Michael K. Smith(1), Scott Marshall(2), John Huggins(2).
(1) Biostatistics and Reporting, PGRD Sandwich; (2) Clinical Sciences, PGRD Sandwich.
Objectives: We wish to optimise the design of a dose-response trial and obtain good estimates of relative potency when information is available on a previous compound within the same drug class. Using Bayesian methodology this prior information can be used quantitatively in the design and analysis of the study. This allows us to bias randomisation towards the new compound and make accurate inferences about the magnitude of the relative potency with a small overall sample size.
Methods: An Emax model was used to describe the dose-response relationship of an existing drug. The estimates from this model were used to provide an informative prior which was used to optimise the design and analysis of a new study to establish the dose-response and relative potency of a related compound from the same drug class. The assumption is made that data from the previous trials and the new study are exchangeable. This can be assured by making the inclusion / exclusion criteria as similar as possible, but departures from this assumption can also be allowed for by increasing the sample size.
Results and conclusions: Simulation results show that a relatively modest overall sample size can yield very informative results about the magnitude of the relative potency using this approach. Biasing the randomisation from the existing drug towards the new compound (in the ratio 1:4) gives sufficient information about the new drug to be able to make decisions about the magnitude of the relative potency with very good precision. Type I and type II errors using this approach are very low. The bayesian approach also allows probabilistic statements about the magnitude of the effect – very useful in decision making. Departures from our assumptions increase the type I and type II errors, but these can be mitigated slightly by increasing the overall sample size in order to allow the study data to influence the posterior estimates. This approach has the potential to allow very efficient dose-response studies where prior information on a previous drug is available and is considered exchangeable with information from a new study.
Reference: PAGE 13 () Abstr 480 [www.page-meeting.org/?abstract=480]
Poster: poster