TABLE 4.

Proportion of rejections of the null hypothesis H₀: NB₁(t) = NB₂(t) based on 500 simulations for comparing estimates of NB for two risk models evaluated on the same validation data, when NB(t) is estimated, using observed risks R in a population assuming that the model is well-calibrated; using risk estimates from a case–control sample when the disease prevalence μ is known; and using observations of (R, Y) in the population

					Proportion rejected H0 for
(α1, β1)	(α2, β2)	AUC 1	AUC 2	t	TR	TCC	TRY
(6.55,124.45)	(6.55,124.45)	0.61	0.61	0.02	0.050	0.036	0.034
				0.03	0.046	0.054	0.054
				0.04	0.054	0.044	0.042
				0.05	0.044	0.062	0.070
				0.06	0.052	0.052	0.050
(6.55,124.45)	(4,76)	0.61	0.64	0.02	0.150	0.108	0.116
				0.03	0.906	0.176	0.218
				0.04	1.000	0.192	0.266
				0.05	1.000	0.240	0.272
				0.06	1.000	0.208	0.234
(6.55,124.45)	(3,57)	0.61	0.66	0.02	0.418	0.286	0.306
				0.03	1.000	0.442	0.524
				0.04	1.000	0.570	0.648
				0.05	1.000	0.560	0.640
				0.06	1.000	0.424	0.522
(6.55,124.45)	(2,38)	0.61	0.69	0.02	0.946	0.684	0.69
				0.03	1	0.882	0.932
				0.04	1	0.934	0.974
				0.05	1	0.924	0.978
				0.06	1	0.872	0.944

Note. Results are based on data sets with N = 10, 000 and 500 simulations for each set of parameters and values of threshold t. For the case–control design, three controls were sampled for each case.