2012 Project G: Multi-Marker Model in Cancer Clinical Trials
January 24, 2012
Dr. Keyue Ding, Senior Biostatisticians at NCIC Clinical Trials Group
High-throughput assays have allowed researcher to discover novel biomarkers
and differentially expressed genes. And various statistical issues for
analyzing such data being raised, and different modelling techniques
have been proposed to deal with issues. This project is to explore a
modelling strategy for multi-marker model building for prognostic and
predictive analysis in cancer clinical trials. The following issues
will be checked:
- Functional form of each marker:
a) using cutoff points (threshold effect);
b) fractional polynomials (continuous).
- Multi-marker model, combining multiple markers:
a) principle components,
b) Compound covariate,
c) SVM based on the prognostic and predictive risk scores.
- Evaluating the added predictive ability of new marker:
a) concordance (C)-index,
b) Net reclassification improvement (NRI),
c) integrated discrimination improvement (IDI).
Investigation will be based on real trial databases, and simulation
studies mimicking the trial data to determine the appropriate
approaches for multi-maker modeling strategy.