Affiliations: [a] Department of Biostatistics & Applied Mathematics, The University of Texas, M.D. Anderson Cancer Center, Houston, TX, USA | [b] Department of Biostatistics, School of Public Health, University of North Texas Health Science Center, Forth Worth, TX, USA | [c] Department of Leukemia, The University of Texas, M.D. Anderson Cancer Center, Houston, TX, USA | Biostatistics Program, Fred Hutchinson Cancer Research Center, Seattle, WA 98109, USA
Correspondence:
[*]
Corresponding author: Donald A. Berry, Ph.D., Department of Biostatistics & Applied Mathematics, M.D. Anderson Cancer Center, Unit 447, 1515 Holcombe Boulevard, Houston, TX 77030, USA. Tel.: +1 713 794 4141; Fax: +1 713 563 4242; E-mail: [email protected].
Abstract: An important aspect of modern cancer research is identifying molecular and genetic markers that predict an individual's risk of recurrence and response to treatment. Increasingly, a large number of biomarkers are included in studies, even though at most a small proportion of them will be prognostic. This article considers the selection and validation of prognostic biomarkers, and proposes a method for building and validating a prognostic index based on potentially many markers. We use simulation to show that our proposed method controls false-positive rate while retaining reasonable power. The prognostic index gets its power from combining prognostic biomarkers. It can be very powerful even when there are few markers that are predictive.
Keywords: Biomarkers, data splitting, false-positive rate, prognostic power, prognostic index
