Table 2.

Summary of prediction performance in the BCSC data. Model 1 included all predictors other than the breast density; model 2 included all four predictors. Since IPA is a relative measure with respect to the null model, the PI for the IPA is calculated as the difference of IPAs between the two models.

	BS	MSEP	IPA
Model 1	3.0271E-03	6.0207E-06	0.042%
Model 2	3.0266E-03	5.4617E-06	0.060%
PI	0.018%	9.280%	0.018%

PI: performance improvement; BS: Brier score; IPA: index of prediction accuracy; MSEP: mean square error for the probability of binary outcome; BCSC: Breast Cancer Surveillance Consortium,