Please click here to obtain permission to reproduce this image.
Click on image to view larger version.
Figure 6 Bayesian analysis of Stroke Prevention Using Oral Thrombin Inhibitor in Atrial Fibrillation (SPORTIF) V trial. Tri-plots showing posterior (thick line) distributions derived from integrating evidence or likelihood (thin line) from SPORTIF V trial and three different priors (dashed line), according to Bayes' theorem. The margin of non-inferiority (M) is indicated by the two vertical dotted lines and is equivalent to a log risk ratio of 0.5 (equivalent to a risk ratio of 1.65). Three priors are used: 1) uninformative prior, intended to add as little as possible to the data (formally expressed as log-RR mean (µ) = 0, standard deviation ( ) = 10; likelihood and posterior distributions are superimposed, hence only one plot); 2) moderately skeptical (i.e., 50% of the distribution is contained within the non-inferiority margin [log-RR µ = 0.250, = 0.374]); and 3) informative, on the basis of prior information derived from SPORTIF III trial (log-RR µ = –0.338, = 0.241). Posterior probability of any effect size can be calculated by computing area under the curve. The probabilities of falling below (<M), within (= M), or above M (>M) are shown on the top right-hand corner of each plot. Probability of non-inferiority is computed as the sum of probability of <M plus = M (e.g., the posterior probability of non-inferiority with uninformative prior [plot # 1] is 0.068 [<M] + 0.736 [= M] = 0.804). Non-inferiority is inferred at a posterior probability of  0.975 (corresponding to a one-sided p  0.025).
|
