A Bayesian zero-inflated binomial regression and its application in dose-finding study

ABSTRACT

In early phase clinical trial, finding maximum-tolerated dose (MTD) is a very important goal. Many researches show that finding a correct MTD can improve drug efficacy and safety significantly. Usually, dose-finding trials start from very low doses, so in many cases, more than 50% patients or cohorts do not have dose-limiting toxicity (DLT), but DLT may occur suddenly and increase fast along with just two or three doses. Although some fantastic models were built to find MTD, little consideration was given to those ‘0 DLTs’ and the ‘jump’ of DLTs. In this paper, we developed a Bayesian zero-inflated binomial regression for dose-finding study, which analyses dose-finding data from two aspects: 1) observation of only zeros, 2) number of DLTs based on binomial distribution, so it can help us analyse if the cohorts without DLT have potential possibility to have DLT and fit the ‘jump’ of DLTs.

KEYWORDS:

• Dose-limiting toxicity
• maximum-tolerated dose
• metropolis algorithm
• zero-inflated binomial regression

Acknowledgments

This work was financially supported by the grant research fund of the Faculty of Science, King Mongkut’s Institute of Technology Ladkrabang. The authors thank the early phase development team in Celgene Corporation for their help on statistical techniques.

Additional information

Funding

This work was supported by the Faculty of Science, King Mongkut’s Institute of Technology Ladkrabang.