Seminars

Clinicians aim toward a more preventative model of attacking cancer
by pinpointing and targeting specific early events in disease
development.  These early events can be measured as genomic,
proteomic, epidemiologic, and/or clinical variables, using
expression or Comparative Genomic Hybridization microarrays, tissue
microarrays (TMA), SELDI-TOF/mass spectra, patient histories, and
pathology and histology reports. The measurements are then used to
develop a patient profile to predict clinical outcomes such
as response to therapy or mortality in the hopes of developing
tailored treatments.

To date, there are no such profiles which have been implemented in
clinical practice. However, numerous methods are available to
unearth biologically driven associations between variables and
clinical outcomes with the intention of revealing potential
biomarkers. These various methods can be investigated to  build
candidate models which elucidate relevant patterns of association in
a given data set. The subsequent challenge is to assess how well these models predict outcomes in independent validation samples.

We will explore a general framework for comparing methods, selecting models, and assessing prediction error in the presence of censored outcomes. This approach is demonstrated in simulations and with tissue microarray data on breast cancer and melanoma cohorts.