After reading your links I think it is more clear that "optimal dose" does not mean optimal dose for the patient or total population. It is optimal dose such th

After reading your links I think it is more clear that "optimal dose" does not mean optimal dose for the patient or total population. It is optimal dose such that a single phase III trial would detect a measurable effect. Once its approved already at a set dose, the "ideal dose" for future use is probably not the same as the "optimal dose" that maximized probability of approval. "We propose a Bayesian utility approach (BUART) to randomized Phase II clinical trials which uses a first-order bivariate normal dynamic linear model for efficacy and safety in order to determine the optimal dose and study population in a subsequent Phase III clinical trial. "