Table 3.

Confidence intervals for select event and nonevent net reclassification indices in the cardiovascular disease study data. Intervals based on bootstrap estimates of the standard error, which we recommend, tend to be wider than intervals based on the formula for the variance of the estimated NRI statistic.. Recall that for a threshold t delineating high risk, ${\mathit{\text{NRI}}}_e^t=\mathrm{\Delta }\mathit{\text{TPR}}$ and ${\mathit{\text{NRI}}}_{\mathit{\text{ne}}}^t=-\mathrm{\Delta }\mathit{\text{FPR}}$.

	${\mathit{\text{NRI}}}_{\mathrm{e}}^{>0}=0.378$	${\mathit{\text{NRI}}}_{\mathit{\text{ne}}}^{>0}=0.319$
formula	(0.252 to 0.503)	(0.294 to 0.344)
bootstrap	(0.275 to 0.481)	(0.257 to 0.382)
	${\mathit{\text{NRI}}}_e^{0.03}=0.029$	${\mathit{\text{NRI}}}_{\mathit{\text{ne}}}^{0.03}=0.055$
formula	(−0.030 to 0.088)	(0.044 to 0.067)
bootstrap	(−0.039 to 0.097)	(0.026 to 0.084)
	${\mathit{\text{NRI}}}_e^{0.1}=0.191$	${\mathit{\text{NRI}}}_{\mathit{\text{ne}}}^{0.1}=-0.027$
formula	(0.125 to 0.258)	(−0.034 to −0.021)
bootstrap	(0.097 to 0.286)	(−0.039 to −0.016)