. Author manuscript; available in PMC: 2017 Oct 15.

Published in final edited form as: Stat Med. 2016 Jun 1;35(23):4136–4152. doi: 10.1002/sim.6997

Table 5. Case study of concordance measures (95% confidence interval) in logistic regression and Cox regression.

The logistic regression model for unfavorable outcome after traumatic brain injury was developed in the International Tirilazad Trial and validated in the North American Tirilazad Trial. The Cox regression model for survival after revascularization was developed in the SYNTAX trial and validated in the CREDO-KYOTO registry.

	Logistic regression		Cox regression
	Apparent validation	External validation	Apparent validation	External validation
SD(Xβ)	1.028	1.112	0.904	0.965
Calibration slope $\hat{γ}$	1.000	1.023	1.000	0.785
Harrell’s c-index	0.749 (0.719 – 0.778)	0.779 (0.750 – 0.808)	0.744 (0.707 – 0.781)	0.725 (0.700 – 0.750)
Uno (τ = 4)			0.743 (0.705 – 0.782)	0.729 (0.687 – 0.771)
mbc = mbc(Xβ)	0.749 (0.721 – 0.778)	0.767 (0.759 – 0.775)^†	0.707 (0.680 – 0.734)	0.719 (0.715 – 0.722)^†
c-mbc = mbc( $\hat{γ}$ Xβ)		0.774 (0.746 – 0.803)		0.684 (0.667 – 0.700)

^†

95% confidence interval based on the variance estimate of mbc(Xβ) under the assumption of true β