617 – Methods in Clinical Trials 1
Robust Platform Adjustment for Calling Breast Cancer Subtypes
Minya Pu
UC San Diego Moores Cancer Center
Karen Messer
UC San Diego Moores Cancer Center
Loki Natarajan
UC San Diego Moores Cancer Center
The PAM50 intrinsic subtypes are prognostic subtypes in breast cancer based on the expression profile of a set of 50 genes. To classify new breast tumors, the data for these genes are median centered, and subtypes are called according to the nearest centroid using the published PAM50 data. To build risk prediction models using over 1,700 breast tumor samples in the WHEL study, we used Nanostring technology to profile the PAM50 geneset. A platform adjustment is needed to translate parameter estimates from the published gene signatures based on microarray/qPCR into Nanostring data. Median centering is robust, but it works well only when the underlying subtype distribution in the new set of samples is similar to the one in the published training data. To avoid this caveat, we created a Nanostring-based centroids using paired Nanostring-microarray data on a subset (n=97) of the published training data. We compared several methods of platform adjustment, including calibration curves from smoothing splines, and simpler more robust approaches. We applied the platform adjusted centroids to data sets with different underlying subtype distributions, and described our recommended approach.