My recent post about Bayesian analysis was in part prompted by this trial, just published in the Lancet, but I never got back to discussing this trial, being distracted by other things, like actual work. Nearly 3000 adults were randomized to either probiotics or placebo, with the primary outcome being antibiotic associated diarrhoea or C difficile. Previous studies have been generally positive for C diff, not so much for AAD overall.
If you look at the Cochrane review of probiotics for preventing intestinal complications of antibiotics, there is about a 60% reduction in C diff,but probably no major effect on overall antibiotic associated diarrhoea.
In this study there was about a 30% reduction in C diff: that has been interpreted as being entirely negative, as the results of this trial are not statistically significant. In fact the confidence intervals of this study overlap the CI of the prior meta-analysis, but the new trial had a far lower incidence of C diff than the previous trials, 1.2% in the controls reduced to 0.8% in the group receiving probiotics.
It is self-evident that with a control frequency of 1.2% it will be almost impossible to prove a reduction in adverse outcomes. Some of the previous studies had control group incidence as high as 25% of C diff.
My interpretation as a non-involved clinician is that it seems very likely that probiotics prevent C diff in at-risk adults, but if the underlying frequency is very low the NNT will be very high (which is hardly surprising). If the control frequency is very low then the cost of the probiotics may become an issue, but if I were in that situation (which I might be in a year or two or thirty) the rather modest cost, the appalling personal cost which C diff sometimes causes, and the safety profile, would probably push me to eat my active culture yogourt every day.
Back to Bayesian analysis; if we take all the previous data in order to determine the significance of what we have from the new trial, it appears to be very close to what you would expect for a trial with a low background incidence of C diff. If you analyze it completely on its own it looks very unimpressive.