Table 2.

Simulation results for comparison of two markers via the difference in respective WMR estimators. We assumed follows a multivariate normal distribution with , , and cor(M_A,M_B) = 0.8. An independent censoring time was generated such that either 20% or 40% subjects were censored. We show the difference between the theoretical AUC for marker A versus marker B (dAUC(t)), the difference between the corresponding estimated WMR statistics (dWMR(t)), the Monte-Carlo SD (MCSD) and the SD estimated using the proposed variance estimator (EstSD) and the coverage (nominal: 95.0)

Log time	dAUC(t)	dWMR(t)	MCSD	EstSD	Coverage
20% censoring
− 2.0	0.101	0.099	0.062	0.054	88.8
− 1.5	0.098	0.096	0.040	0.040	94.5
− 1.0	0.091	0.090	0.032	0.032	94.6
− 0.5	0.080	0.078	0.029	0.029	94.8
0.0	0.070	0.069	0.030	0.029	94.9
0.5	0.060	0.060	0.037	0.035	93.2
1.0	0.051	0.053	0.054	0.051	93.3
40% censoring
− 2.0	0.101	0.099	0.062	0.054	88.8
− 1.5	0.098	0.096	0.041	0.040	94.7
− 1.0	0.091	0.090	0.033	0.033	95.0
− 0.5	0.080	0.079	0.031	0.031	94.8
0.0	0.070	0.071	0.035	0.035	94.5
0.5	0.060	0.061	0.048	0.047	94.8
1.0	0.051	0.053	0.100	0.089	93.3