The goal of this application is to enable cancer researchers to develop and evaluate predictive cancer biomarkers more quickly and easily. We will do this by producing biostatistical methods aimed at eliminating two common frustrations in cancer biomarker research: (1) promising biomarkers must be abandoned because the assay cannot be implemented in its current form in a clinical trial, and (2) modifications in a biomarker assay which may reduce cost or improve marker performance cannot be explored because of the high additional cost of such studies. It is frequently required or desirable to modify some aspects of an assay before moving to the clinical validation step. The assay may be modified for many reasons. We will call all of these changes assay modifications, and refer to the result as a modified assay. Assay modifications can stymy biomarker development because, if the modification is not trivial, then any previous associations between the biomarker and clinical outcomes must be re-investigated. This application will develop methods that allow the clinical performance of the modified assay to be estimated using data from a reproducibility study. The reproducibility study will compare the original assay and the modified assay on a set of patient samples. Importantly, these samples need not have clinical outcomes associated with them. In sum, if this application is successful, investigators will no longer need to feel locked into the assay as it was originally developed, or be forced to abandon an assay because modifications are required. More predictive biomarker assays will therefore be developed, and the benefits they provide to public health, patients and physicians will be realized.
The specific aims are therefore: 1. Develop sets of estimators for the change in biomarker performance associated with modifying an assay. 2. Using a combination of real data resampling studies, simulations and mathematics, evaluate how well different reproducibility metrics capture changes in biomarker performance. 3. Using the best metrics identified in Specific Aim 2, develop optimal statistical design and sample size methods for biomarker reproducibility studies.
Biotechnological developments are making it possible to precisely measure cancer patients to determine which cancer therapy will be most effective. Clinical tools used to predict which cancer treatment will benefit a patient the most are called 'predictive biomarkers.' This project will produce biostatistical methods to clear away roadblocks which stymy the development of predictive biomarkers.
| Wang, Ching-Yun; Cullings, Harry; Song, Xiao et al. (2017) Joint nonparametric correction estimator for excess relative risk regression in survival analysis with exposure measurement error. J R Stat Soc Series B Stat Methodol 79:1583-1599 |
| Wang, Ching-Yun; Song, Xiao (2016) Robust best linear estimator for Cox regression with instrumental variables in whole cohort and surrogates with additive measurement error in calibration sample. Biom J 58:1465-1484 |
