The Role of the Cutoffs in Developing the Genomic Classifier for Clinical Utility Assessment

• Activity Number: 112
• Type: Topic Contributed
• Date/Time: Monday, August 1, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Samir Lababidi*+ and Sue-Jane Wang
• Companies: U.S. Food and Drug Administration and U.S. Food and Drug Administration/CDER
• Keywords: Gene Expression ; Prediction performance ; cutoff ; MAQC-II ; genomic classifier

Abstract

With increasingly sophisticated profiling techniques, physicians and pharmaceutical investigators have gradually learned to appreciate the complexity of disease in individual patients, and thus the unique response of certain patients to specific treatments. Genomic classifiers can be incorporated into randomized controlled trials to evaluate if a new treatment is effective or more effective relative to its comparator in patients who are classified positive. However, the choice of cutoffs of classifiers plays a critical role in the prediction performance of these genomic models. In this paper, we assess the prediction performance of three genomic models (perceived to be the best, the reasonable, and the worst) developed by data analysis teams in the recent MAQC-II project by evaluating the clinical utility of the genomic classifiers beyond the clinical covariates at pre-specified cutoffs of the genomic classifier. We show that the effective use of clinical covariates critically depends on the particular cutoff for which the genomic classifier is constructed. This highlights the challenges in developing microarray-based gene expression models in the area of personalized medicine.