Bayesian Models for Clinical Studies
SummaryThe concept of Bayesian statistics, based on the model of random parameters with appropriate a priori distributions, is discussed and applied to the analysis of clinical studies (i. e. treatment comparisons). It is shown that assumptions about the a priori distribution can be eliminated if analysis is restricted to the class of conjugate prior distributions and the number of data is sufficiently high. For treatment comparisons the concept of “preferences” is introduced, i.e. the a posteriori probability for special rankings of the effect parameters. This concept is an alternative to hypothesis testing and error probabilities which is meaningless in Bayesian models. With this concept it is not necessary to formulate the hypotheses before the study or fix sample size or stopping rules in advance. It is also not necessary to restrict the analysis to the test of one or few hypotheses. On the other hand, the physician will not get error probabilities for his statements but “preferences” of the relevant rankings of the treatments.