|
Resolution: standard / high Figure 3.
Performance, model size distribution and variable stability of reduced models as described
in Figure 1 for node-negative (node(-)) and hormone receptor positive (+) subpopulation. We included patients whose tumors were estrogen receptor (ER) positive, progesterone
receptor (PR) positive or both. The left column shows the models for all variables
excluding nodal status, and the right column shows models for the clinico-pathological
variables alone (tumor size, nuclear grade, age, human epidermal growth factor receptor
(HER) 2, ER, PR). The compact, reduced model (RM) derived from molecular and clinico-pathological
covariates dramatically outperformed the full models (FM), and included AURKA, tumor
size, HER2, CD68 and nuclear grade. ATD-AUCROC, average time-dependent area under
the receiver operator characteristic curve.
Parisi et al. Breast Cancer Research 2010 12:R66 doi:10.1186/bcr2633 |