Estimation from a Two-Stage Biomarker Study Allowing Early Termination for Futility

• Activity Number: 359
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Shanshan Zhao*+ and Ziding Feng
• Companies: Fred Hutchinson Cancer Research Center and Fred Hutchinson Cancer Research Center
• Keywords: diagnosis biomarker ; two-stage design ; conditional estimation ; ROC curve

Abstract

Biomarkers are widely used for disease screening. However, many candidate biomarkers identified in a biomarker discovery study will not perform well in a validation study. Thus, a two-stage biomarker validation study allowing early termination for futility is preferred to save specimens and funding. In this study, we consider how to estimate biomarker performance conditional on reaching full study enrollment. Previous study showed that the estimates of biomarker performance conditional on passing the first stage is more efficient than the estimates based on stage 2 data only, when there is only one biomarker of interest. This study extends the conditional estimation to allow a panel of biomarkers. We examine the performance of conditional estimates for several ROC curve related measures (e.g., sensitivity, specificity) under different combination rules, and show the efficiency gain is a function of the number of biomarkers of interest.