Bayesean Analysis as a Predictor of outcome rate.

Abstract

Power calculations in clinical trials depend on an estimation of a change in effect size, and this depends on an estimation of the rate of the (measured) event in the population. It may be that using a Baysian approach authors would be able to better estimate the rate of events, which would help with estimation of power. We used a data-set on early intervention described elsewhere (Gale, in press). We identified three where a transition rate and was available and we could estimate a post screen probability of transition to psychosis. Using Bayes' theorem, we estimated the screen positive (using the CAARMS) transition rate in the first trial at 11.1%. Screen positive participants, at six month follow-up, had a transition rate of 10.1%. (Yung 2006). The same authors reported two year follow-up: we estimated the transition rate at 17.2% and the reported rate was 16.0%. Another author used the PRIME questionnaire: we estimated the rate of transition at 17.2% and the reported six month follow-up rate was 11.9%. These results are preliminary. However, this may be a more appropriate method than using earlier trials, particularly if the rate of events is changing (McGorry, 2002)