The bivariate continual reassessment method. extending the CRM to phase I trials of two competing outcomes

Abstract

Traditional phase I studies find the therapeutic dose of an investigational drug based solely on toxicity and without regard to the drug's efficacy. We propose extending the continual reassessment method (CRM) to a bivariate trial design in which the maximum tolerated dose is based jointly on both toxicity and disease progression. We call our study design bCRM, for bivariate CRM, which we apply to a study of bone marrow patients seeking to find the optimal time after bone marrow transplant at which to taper immune suppression and eventually begin donor leukocyte infusions. We demonstrate through simulation that bCRM has excellent operating characteristics and compare the performance of bCRM in a bone marrow transplantation study to designs proposed by previous investigators. We also attempt to provide some direction to future investigators with regard to plausible models and distributions to use with a bivariate study design.