Designs Accounting for Potential Missing Trials in Multiple-objective Dose-response Studies

• Activity Number: 407
• Type: Contributed
• Date/Time: Thursday, August 12, 2004 : 8:30 AM to 10:20 AM
• Sponsor: General Methodology
• Author(s): InYoung Baek*+ and Wei Zhu
• Companies: SUNY, Stony Brook and SUNY, Stony Brook
• Keywords: Bayesian optimal design ; dose-response ; missing data ; multiple-objective

Abstract

The toxicity and efficiency of a therapeutic drug are determined before it is approved for general use. Quantal dose-response experiments are routinely conducted to examine the response rates at various dose levels of interest. However, due to side effect or lack of efficacy or other problems, responses are not always observed at all selected dose levels. Furthermore, the missing data patterns are often dose-dependent. Missing trials would invariably lead to reduced precision in parameter estimation and power in hypothesis testing. Depending on different missing data mechanisms, one can protect against potential missing trials by incorporating prior information in the design stage of the experiment. We examine missing data patterns from previous dose-response studies and subsequently design future multiple-objective dose-response studies accounting for potential missing observations in a Bayesian approach. The performances of designs with and without accounting for potential missing trials are compared in terms of the precision of related parameter estimation and the power of related tests.