Model-Based dose escalation Designs in R with crmPack

Bové, D.S. and Yeung, W.Y. and Palermo, G. and Jaki, T. (2019) Model-Based dose escalation Designs in R with crmPack. Journal of Statistical Software, 89 (10). ISSN 1548-7660

Abstract

Model-based dose escalation designs have gained increasing interest due to the need for more efficient and informative Phase I trials. The wide-spread implementation of such designs has been hindered by the need for either licensing specialized commercial software or programming the design and simulations from scratch for each project. The R package crmPack provides a simple and unified object-oriented framework for model-based dose escalation designs. This enables the standard use of such designs, while being able to flexibly adapt and extend them. The framework comprises classes and methods for the data structure including the dose grid, statistical models including prior specification, rules for maximum increments, next best dose, and adaptive stopping and cohort sizes. In addition to multiple modified classic continual reassessment method and escalation with overdose control designs with possibly advanced prior specifications (e.g., minimal informative and mixture priors), crmPack currently features dual-endpoint (safety and biomarker) designs and two-part designs. Optional assignment of a small number of patients in each cohort to placebo instead of treatment enables the use in trials outside oncology.