Variability in testing policies and impact on reported Clostridium difficile infection rates: results from the pilot Longitudinal European Clostridium difficile Infection Diagnosis surveillance study (LuCID)

Lack of standardised Clostridium difficile testing is a potential confounder when comparing infection rates. We used an observational, systematic, prospective large-scale sampling approach to investigate variability in C. difficile sampling to understand C. difficile infection (CDI) incidence rates. In-patient and institutional data were gathered from 60 European hospitals (across three countries). Testing methodology, testing/CDI rates and case profiles were compared between countries and institution types. The mean annual CDI rate per hospital was lowest in the UK and highest in Italy (1.5 vs. 4.7 cases/10,000 patient bed days [pbds], p < 0.001). The testing rate was highest in the UK compared with Italy and France (50.7/10,000 pbds vs. 31.5 and 30.3, respectively, p < 0.001). Only 58.4 % of diarrhoeal samples were tested for CDI across all countries. Overall, only 64 % of hospitals used recommended testing algorithms for laboratory testing. Small hospitals were significantly more likely to use standalone toxin tests (SATTs). There was an inverse correlation between hospital size and CDI testing rate. Hospitals using SATT or assays not detecting toxin reported significantly higher CDI rates than those using recommended methods, despite testing similar testing frequencies. These data are consistent with higher false-positive rates in such (non-recommended) testing scenarios. Cases in Italy and those diagnosed by SATT or methods NOT detecting toxin were significantly older. Testing occurred significantly earlier in the UK. Assessment of testing practice is paramount to the accurate interpretation and comparison of CDI rates.

A recent US study, based on data from 3.6 % of the population, estimated a total of 453,000 cases in 2011 [4]. Notably, a sensitivity analysis explored the effects of testing method: the total number of cases could vary from 286,300 to 701,100 if the proportion of samples tested for CDI by polymerase chain reaction (PCR) ranged from 0 to 100 % [4]. Using single C. difficile targets (e.g. toxin gene) can exaggerate reported case rates, as such testing policies do not differentiate between colonisation and true CDI [10][11][12]. The detection of C. difficile toxin correlates with mortality, disease severity and complications [11,13], but available assays are sub-optimal, as standalone toxin tests (SATTs) have sub-optimal sensitivity and specificity.
In addition to testing methodology, diagnostic intensity (mostly driven by CDI awareness) is a key determinant of reported rates [8,14,15]. In 2012, only ∼40 % of 482 hospitals across Europe were using recommended laboratory methods to diagnose CDI [8]. Furthermore, 23 % of all CDI-positive samples identified at study coordinating laboratories were never even tested at the submitting hospital due to a lack of clinical awareness [8]. We have used a systematic, observational prospective large-scale sampling approach to investigate variability in C. difficile sampling, testing and reported CDI rates.

Methods
Sixty hospitals across France, Italy and the UK (20 per country) were recruited by national coordinators to represent a wide geographical area. Via a questionnaire, hospitals provided institutional data (size and type of institution) and details of current CDI laboratory diagnostic methods and policies (April 2014 to March 2015). We obtained monthly data for hospital in-patients on the numbers of enteropathogen and CDI tests, and CDI-positive cases, along with case demographic data. Cases were defined as 'primary' if the first case in the patient or 'recurrent' if a second positive sample occurred within 2-8 weeks of a previous positive. All data were uploaded prospectively by participating hospitals to a secure, dedicated, webbased, study database. All data analyses were conducted by the European coordinator.

Data analysis
The testing methodology, testing/CDI rates and case profiles were compared between countries and different sized institutions. Hospitals were classified as small (<100,000 pbds per annum), medium (100,000-500,000) or large (>500,000). Annual testing and case rates were calculated per 10,000 pbds for each hospital, and mean rates were compared between countries and between different sized institutions. The mean CDI case incidence and patient profiles were compared between hospitals using a recommended testing algorithm (GDH/toxin or NAAT/toxin) versus methods not detecting toxin (e.g. GDH/NAAT or NAAT alone), SATT (e.g. toxin EIA alone) or a nonrecommended algorithm (e.g. GDH/toxigenic culture) [1]. CDI and testing rates were compared by analysis of variance (ANOVA), age distributions by Kruskal-Wallis and proportions were compared by Chi-squared. Analysis was performed on SPSS 19 (IBM). This surveillance study was granted ethical approval by the University of Leeds (SoMREC13032) for UK data collection and European-wide analysis, and by the National Institute for Infectious Diseases 'Spallanzani', Rome for Italian data collection. Ethical approval was not required in France.

Institutional data
Fifty-nine of 60 hospitals completed the questionnaire (one missing in the UK). There were five small, 40 medium and 12 large hospitals; data on the number of pbds were not available for a further two hospitals. There were more small hospitals in Italy than in either France or the UK; Italy had no large hospitals (Table 1).

Testing and case rates: comparison between countries and different sized institutions
The mean annual overall enteropathogen testing rate per hospital was 107.4 tests/10,000 pbds, with significantly more tests carried out in Italy than the other two countries (p < 0.001) ( Table 1). The UK had the highest rate of CDI patient tests/10,000 pbds compared with Italy and France (50.7 vs. 31.5 and 30.3, respectively, p < 0.001) ( Table 1). If the number of in-patient samples sent for enteropathogen detection is used as a proxy for the number of diarrhoeal samples submitted to the laboratory, then, overall, only 58.4 % of diarrhoeal samples were tested for CDI across the three countries. The UK and France tested higher proportions of inpatient diarrhoeal samples for CDI (68.5 and 59.4 %, respectively) than Italy (31.8 %).
The mean annual hospital-reported CDI rate was significantly higher for those institutions using methods that do not detect toxin or SATT (5.2 and 4.0/10,000 pbds, respectively) versus hospitals using recommended algorithms (2.0/10,000 Table 1 Annual testing, Clostridium difficile infection (CDI) and recurrence rates/10,000 patient bed days (pbds) per hospital for each country, and for small (<100,000 pbds per annum), medium (100,000-500,000 pbds) or large (>500,000 pbds) hospitals France No data 1 n/a n/a n/a n/a n/a n/a n/a France Small 0 n/a n/a n/a n/a n/a n/a n/a Italy No data 0 n/a n/a n/a n/a n/a n/a n/a Italy Large 0 n/a n/a n/a n/a n/a n/a n/a Italy Overall 20 150.4 99.2 39.2 31.5 5.0 4.7 0.6 UK No data 1 n/a n/a n/a n/a n/a n/a n/ n/a n/a n/a n/a n/a n/a n/a Overall There was a significant difference in the median age and distribution of ages of cases between countries; CDI patients in Italy were older than those in France and the UK (p < 0.001) ( Table 2). Additionally, the median ages of cases diagnosed in hospitals using methods either not detecting toxin (77 years) or SATT (81 years) were significantly older than those tested in hospitals using the recommended algorithm (75 years, both p < 0.001). Also, patients with recurrent CDI were significantly older than those with primary infection (76 vs. 74 years, p = 0.007). Specialty data were available for 5919 CDI cases; 73.9 % were in medical specialities, 12.7 % surgery, 7.7 % ITU/HDU, 1.3 % obstetrics/gynaecology and 3.0 % paediatrics (    There were more UK surgical patients than in France or Italy (p < 0.001), and more ITU/HDU patients in France than in Italy or the UK (p < 0.001).

Discussion
This pilot study demonstrates the feasibility of collecting and comparing prospective data on CDI testing across hospitals and countries. We have clearly shown the impacts of testing policies and methodologies on the reported rates of CDI and case demographics. There are three main drivers of CDI rates that can distort true incidence: diarrhoea sampling frequency; rate of CDI testing (testing/requesting frequency) and laboratory method(s). A European study in 2008 showed a close correlation between testing frequency and reported CDI rate [8]. Five years later, albeit using a considerably larger hospital cohort, this association was much weaker, likely reflecting greater heterogeneity of testing methods [8].
Targeted testing of 'at-risk' patients will also likely impact on CDI rates. CDI patients in Italy were significantly older than those in the other two countries. Either CDI cases in Italy are genuinely older or, more likely, C. difficile testing is generally performed on older patients. Although older age is often reported as a risk factor for CDI, this may, in fact, be a proxy marker for co-morbidity risk [16]. We have shown previously that CDI under-diagnosis is more likely to occur in younger individuals, presumably due to the lack of clinical suspicion and/or testing policies that give insufficient prominence to (real or perceived) lower risk groups [8]. In the present study, patients diagnosed by non-recommended methods were significantly older than those identified by recommended testing algorithms. It appears, therefore, that targeted testing and the use of non-recommended testing methods may be linked.
Interestingly, whilst UK hospitals appear not to be targeting C. difficile testing in the elderly, there is a clear bias towards earlier sampling/testing, compared with practice in institutions in either France or Italy; CDI cases were tested a median of 3-7 days earlier in the former (median 3 vs. 6 and 10 days, respectively; p < 0.001). Earlier CDI testing in the UK likely reflects the desire to apportion cases as 'community-' (diagnosed before or within the first 2 days of hospital admission) rather than 'hospital-acquired' (diagnosed after 2 days of hospital admission); notably, there are financial penalties associated with excess 'hospital-acquired' CDI cases in the UK. Earlier testing is also likely to be partly driven by differing country policies for sample submission; in France and Italy, sampling is recommended after 48 h of diarrhoea, whereas in the UK, this can occur after only one diarrhoeal episode [17]. It is possible that such earlier testing will result in higher reported CDI rates, as some cases would have resolved without treatment.
Sub-optimal CDI diagnosis is still prevalent within Europe [8,14,15], despite guidelines having been issued in 2009 [1]. Only 64 % of the 60 hospitals in the present study used recommended laboratory testing algorithms [1]. More UK hospitals used recommended algorithms for CDI diagnosis (85 % vs. 40-65 %), reflecting performance management of national guidelines that were issued in 2012 [17]; indeed, the UK rate of 'recommended' testing has increased since 2013 (76 %) [8]. According to our data, almost a third of Italian hospitals are still using SATT for CDI diagnosis, despite clear evidence of their poor prognostic performance (i.e. sub-optimal positive and negative predictive values) [18,19]. Small hospitals were significantly more likely to use SATT, although all 3/5 such hospitals were in Italy; thus, this association may be a country effect and/or related to institutional size. There was, however, a significant inverse relationship between hospital size and both C. difficile testing rate and CDI case frequency (Fig. 1). It is likely that high CDI rates in small hospitals reflect more testing. However, it is possible that there is also some confounding here, as the use of SATT is associated with a high false-positive CDI rate [11,18,19]. In addition, smaller hospitals may have a different patient and ward mix to that seen in larger institutions; this again may impact on CDI rates.
In hospitals using diagnostic methods that only detected toxin or did not detect toxin at all, CDI rates were significantly higher than in institutions using recommended algorithms, despite testing rates being similar (Table 3). It could be argued that assays not detecting toxin (including standalone NAAT testing) are more sensitive and, thus, are detecting more cases, which are missed by other methods. However, several studies have shown that diarrhoeal patients that are C. difficile NAATpositive only do not have higher mortality or CDI complications than controls, in contrast to cases defined by the presence of faecal C. difficile toxins [10][11][12][13]. It is likely, therefore, that the higher case rates seen in hospitals not using toxin detection are due to the detection of colonised patients in addition to those with true CDI.
There are several limitations to this pilot study. Firstly, this was an observation study that collected data at the laboratory level, with no information available on symptom severity; we presumed that diarrhoea was present, based on assumed clinical decisions to sample and/or request C. difficile testing. Institutions that participated in this study were chosen by the study coordinators and may have had a higher CDI awareness and/or rates; however, institutional selection was intended to be 'representative' of hospitals nationally, and to cover a wide geographical spread. We did not collect demographic data for patients with samples that tested negative for C. difficile/CDI. This would provide better evidence for the age distribution of cases/controls and help to show how testing is targeted. Such data will be collected in the follow-on full LuCID study, which will also expand to include a total of 40 hospitals in each country and two more countries (Spain and Germany). The additional countries and hospitals will increase the sizes of the (at present) small sub-groups, such as small hospitals and CDI recurrences. As patient bed day data were collected annually rather than per month, we were not able to determine with confidence possible seasonality of either CDI testing or case rates. These limitations will also be addressed in the full LuCID study.
Our pilot results show that it is important to understand the context of sampling, testing and methodology in order to interpret reported CDI rates. It is essential that studies either take such potential confounders into account when interpreting data, and, ideally, that large, multi-centre, multi-country studies are used to determine the true epidemiology of CDI. Additionally, we have shown that inclusive testing with recommended diagnostic methodologies is associated with lower reported rates of CDI. Theoretically, increased awareness of true CDI cases enhances opportunities to implement appropriate and targeted infection prevention and control measures. 'Missed' cases, either through lack of clinical suspicion or the use of non-recommended laboratory diagnostics, may facilitate the transmission of C. difficile because of unrecognised reservoirs of infection. Conversely, false-positive cases may be receiving antibiotics for assumed CDI that could be detrimental, potentially having deleterious or resistance selection effects on gut microbiota and possible induction of true CDI in some instances.