The present study used data in an endemic setting from three medical and/or surgical centers of the multicenter prospective cohort OUTCOMEREA™, with homogeneous procedures for microbiological diagnosis of CDI. Patients were included between January 1999 and January 2009. ICU-acquired CDI was defined as watery or unformed stools, according to the Bristol stool chart 9, in a 24-hour period occurring ≥ 72 hours after ICU admission with a laboratory confirmation of a stool sample positive for C. difficile toxin A or B by an immunoassay enzyme 10. Two control groups were chosen, the first including patients hospitalized at the same time in the same unit with watery or unformed stools in a 24-hour period occurring > 72 hours after ICU admission, but with a stool sample negative for C. difficile toxin A or B and a negative stool culture. The second one comprised patients hospitalized at the same time and the same unit. Pseudomembranous colitis was defined as the presence confirmed by endoscopy of typical plaque-like lesions of the pseudomembrane in patients with CDI as defined above.

All codes and definitions were established prior to the study initiation. All practitioners used the same definition before any testing. Moreover, the Quality of the Database was systematically controlled. The data-capture software automatically conducted multiple checks for internal consistency of most of the variables at entry in the database. Queries generated by these checks were resolved with the source ICU before any incorporation of the new data into the database. At each participating ICU, the data quality was controlled by having a senior physician from another participating ICU checking a 2% random sample of the study data. A one-day coding course is organized annually with the study investigators and clinical research organization monitors.

The following data were collected: admission characteristics - age, sex, and origin; body weight; diagnosis at ICU admission; admission category - main reason for ICU admission; chronic diseases; McCabe score; main clinical features; and treatments used, including antimicrobials. The following scores were computed at admission, then once a day: Simplified Acute Physiologic Score (SAPSII) 11, Logistic Organ Dysfunction (LOD) 1213, and Sequential Organ Failure Assessment (SOFA) 1214. Daily data about use of procedures, antibiotic consumption and proton-pump inhibitor were also collected. We recorded the durations of invasive mechanical ventilation, of the ICU and hospital stays, vital status at ICU and at hospital discharge. According to French law, this database study did not require informed consent.

Results are expressed as frequencies and percentages for categorical variables, and as medians and quartiles for continuous variables. Independent risk factors of ICU-acquired CDI were identified using multivariate logistic regression (See Additional file 1). Patients were followed from ICU admission to the occurrence of one event, or censored at ICU discharge. Two different analyses were performed using either the overall population or only the patients with diarrhea and sampled for CDI.

In the overall population analysis, univariate risk factors of ICU death were detected using a Cause Specific Hazard model 15. ICU admission was considered as time 0. Death in the ICU was the variable of interest, whereas discharge alive from ICU was considered as a competing event with ICU death 16. ICU-acquired CDI was included as a time-dependent variable, which equals to 0 before infection, and to 1 from the day of CDI until the end of the follow-up. Last, a Cause Specific Hazard model was conducted to assess the impact of CDI on prognosis, with adjustment on time-fixed and time-dependent confounding factors, such as iatrogenic events occurring between admission and the CDI, bloodstream infection during ICU stay (BSI), ventilator-associated pneumonia (VAP), surgical site infection, pneumothorax during ICU stay, gastrointestinal bleeding during ICU stay, and severe hypernatremia 1718.

In the second analysis, we only used patients with diarrhea. The time of CDI test performance was considered as Day 0, and CDI infection was considered as a time-fixed covariate. Other covariates were introduced in a Cause Specific Hazard model as previously described.

Results were presented with Cause Specific Hazard ratios (CSHRs) and 95% confidence intervals (95% CI). Models were stratified by center.

Finally, we estimated the prolongation of ICU stay using the disability model approach 19. We used a multi-state model with four states, and all diarrheic populations started in an initial state. Then, prolongation of ICU stay was determined by reaching one of two competing absorbing states, (death or discharge alive), by taking into account the intermediate state (ICU-acquired CDI). Finally, we computed standard error estimation for prolongation of ICU stay thanks to the bootstrap method and 2,000 random samples with replacement and computed P-value using the Wald test. P-values < 0.05 were considered significant. Statistical analysis was performed using SAS 9.1 (SAS Institute, Cary, NC, USA). Length of stay prolongation was calculated with R software (R foundation, Vienna, Austria), using the change LOS library.

Assuming a 40% rate of hospital death in the diarrheic population, 471 patients were necessary to detect a hazard ratio (HR) of 2 for death with greater than 90% power and a type I error of 0.05 20. Similarly, 4,290 patients were necessary, assuming a 35% rate of hospital death in the whole population.

According to French law, this study did not require patient consent, as it involved research on a database. The study was approved by the institutional review board of the Centre d'Investigation Rhône-Alpes-Auvergne.

The impact of CDI on mortality was homogeneous across centers (data not shown). ICU death in patients with CDI infection was associated with a high LOD score (P = 0.01), a high McCabe score (P = 0.02), and with immunosuppression (P = 0.02). Two different groups were used to analyze the impact of ICU-acquired CDI on patient's outcome. The first analysis compared patients discharged alive (n = 4,135) versus those dead (n = 1,125), and showed that CDI had no significant effect on mortality as a crude (CSHR = 0.64, 95% CI 0.34 to 1.21, P = 0.17) or adjusted factor (CSHR = 0.71, 95% CI 0.38 to 1.35, P = 0.3) (See Table 3).

The second analysis compared ICU-acquired CDI patients with diarrheic patients with negative stool culture: the crude effect of CDI on mortality was still not significant (CSHR = 0.70, 95% CI 0.36 to 1.35, P = 0.3). Moreover, after adjustment on confounding factors and iatrogenic events between admission and occurrence of diarrhoea, the effect on mortality remained not significant (CSHR = 0.81, 95% CI 0.40 to 1.64, P = 0.6) (See Table 4).

Results remained similar when excluding patients with metronidazole (IV or oral) or vancomycin (oral) before the diagnostic test (adjusted CSHR = 0.84, 95% CI 0.41 to 1.71, P = 0.6), or when only taking into account cases with pseudomembranous colitis acquired in ICU (n = 24 cases, CSHR = 0.80, 95% CI 0.31 to 2.03, P = 0.6).

Results would only be slightly different if the 95/443 (21.4%) given metronidazole or vancomycin within 48 hours following a negative test were excluded from the statistical analysis. (adjusted CSHR_Death = 0.80 (0.39 to 1.65), P = 0.5, CSHR_discharge = 0.68 (0.46 to 1.00), P = 0.0475).

The median length of ICU stay in the whole population of diarrheic patients was 17 (8 to 34) days, whereas the median length of ICU stay in the CDI population was 20 (12 to 42) days. Using a multistate model, the estimated prolongation of ICU stay for the diarrheic population due to C. difficile was 8.0 days ± 9.3 days, P = 0.4.

Moreover, the median length of ICU stay in the whole population was 4 (3 to 9) days, whereas the median length of ICU stay in the ICU-acquired CDI group was 20 (12 to 42) days. The estimated prolongation of ICU stay due to C. difficile was 6.3 days ± 4.3, P = 0.14

First, the choice of control groups may influence findings, as this has been outlined in a number of epidemiological publications 2223. Our methodological approach was to minimize bias due to the characteristics of the control group by comparing patients with ICU-CDI, to patients with diarrhea not linked to C. difficile, and to the whole ICU population. Indeed, there is much potential selection bias that arises if we choose only patients with diarrhea as a control group. On the other hand, controls should be selected from the same source population or study base that gives rise to the cases. The patients whose stools have been sampled are possibly different from the ones that have not been sampled.

However, the variability of the patient populations might also explain the variability in the association between mortality and CDI disease in the patient populations under study. Our study population included all ICU patients, and was different from that of other studies that were interested in specific selected populations, such as older persons, ill patients or burn unit patients.

Finally, our epidemiological situation is different from North America's, as none of our patients had been infected with NAP1/O27 isolates. As this strain seems to be more virulent comparatively to others, our lower mortality rate could be explained by this microbiological difference. Indeed, in recent years with the emergence of a hypervirulent strain, the annual frequency of and the case fatality due to CDI have doubled in the United States 22425. Moreover, authors 1 demonstrated a higher mortality rate among inpatients in which nosocomial CDI developed compared to control subjects without CDI, matched for sex, age and disease severity; but this attributable mortality was measured during the CDI epidemic in Quebec caused by the hypervirulent strain NAP1/O27. Finally, the antimicrobial treatment was instituted early in CDI patients and may have decreased the impact of CDI on mortality and length of stay.

A second consideration that may explain differences in findings among studies conducted to date is in the analysis with adequate adjustment for confounding variables and competing events for mortality. Failure to adequately adjust for factors differently distributed among patients with or without CDI that also affect their outcome may lead to different conclusions. A number of factors could explain mortality in the ICU, such as advanced age and severity of illness at onset, and the presence of sepsis or septic shock. We used a modern statistical model that is frequently applied in other medical fields, such as cancer epidemiology. This approach is based on event histories, model time-to-event and may focus on time-dependent risk factors, such as nosocomial infections. Modern statistical methods are further able to simultaneously analyze different endpoint types, and they explicitly account for the timing of events 16. Indeed, a case-control study could have led to different findings. It is important to underline that nosocomial infection is a time-dependent event. Occurrence of nosocomial infection is a time-dynamic process, and the discharge acts as a competing risk when estimating the relationship between nosocomial infection and death. Both factors may bias the attributable mortality estimate. Matching patients with and without CDI infection on ICU duration and then performing conditional logistic regression is a widely used method to evaluate nosocomial infection (CDI here). The attributable mortality method is also used for other events that are dependent on the duration of exposure to a risk factor. With this method, each patient is classified as being exposed to CDI or unexposed (no CDI). In exposed patients, the data are handled as if the exposure was present at the study initiation (although exposure status is determined at study completion). Thus, the excess risk of death associated with the exposure is assumed to be present throughout the ICU stay, that is, both before and after the occurrence of the exposure. In other words, the exposure is handled as a time-independent variable. If the exposure is actually time-dependent, then a bias is introduced. Therefore, the impact of a time-dependent exposure on mortality is overestimated with this method. Our statistical model, in contrast, considers that the excess risk of death associated with the exposure exists only after the exposure onset. In this multistate model, each patient goes through two or more states. Thus, at study initiation, all patients are classified as being in the unexposed state. Over time, some patients acquire the exposure of interest (here, CDI); therefore switching to the exposed state, at different time points during the ICU stay. Eventually, the model fits reality far more closely than does the matched cohort design, resulting in the narrowest confidence intervals. The main advantage of using the multistate model for complete data is that mortality can be estimated over time. Therefore, the changes in the mortality rate over time can be detected.

We previously showed that about a quarter of ICU patients experienced more than one adverse event, and that nosocomial infections, such as ventilator-associated pneumonia, ICU-acquired bloodstream infections, deep and organ/space surgical site infection without BSI, and adverse events, such as pneumothorax, gastrointestinal bleeding 18 and hypernatremia 17, were independently associated with mortality. The multistate model we used allows us to avoid the estimation bias associated to these events 1626.

Procedure for C. difficile detection is clearly defined in all study centers, and is only used in cases of watery stools. The toxin assay we used possesses an excellent specificity, but an 80% sensitivity 27. It is, therefore, possible that some patients may have been falsely classified as belonging to the diarrheic CDI-negative control group, and thus decreased the study power (that is, the probability to find a difference if it exists). However, in the diarrheic CDI negative patients, no instance of hospital-acquired CDI was diagnosed after ICU discharge. Finally, our study was conducted in three French ICUs (in Grenoble and the Paris region), so our results cannot be extrapolated to the whole of France.