Optimal Dose de-escalation 
Trial Designs for novel contraceptives in women
Christoph Gerlinger - Senior Director - Bayer Schering Pharma AG

Dose finding for classical hormonal contraceptives for women is usually done by investigating the surrogate endpoint inhibition of ovulation. For novel compounds such an approach is not feasible because they do not necessarily inhibit ovulation and no other surrogate endpoint for pregnancy is currently available. The only way to assess the efficacy of such a product is the direct measurement of the contraceptive efficacy, e.g. by the Pearl Index. However, a classical parallel group dose response trial investing several doses including at least one ineffective dose is not possible due to ethical considerations. Therefore, an alternative trial design to determine the lowest effective dose of a new compound that minimizes the number of unwanted pregnancies occurring during the dose finding trial is needed.

  We investigated six dose escalation designs used to find the maximal tolerated dose in cancer trials to our problem of determining the minimal dose that is at least 99% effective (least effective dose LED) in preventing pregnancies over 1 year. We elucidated the statistical properties of these designs by a simulation study.

  The biased coin and the r-in-a-row designs proved to be not feasible for our problem because they require a sequential treatment, i.e., a cohort size of 1, which takes far too long in the case of contraceptives. The Bayesian ADEPT method was also not applicable for our problem.

  The most suitable dose de-escalation designs to determine the LED of a new contraceptive that minimizes the number of unwanted pregnancies occurring during the trial were the continual reassessment method and a design derived from the classical 3+3 design in cancer, but with a cohort size of 100 instead of 3. Both dose-finding designs substantially reduced the expected number of pregnancies to less than 4 pregnancies compared to 16.9 in the classical dose-finding design with a similar total sample size. However, this clear advantage comes at the price of a 5-fold increase in trial duration.