On the consistency of the continual reassessment method with multiple toxicity constraints

Conventional dose finding methods require the dichotomization of toxicity outcome measures generally collected in an ordinal scale. To improve efficiency and include more information on the gradation of toxicities, a sequential likelihood procedure that accounts for multiple toxicity constraints is proposed to differentiate the tolerance for toxicity of various degrees of severity under a novel class of multiplicative models, and the asymptotic properties of the procedure under certain model misspecification are established.