Integrating Preclinical Insights for Adaptive Dose Escalation in Phase I Oncology Trials

Leveraging preclinical prior information has the potential to enhance the efficiency of Phase I oncology trials if used appropriately. In this paper, a comparison of the meta-analytic predictive (MAP) prior approach and the power prior approach is undertaken in application to a setting using the Bayesian Logistic Regression Model (BLRM). A novel methodology to determine the parameters of the two approaches based on external data is introduced. Moreover, the escalation with overdose control criterion, commonly used in conjunction with the BLRM, is extended via an additional criterion, which allows for less conservative escalation. It is found that the inclusion of animal data is recommended to be done using the flexible MAP prior, with a justified specification of prior exchangeability. However, the benefit over not including animal data must be weighed against the potential losses.