5.2 Exemple 2

Dans cet exemple [2], plusieurs types d’apriori ont été utilisés donnant chacun un résultat différent

Abbreviations: aRR, adjusted risk ratio; P-TB, posterior probability of treatment benefit (risk ratio <1.0); RR, risk ratio

In Bayesian analyses, the probability of treatment effect (posterior probability) is estimated after the trial and incorporates the prior probability estimated from the best data from previous studies (clinical trials or pilot trials). Judgment of the prior probability may vary and be neutral, enthusiastic, or skeptical. Therefore, analyses were performed using 3 different prior probabilities: (1) a neutral prior, assuming no treatment effect (RR, 1.0); (2) an enthusiastic prior, assuming a 28% reduction in the risk of death or disability as in the earlier Neonatal Research Network trial (RR, 0.72); and (3) a skeptical prior, assuming a 10% increase in the risk of death or disability (RR, 1.10). Whether neutral, enthusiastic, or skeptical, assessments of prior probability involve uncertainty about the minimum and maximum likely treatment effects. To reflect this uncertainty in each analysis, a probability distribution for the treatment effect with the 95% credible intervals that ranged from half to twice the assumed RR (SD, 0.35 in the log scale) was used. For example, the probability distribution for the neutral prior was centered at an RR of 1.0 (mean of 0 in the log scale) with a 50% prior probability of a better outcome, a 50% prior probability of a worse outcome, and a 95% credible interval for the RR of 0.5 to 2.0