Systematic literature review and search strategy

This systematic literature review was carried out following the guidelines set forth by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA), ^{Reference Liberati, Altman and Tetzlaff19} and the Bayesian Analysis Reporting Guidelines (BARG). ^{Reference Kruschke20} The review was registered with the International Prospective Register of Systematic Reviews (PROSPERO) on March 21, 2025 (CRD420251016925), with its protocol included. Approval from the Institutional Review Board was not necessary for this study and no patient informed consent was required either.

Our search strategy was developed in March 2025 with the assistance of a health sciences librarian. We explored Embase, PubMed, Scopus, Web of Science, Cochrane, and CINAHL. The literature search included manuscripts published from the inception of each database up to May 21, 2025. The detailed search strategy is available in the Supplementary Table 1.

Studies were included if they evaluated the incidence of CDI in patients who received oral vancomycin prophylaxis compared to those who did not. CDI was defined as detection of a positive nucleic acid amplification testing (NAAT) and/or reflex to toxin enzyme-linked immunosorbent assay for C. difficile, ideally in a two-step testing, in patients with new onset diarrhea (24 h or less). ^{Reference Turner, Grambow and Polage9} Primary prophylaxis was defined as the use of a drug to prevent disease in at-risk individuals with no prior history, while secondary prophylaxis aimed to prevent the recurrence of a disease in patients who have already experienced it. The intervention was defined as providing oral Vancomycin 125 mg twice daily, starting on the day of inpatient admission and continued until discharge. We excluded comments or reviews, studies that focused on secondary CDI, studies without a comparable control group, studies where controls received a different intervention (other than standard of care), pilot studies, and studies that used the same hospital population of an already included study.

We identified a total of 164 studies from our literature search. After removing duplicates, 132 studies were screened using title and abstract by two independent reviewers (WT and EV). From this initial review, 23 full-text studies were independently assessed (WT and EV). Discrepancies were resolved through discussion or consultation with a third reviewer (EM). Ultimately, six studies met inclusion criteria and were included in this systematic review.

Data abstraction form and quality assessment

Two independent reviewers (WT, EM) abstracted data for each article using a standardized abstraction form. We recorded data regarding authors, publication year, study period, design, population selection, setting, inclusion and exclusion criteria of each study, studied groups, prophylaxis regimen, CDI diagnostic criteria, number of participants, total number of participants receiving the intervention, number of CDI per group, and secondary outcomes analyzed. Our primary outcome was the incidence of CDI in patients who received oral vancomycin prophylaxis and those who did not. Our secondary outcomes were bloodstream infections, vancomycin-resistant Enterococcus cultures, GVHD, and length of hospital stay.

We used the Downs and Black scale to assess quality of studies, which is specifically designed for randomized and non-randomized studies. ^{Reference Downs and Black21} All questions of the original published scale were answered for each reviewed paper and the total score calculated. We adapted question 27 of the Downs and Black scale replacing the multiple-choices options with a yes/no answer. The maximum possible score was 28. Downs and Black score ranges were given corresponding quality levels: excellent (26–28); good (20–25); fair (15–19); and poor (≤14). The reviewers (WT, EM, EV) performed the quality analysis independently and inconsistencies were resolved by discussion.

Statistical analysis

Effect sizes and their standard errors were calculated from the sample sizes and reported cases for each study using the metafor package in R (version 4.8-0). The meta-analysis was conducted with the bayesmeta package version 3.4, employing Bayesian random-effects models. ^{Reference Röver22} The Bayesian approach is especially useful to account for between-study heterogeneity. Additionally, unlike frequentist methods, it does not require a larger number of studies and directly provides credible intervals for the pooled mean effect. Half-normal prior was applied for both the overall effect and heterogeneity parameters. Heterogeneity is calculated using tau (τ), an estimate of the amount of true variability between the effect sizes of the included studies in a random-effects meta-analysis model. Results were summarized by reporting the posterior mean as an odds ratio (OR), along with 95% credible intervals (95% CrI) for the overall effect size and the tau statistic to quantify heterogeneity among the studies. And the prediction describes the consistency of data and model by comparing the actual data to data sets predicted via the posterior distribution. ^{Reference Röver22}

The potential for publication bias was assessed using funnel plot, Egger’s test, and the Robust Bayesian Meta-Analyses (RoBMA) package version 3.4.0. ^{Reference Maier, Bartoš and Wagenmakers23} RoBMA results use Bayes factors (BF), a continuous measure of evidence in favor of the presence or absence of effect, heterogeneity, and publication bias. Bayes factor values above 10 indicate very strong evidence, from 3 to 10 moderate evidence, and from 1 to 3 weak evidence and <1 no evidence. ^{Reference Wagenmakers and Lee24} Sensitivity analyses assessed the robustness of findings by excluding studies with unpublished full-text manuscripts. All analyses were performed using R version 2024.12.0 + 467.