Statistical analysis.

Acquisition of ARE, among patients not colonized with ARE on admission, was the primary outcome. Acquisition was assumed to happen exactly between the last negative and the first positive swab. Patients who did not acquire ARE colonization were censored at discharge (when the last negative culture was obtained <48 h before) or 1 day after the last screening sample. Once colonized, patients were no longer considered at risk for acquisition.

The data were analyzed with a Cox proportional hazards model with the intervention, probiotics, incorporated as a dichotomous time-dependent covariate. In a second analysis this variable was replaced by study period (intervention or control) to study the ecologic effect of probiotics. The following variables were designated potential confounders: age and ward as time-constant covariates and colonization pressure, isolation, and use of antibiotics as time-dependent covariates. Colonization pressure was calculated per day for each patient at risk.

Colonization on admission was defined as ARE carriage documented <48 h after admission. If the first culture was taken >48 h after admission and yielded ARE, the patient was considered to be colonized from the midpoint between the day of admission and the first positive swab. There were no infection control measures for ARE during the study periods, and general hygiene and infection control policies did not change.

Antibiotics active in the intestinal compartment were distinguished into two groups, based on their effect (or lack of effect) on ARE (Table ​(Table1)1) (20). Here, we considered piperacillin-tazobactam to be intestinally active against ARE, based on its ability to achieve high biliary concentrations in human bile exceeding the MICs for most multiresistant enterococci (5, 21). A third antibiotic group was designed to account for an “ecological postantibiotic” effect of 3 days, as restoration of the indigenous flora takes time after antibiotic discontinuation. Sensitivity analyses were done with 0 and 5 days, and statistical interaction between antibiotics and probiotics was considered.

TABLE 1.

Antibiotic use by patients at risk for ARE acquisition

Antibiotic	No. of patients	No. of days
Antibiotics active in the intestinal compartment
Oral
Amoxicillin	3	12
Amoxicillin-clavulanic acid	38	204
Azithromycin	2	6
Ciprofloxacin	41	279
Clarithromycin	1	5
Clindamycin	2	13
Cotrimoxazol	42	361
Erythromycin	3	24
Norfloxacin	4	45
Tetracycline	2	20
Trimethoprim	5	26
Vancomycina	1	6
Flucloxacillin	2	7
Intraperitoneal
Cefazolin	1	3
Cephalothin	1	2
Intravenous
Amoxicillin	3	10
Amoxicillin-clavulanic acid	48	237
Cefazolin	4	5
Cefotaxime	1	7
Ceftazidime	1	8
Ceftriaxone	61	403
Ciprofloxacin	6	11
Erythromycin	2	6
Meropenem	5	36
Metronidazole	2	2
Piperacillina	1	2
Piperacillin-tazobactama	4	21
Antibiotics without activity in the intestinal compartment
Intraperitoneal
Vancomycin	1	5
Intravenous
Cefuroxime	7	7
Gentamicin	20	35
Vancomycin	2	5
Benzylpenicillin	1	29
Flucloxacillin	5	27

Open in a separate window

^aAntibiotics active against ARE in the intestinal compartment.

Allowing covariates to change in time might lead to biased estimates, as the relationship between determinants and outcome is longitudinal (6). Moreover, there will be a delay between actual acquisition and reaching the detection limit of the culture method. Therefore, risks of acquisition (or, more correctly, of culture positivity) will be influenced by the values of covariates on 1 or more days before that time point. This was investigated by fitting a full model (with all covariates) to different values of lagged time (0 to 3 days) for the time-dependent covariates, which were compared using Akaike's information criterion (AIC; in this case the AIC is equivalent to the likelihood, as the numbers of parameters are similar in all models), and the model with the best fit was selected for further analyses (1).

The Cox model was fitted with the survival and spline packages in R 2.8.0, using the Efron method for ties. We used AIC to determine the model with the best balance of parsimony and fit of the data (i.e., the model with the lowest AIC value). The proportional hazards assumption was assessed by formal tests and graphically, using scaled Schoenfeld residuals.

In a sample size calculation we determined that, assuming equal proportions of patients in both the probiotic and control groups, at least 66 acquisitions were required to detect a hazard ratio (HR) of 0.5 at a 5% significance level (two sided) with a power of 80% (22). Based on previous admission data (on average 850 and 250 admissions per year with a length of stay of >48 h on the gastroenterology/nephrology and hematology wards, respectively), an assumed inclusion rate of 70%, and a formerly documented acquisition rate of 15%, we set the length of the study to at least two times 3.5 months (4).

Comments.

In a setting with a high prevalence of intestinal carriage with multiresistant enterococci, daily intake of probiotics did not reduce ARE acquisition rates nor did it change the enterococcal ecology on a ward level. Exposure to antibiotics for which ARE is resistant, though, increased the hazard of acquisition, both during treatment and during (at least) the 3 days after their discontinuation.

We previously demonstrated that ARE epidemiology in our hospital closely mimics the epidemiology of VRE in American hospitals (4, 10). Moreover, vancomycin is the only difference in antibiotic susceptibility between ARE and VRE. Since only one patient in our study received oral vancomycin (during a total of 5 days), we consider our findings on the effects of probiotics and antibiotics generalizable to settings where VRE, instead of ARE, is endemic.

The multispecies probiotic powder that was used in the present study was selected for its ability to inhibit in vitro growth and biofilm formation of hospital-acquired E. faecium, in combination with its capacity to reach the intestinal tract in active form (14). Moreover, it contained an E. faecium strain to enhance competition with ARE for an intestinal niche. Although this strain is susceptible to ampicillin and, therefore, would not be identified in the selective microbial procedure used in our study, we confirmed with MLVA that all isolated ARE strains were genetically different from the probiotic E. faecium strain. In our setting, where both the intestinal flora and ingested probiotic strains are under high selective pressure of antibiotics, the hypothesized preventive effect could not be demonstrated. In another study, though, successful eradication of VRE colonization was reported for patients randomized to receive L. rhamnosus GG yogurt (n = 14) instead of normal yogurt (n = 13) during 3 weeks (18). However, study size and absence of detailed information on baseline characteristics and antibiotic use call for prudence in interpretation.

In general, probiotics are considered harmless food supplements, and the probiotic mixture evaluated in this study had been used with healthy volunteers without any side effects (14). Despite the recently reported adverse events of probiotic administration via a nasojejunal tube in patients with severe acute pancreatitis, there was no evidence that probiotic use, given orally in addition to a normal diet, was unsafe in our population of patients who were not critically ill (2).

Up until now, few studies have evaluated the usefulness of probiotics as a measure to control nosocomial spread of multiresistant pathogens. The strengths of our study include its size and the detailed microbiological analysis with extensive genotyping. However, our study also suffers from some limitations. Although all patients with a length of stay of >48 h were offered probiotics on a daily base, some refused. This might be related to the fact that intestinal colonization is asymptomatic and that, therefore, the potential benefits of probiotics remained intangible for patients. Nevertheless, with 92 acquisitions we still would have been able to demonstrate the effect at which we originally aimed. Moreover, it is unlikely that an increased sample size would have led to a positive effect of probiotics, as the estimate of the hazard points in the opposite direction. Yet, unwillingness to take probiotics, although not related to specific patient groups, introduces the risk of selection bias. This was reflected by some differences in baseline characteristics, for which we adjusted in multivariate analysis. Although we cannot rule out residual confounding, we expect it to have minimal impact, as we adjusted for most important risk factors (4, 9, 24). Furthermore, since cultures were taken twice per week, the exact moment of ARE acquisition was not known and was therefore assumed to occur exactly between the last negative and first positive culture. Sensitivity analyses of the timing of acquisition, however, did not change conclusions. Finally, microbiological culture techniques are imperfect in detecting colonization immediately after acquisition, since time is needed to reach the detection limit of culture methods. This delay between actual acquisition and culture positivity will be important when estimating the effects of time-dependent risk factors. We, therefore, performed time-dependent sensitivity analyses to take this uncertainty about the exact time point of acquisition into account.

In conclusion, the results of this study indicate that there is no role for probiotics in the prevention of colonization with multiresistant enterococci, such as ARE and VRE, in nosocomial settings where they are endemic and selective antibiotic pressure is high.