This meta-analysis was conducted and is presented in adherence with the PRISMA statement recommendations.
We conducted a comprehensive literature search of MEDLINE, Cochrane Controlled Trial Register and EMBASE databases (1966–2011). The search terms included combinations of the following keywords: probiotic, diarrhoea, antibiotic therapy and randomised controlled trials (RCTs) as well as common probiotic species as Lactobacillus, S. boulardii, Saccaromyces, LGG, Bifidobacterium. For example, the PubMed search text was: [(probiotics OR bifidobacterium OR saccharomyces boulardii OR lactobacillus) AND (antibiotics AND diarrhoea)] AND (Clinical Trial[ptyp] AND English[lang]). A manual search of the references listed by studies retrieved from the online databases and from previously published systematic reviews was also performed to identify additional studies of interest. The search was limited a priori to studies that were double-blinded, placebo-controlled, parallel group RCTs published as a full text papers in English.
Selection of Studies
Details of inclusion criteria and definition of outcome variables assessed and extracted are outlined in Table 1. We included studies with concurrent administration of probiotics and antibiotics where the primary or secondary aim was to study the incidence of diarrhoea during or after antibiotic administration in a probiotic and placebo group. Specific inclusion criteria were the following: (i) description of diarrhoea as a definite outcome, (ii) report of the number of subjects experiencing diarrhoea in the therapeutic or placebo group at the end or during the antibiotic treatment, and (iii) no use of other antidiarrhoeal medications as part of the therapeutic regimen during antibiotic administration (Table 1). The definition for diarrhoea was left to the individual studies.
Two investigators (FC, EV) extracted data. Any discrepancies regarding individual study inclusion, data extraction and interpretation were resolved by consensus prior to the final analysis. Details of the extracted data are presented in Table 2. Study variables were grouped in the following categories: study design, demographics, antibiotic(s) used, treatment duration, diarrhoea incidence, probiotic(s) used and risk of bias. Risk of bias was assessed according to the recommendations of the Cochrane Group.
To avoid the inclusion of duplicated data that may lead to an overestimation of placebo effects in final analysis, retrieved studies were carefully appraised and examined by comparison of geographical locations, author names and period of study, as discussed in the Cochrane Handbook for Systematic Reviews of Interventions. The six domains assessed for bias are listed in Table 1 and the criteria used to judge a study as 'high' or 'low' risk were those outlined in Chapter 8 of the Cochrane Handbook. A study was given an overall rating of 'high risk' if one or more of the six domains was assessed to be high risk, a rating of 'low risk' if all domains were assessed to be 'low risk' and 'unclear risk', if all domains were either 'low risk' and 'unclear risk'.
Meta-analysis and Subgroup Analyses
All statistical analyses were performed using JMP v.7, RevMan v. 5.1 and Comprehensive Meta-Analysis 2.2. First, the RR and 95% CI of diarrhoea on probiotic treatment vs. placebo treatment was calculated for each study based on the per protocol (PP) sample size, and we performed a sensitivity analysis based on primary authors' defined ITT sample. Statistical heterogeneity across the various studies was then tested with the use of Q-statistic. A P value <0.10 indicated a significant statistical heterogeneity across studies, thus a random effects model was used. In addition to the main (overall) analysis, multiple subgroup analyses were performed, yielding a pooled RR including and excluding studies of H. pylori treatment, paediatric populations and grouping studies by risk of bias, relative duration of antibiotics and probiotics and by probiotic species used. The overall number needed to treat (NNT) and its 95% CI were also computed using pooled raw events for the overall analysis and for the H. pylori subgroup analysis.
Potential association between the response to probiotic administration and study-related variables was studied by meta-regression using unrestricted maximum likelihood models. The log odds ratio of diarrhoea during antibiotic therapy was the dependent variable in the meta-regression model. Independent study-level variables were: (i) the risk of bias (using a scale of low, uncertain and high risk of bias), (ii) individual study mean patient age (iii) probiotic type and (iv) incidence of diarrhoea in the placebo group.
Assessment of Publication Bias
Publication bias was determined by the funnel plot of Beggs & Mazumbar's rank correlation test and by Egger's intercept test using a random effects models.