Meta-analysis: Probiotics in Antibiotic-Associated Diarrhoea

E. J. Videlock; F. Cremonini


Aliment Pharmacol Ther. 2012;35(12):1355-1369. 

In This Article


Eligible Studies

The literature search identified 203 articles that met the search criteria. Figure 1 shows the flow of this meta-analysis. Sixteen studies were excluded after full text review.[16–31] After the screening process, 34 parallel-design, double-blinded, placebo-controlled, randomised trials published in English were included.[32–65] The studies were published between 1979 and 2011.

Figure 1.

Meta-analysis flow.

Patients' characteristics, study design and interventions of the trials included are summarised in Table 2. Ten trials were conducted in paediatric populations, the remainder in adult patients. Mean age in the paediatric and adult trials was 5 and 52 years respectively. The probiotics used included lactobacilli, bifidobacteria, enterococci, streptococci, the yeast S. boulardii and combinations of these species. The duration of antibiotic and probiotic supplementation was variable but ranged from a minimum of 3 days to several weeks.

Fourteen studies were assessed to have a low risk of bias. In 10 studies, the risk was unclear and 10 studies were assessed as high risk. The justification for these classifications is outlined in Table 3.

The definition of the event of diarrhoea varied across trials, with most studies considering as diarrhoea a daily bowel frequency of at least 3 bowel movements daily, while some also used stool consistency. When combined, the studies included enrolled 4138 patients, of which 2154 received probiotic supplementation.

There was significant heterogeneity across studies (P = 0.0005), thus a random effects model was thus chosen to calculate pooled estimates of treatment effects.

Effect of Probiotics on Antibiotic-associated Diarrhoea

When including all studies of adult and paediatric populations, probiotics resulted in a pooled RR of AAD of 0.53 (95% CI 0.44–0.63) compared to placebo (Figure 2). This risk reduction corresponded to an average NNT of 8 (95% CI 7–11).

Figure 2.

Forest plot showing random effects (RE) relative risk (RR) estimates with the corresponding 95% CI for the development of AAD (probiotic vs. placebo arms) in all studies.

The pooled RR among the 10 paediatric studies was 0.48 (95% CI 0.35–0.65) similar to the pooled RR of 0.53 (95% CI 0.43–0.66) from the 24 adult studies (Figures 3a and b).

Figure 3.

Forest plots showing random effects (RE) relative risk (RR) estimates with the corresponding 95% CI for the development of AAD (probiotic vs. placebo arms) in adult (Figure 3a) and paediatric studies (Figure 3b).

The pooled RR from 6 studies during H. pylori treatment was 0.37 (95% CI 0.20–0.69) corresponding to a NNT of 5 (95% CI 4–10), while the pooled RR excluding studies during H. pylori treatment was 0.56 (95% CI 0.46–0.67), corresponding to a NNT of 9 (95% CI 7–12).

An analyses based on the ITT sample size, where provided by authors, yielded similar RRs to the analysis based on the PP sample size (data available upon request).


No effect on the outcome was detected by risk of bias (intercept −0.75, slope 0.10, P = 0.12) mean patient age (intercept −0.71, slope 0.0046, P = 0.06), duration of antibiotics (intercept −0.84, slope 0.040, P = 0.26), or incidence of diarrhoea in the placebo group (intercept −0.67, slope 0.003, P = 0.25).

Subgroup Analyses

Table 4 summarises the pooled RRs from subgroup analyses of trials according to probiotic used, indication, population, setting, relative treatment duration of antibiotics and probiotics and risk of bias. Lactobacilli were the most used probiotic species, with 24 studies, including 8 studies testing the effect of Lactobacillus GG. Saccharomyces boulardii was tested in 7 studies and 10 studies used bifidobacteria (including, as part of a combination). The pooled estimates of the preventative effects on AAD appeared consistent across the probiotic species administered. The pooled RR for studies with low risk of bias was obtained, combining 14 studies assessed as low risk yielding a RR of 0.50 (95% CI 0.37–0.67), which was similar to the RR of 0.45 (95% CI 0.33–0.61) for the 10 studies with unclear risk of bias.

Publication Bias

The funnel plot using random effects (Figure 4), and Egger's intercept test (intercept −1.41, P = 0.001) suggested a lack of publications reporting negative results, thus supporting the presence of a negative publication bias.

Figure 4.

Begg's funnel plot SE (effect estimate) versus the effect estimate for each study. Each study is represented by a single dot. Log relative risk (RR) from the random-effect model was used as the effect estimate. The overall effect estimate is indicated by the vertical line. The two diagonal lines represent 95% CI.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.