patients had a median age of 50 years. Of the CDI patients, 60% were male, compared to 50% and 54% of the non-CDI and control patients, respectively. The time span between admission and the start of diarrhoea did not significantly differ between CDI and non-CDI patients.

Characteristics and risk factors

We present the baseline characteristics and risk factors for CDI and non-CDI diarrhoea in Tables 1 and 2. The use of antibiotics as a risk factor for CDI and non-CDI is depicted in Table 3. All of the following results reached statistical significance in the multi-variable analysis (MVA), unless otherwise stated.

Table 2 Crude and adjusted odds ratios (ORs) for the development of CDI and non-CDI diarrhoea
Full size table
Table 3 The use of antibiotics expressed in defined daily doses (DDDs) in patients with CDI and non-CDI diarrhoea and control patients
Full size table

Clinical course

Severe diarrhoea was present among 51 hospitalised patients with CDI (58.6%) and 25 patients with non-CDI diarrhoea (39.7%) (OR 2.22; 95% CI 1.14–4.30). No significant differences between CDI and non-CDI diarrhoeal patients were found regarding the frequency of fever (55.6% vs. 43.3%), bloody stools (12.2% vs. 12.9%) or abdominal pain (54.5% vs. 48.2%). CDI patients did, however, have a higher white blood cell count (≥15 × 109/l: 49.9% vs. 30.0%, OR 2.28; 95% CI 1.13–4.59). Most patients with CDI were treated with metronidazole (n = 57; 63.3%), two patients (2.2%) were treated with vancomycin and in 27 patients (30.0%), no specific CDI treatment was initiated. The 30-day and 60-day mortality rates are depicted in Fig. 1. At one month follow-up, a complicated course was observed in nine CDI patients (10.3%), comprising two colectomies, four ICU admissions due to CDI and seven deaths (7.5%). CDI contributed directly to three of these deaths, but was not the primary cause. One non-CDI patient (1.6%) and none of the control patients died at one month follow-up. No significant associations were detected between the severity of the diarrhoea, treatment or outcome.

Fig. 1
figure 1

Survival curve of patients with Clostridium difficile infection (CDI), non-CDI diarrhoea and control patients in a period of 60 days after the reference date

Full size image

Discussion

In this 34-month prospective case–control study, risk factors for CDI were studied in an endemic setting with a low incidence rate of CDI. The inclusion of a control group of patients with diarrhoea, tested negative for CDI, enabled us to discriminate between risk factors for CDI and for diarrhoea in general.

Common risk factors for CDI outbreaks, such as age above 65 years and a high co-morbidity index, were recognised as trends in our study. This may be due to the fact that these risk factors are of less importance in endemic settings, resulting in a lack of power to discern these risk factors. Other well known risk factors for CDI, such as the use of second-generation cephalosporins and previous (ICU) admission, were also found in this endemic situation [3, 10, 15]. Conversely, the use of fluoroquinolones or proton pump inhibitors (PPIs) was not a risk factor for CDI. Furthermore, the previous use of third-generation cephalosporins was a risk factor for diarrhoea in general.

The CDI incidence in our hospital was lower than that described in other studies in endemic situations, but comparable to the incidence of 18 per 10,000 hospital admissions found in other Dutch hospitals [16]. Recently, a retrospective study analysing risk factors for CDI in an endemic setting in the USA reported an incidence rate of CDI of 106 per 10,000 hospital admissions, which is a factor of 5 higher than what we found in this study [10]. There seems to be a considerable difference, per hospital and per country, in the application of the definition of endemic CDI. Therefore, reported rates of endemic CDI may merely reflect a baseline incidence.

In outbreak situations, the previous use of fluoroquinolones has been recognised as an important risk factor for CDI [9, 11, 17, 18]. This association may be due to disruption of the gut flora by newer fluoroquinolones or the high fluoroquinolone resistance found among hypervirulent type 027 strains [19]. Although fluoroquinolones (mainly ciprofloxacin) were frequently prescribed in this study, we found no association with CDI. An explanation could be that we did not encounter type 027 in our hospital. The most frequently found PCR ribotypes in our study (types 014, 078 and 015) are commonly found in the Netherlands and Europe, and are more susceptible to fluoroquinolones than type 027 [11].

The use of vancomycin was previously recognised as a risk factor for endemic CDI [10]. Instead, in this study, the association between vancomycin and CDI was strongly confounded by the concomitant use of second- and, especially, third-generation cephalosporins (the combination is part of the in-house empirical sepsis therapy) and was not a risk factor for CDI.

PPIs raise the gastric pH, which is associated with enhanced bacterial colonisation of first part of the gastro-intestinal tract. Studies on the use of PPIs in association with CDI revealed conflicting conclusions [20, 21]. In our study, we found no association of the use of PPIs with CDI. It should be noted that half of the non-CDI and control patients also used PPIs.

Earlier studies have found high contamination and colonisation rates with C. difficile spores in the hospital environment, among hospitalised patients and asymptomatic carriers [22, 23]. A high colonisation pressure on a ward (exposure in time to multiple colonised or infected patients) is associated with an increased risk of CDI [10]. To ensure that CDI and control patients were exposed to a similar colonisation pressure, we selected control patients from the same ward as CDI patients using the same time period between admission and reference date [24].

We observed a contributable and overall mortality of 3.2% and 7.5% after 30 days follow-up, respectively. These proportions are in between the overall 30-day mortality of 4.7%, found in an endemic setting in Canada, and 20% mortality after 60 days in a USA study [25, 26]. These mortality rates are much lower than those reported during outbreaks caused by the type 027 strain [3, 11, 26, 27]. In the Netherlands, a complicated course due to type 027 was described in 12.5%, with an attributable mortality rate of 6.3% [9].

Our study has several limitations. First, we used the presence of toxins in faeces as a screening test for CDI, which is in agreement with the European recommendations [28]. An alternative standard for diagnosing CDI is the detection of C. difficile in faeces by toxinogenic culture or PCR. Application of this definition could have resulted in a different case and non-CDI control group. However, none of the patients with non-CDI diarrhoea developed CDI at a later time during admission, which was in accordance with the high negative predictive value of our toxin test. Second, although the endemic incidence found in our study is comparable to that in other Dutch hospitals, it is lower than the incidence rates reported in other studies in endemic situations, which can imply that our findings may not be applicable to endemic situations in other countries [8, 26, 29].

In conclusion, in this endemic setting, some risk factors for CDI were similar to those found in outbreak situations, but some risk factors that have been ascribed to CDI earlier were, in this study, not specific for CDI, but for diarrhoea in general. The use of fluoroquinolones and PPIs did not influence the risk of endemic CDI. CDI patients were more severely ill than non-CDI diarrhoeal patients, as illustrated by a higher leukocyte count and the relatively high 30- and 60-day mortality. Because CDI is the most important cause of nosocomial diarrhoea, more studies are needed in order to determine the long-term outcome associated with C. difficile infections.