Abstract:
|
In many phase I clinical trials, it is desirable to use a dose-selection algorithm capable of incorporating toxicity and efficacy data. For example, in the trials for rare disease such as hemophilia B where identifying and recruiting patients is a challenge, patient response data should be utilized whenever possible. A bivariate continual reassessment method (bCRM) was proposed by Braun in 2002. Bearing the characteristics of adaptive design, bCRM not only addresses limitations in the historically simplistic phase I maximum tolerated dose (3+3) design, it also incorporates efficacy information when identifying dose escalation. The original bCRM algorithm did not consider the case when there are missing data. In practice, it is very likely that either toxicity or efficacy outcome is not collected and therefore missing. For example, some trial might have a big window between collection of toxicity data and efficacy data. In this scenario, the likelihood of missing either the efficacy or toxicity data at a specific visit is very high. We plan to investigate several imputation rules for different scenarios and propose the most robust imputation rule and most conservative imputation rul
|