Reducing the number of patients required for a clinical trial is important for shortening development time. Phase II cancer trials assess the tumour-shrinking effect of a novel compound through a binary end-point formed from the percentage change in total lesion diameter. We compare single-arm two-stage designs which use the binary end-point to those which directly use the continuous end-point.
Using the continuous end-point results in lower expected and maximum sample sizes. For larger trials the reduction is around 37%. This assumes that the dichotomisation point of the continuous end-point is chosen to give the best sample size, with the trial design using the binary end-point performing even worse otherwise. We consider a previous trial designed using a Simon two-stage design and show that if the continuous end-point had been used, the expected and maximum sample sizes of the trial would be reduced by around 50%.
Using the continuous end-point in a two-stage cancer trial results in large sample size reductions. The methods discussed in this paper work best when the number of complete responses is low, as is true in several types of cancer. We discuss what could be done if this is not the case.