When It’s Too Trying for Trials: A Commentary on a Different Approach for Testing a Probiotic, Saccharomyces boulardii, to Prevent Healthcare-associated Clostridioides difficile Infections

(See the Major Article by Wombwell et al on pages e2512–8.)

Globally, Clostridioides difficile infections (CDI) continue to plague healthcare facilities and long-term care facilities with outbreaks of disease with an estimated number of cases per year of over 495 000/year, costing 5–6 billion US dollars for healthcare [1, 2]. Although major strides in primary prevention infection control practices have reduced the incidence of CDI, outbreaks continue to occur, calling for innovative measures to prevent CDI. One addition to standard infection control protocols (typically including enhanced handwashing, terminal room disinfection, and antibiotic stewardship programs) has been the use of probiotics for high-risk patients [3]. Randomized controlled trials have proved to be problematic when testing new infection control programs due to large required study sizes necessitated by typical low (1%–5%) prevalence of C. difficile in nonoutbreak situations. Thus, quasi-experimental study designs have been used to test new infection control protocols using either a historical (“pre”) period compared to a test period when the new infection control practice is implemented or conducting a prospective study using a similar ward at the same hospital or a nearby sister hospital comparing CDI rates during the intervention period. In the retrospective cohort study by Wombwell and colleagues in this issue of Clinical Infectious Diseases [4], they report the effect of placing 1 probiotic (Saccharomyces boulardii) on their hospital formulary on the rate of CDI. A new prescription of an antibiotic considered at higher risk of causing CDI for an inpatient at their hospital caused an electronic flag to be triggered, which offered an option to also start the probiotic. This change in formulary started in December 2015, and hospital-onset cases of CDI were counted from January 2016 through March 2017. Then patients given S. boulardii were compared to patients not receiving the probiotic for healthcare-onset CDI and other risk factors to determine the effect of the probiotic. The authors failed to provide the strain or manufacturer of the probiotic, however.

Although not a randomized, controlled trial, this quasi-experimental study design allows perhaps a more realistic evaluation of a new intervention implemented in a setting with diverse challenges, including the lack of control of who receives the new treatment (the choice was made by the physician after receiving an electronic flag) to differences in compliance to different durations probiotic may have been taken. Outside of the randomized trial setting, where specific protocols are followed and only eligible enrolled patients receive the new treatment and then closely followed, this design allows the somewhat open choice to begin the probiotic and does not follow each individual patient, but relies on hospital-wide rates of CDI to assess the effectiveness of the probiotic. Another advantage of this type of design is that it allows an in situ evaluation over a large population, and this study included 8763 patients. The availability of hospital-wide data on antibiotic and medication use and other risk factor data are another advantage of this type of study.

One disadvantage of this type of study design is that it does not account for other changes in the hospital population during the time of the study that may impact CDI rates not related to the probiotic itself, such as changes in antibiotic use, changes in infection control practices, changes in medication usage or changes in patient populations. Wombwell et al addressed this issue by tracking changes in infection control practices (there were none during the study period) and by adjusting for the type of antibiotics received, use of proton pump inhibitors (PPIs), and so forth using a logistic regression model.

Another disadvantage of this study design arises because the assignment of which patient receives the probiotic and which does not is not randomized; thus, selection bias might result in the inclusion of higher or lower risk patients in one group (those receiving the probiotic) compared to the other group (nonprobiotic group). Of 8763 eligible patients, 63% received the probiotic. We can assume there were other reasons besides their exclusion criteria that 3276 patients did not receive the probiotic, but these are unknown. Wombwell et al did find that the group receiving S. boulardii was different than the group who did not receive the probiotic (the probiotic group was older, received more PPIs, quinolones, and cephalosporins, and had fewer intensive care unit admissions). As the choice of which patient received the probiotic was under their physician’s discretion, the choice of initiating probiotic treatment may be influenced by the other factors, such as whether the patient was considered at higher risk (older age, history of prior CDI episodes, immunosuppression, etc.) of CDI. The authors did address this limitation by collecting data on patients relating to CDI for each group and using a logistic model with a propensity score to adjust for most of these factors.

Wombwell et al concluded that S. boulardii administered to hospitalized patients prescribed high-risk antibiotics was associated with a reduced risk of hospital-onset CDI. Although this type of study design does not lend itself toward proving causation, it may show an association with a protective effect of adding a specific probiotic to a hospital formulary. In addition, other improvements may be noted during the observation period, as this study found a stronger protective effect if the probiotic was given within 24 hours of the antibiotic initiation.

Future studies might benefit from asking healthcare providers the reasons why they chose to begin the probiotic or not, as this might help to define bias arising from the lack of randomization. As this is a hospital-wide study, future studies might also benefit from collecting more data on important risk factors for CDI (such as a history of a prior CDI episode, prior hospitalizations, etc.) that could be adjusted for in a multivariate model.

The study design used by Wombwell et al allows clinical studies to explore new treatments in the environment for which they would be used in when limitations imposed by randomized clinical trials pose too much of a barrier, but caution should be exercised when attributing causation solely to the probiotic intervention.

Note

Potential conflicts of interest. The author is on the Microbiome Board of Advisors for Biocodex (France) and a paid lecturer. The author does not own stock or equity in the company. The author has submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.

References