Table 1.

Impact of distinct mechanisms of resistance on the area under the curve (AUC) of receiver operating characteristic (ROC) curves derived with the predictive gene signatures generated.

Proportion of resistant cases	Optimal signature (2.0-fold)					p-value for trend
	Number of resistance mechanisms
	1	2	3	4	5
0.5*	1.000 (1.000-1.000)	1.000 (0.999-1.000)	0.992 (0.987-0.997)	0.971 (0.955-0.983)	0.931 (0.886-0.955)	< 0.0001
0.6	1.000 (1.000-1.000)	0.999 (0.997-1.000)	0.992 (0.985-0.996)	0.966 (0.943-0.980)	0.930 (0.882-0.954)	< 0.0001
0.7	1.000 (1.000-1.000)	1.000 (0.998-1.000)	0.991 (0.977-0.996)	0.967 (0.933-0.980)	0.927 (0.888-0.953)	< 0.0001
0.8	1.000 (1.000-1.000)	1.000 (0.998-1.000)	0.990 (0.974-0.996)	0.963 (0.933-0.980)	0.920 (0.870-0.947)	< 0.0001
0.9**	1.000 (1.000-1.000)	0.999 (0.998-1.000)	0.987 (0.967-0.995)	0.943 (0.898-0.967)	0.866 (0.786-0.919)	< 0.0001
0.95	1.000 (1.000-1.000)	0.999 (0.994-1.000)	0.951 (0.864-0.987)	0.803 (0.728-0.881)	0.687 (0.619-0.754)	< 0.0001

Proportion of resistant cases	Weak signature (1.4-fold)					p-value for trend
	Number of resistance mechanisms
	1	2	3	4	5
0.5*	0.999 (0.996-1.000)	0.968 (0.935-0.982)	0.900 (0.844-0.936)	0.818 (0.762-0.855)	0.736 (0.689-0.784)	< 0.0001
0.6	0.999 (0.996-1.000)	0.970 (0.949-0.984)	0.892 (0.843-0.925)	0.812 (0.747-0.855)	0.729 (0.682-0.771)	< 0.0001
0.7	0.999 (0.996-1.000)	0.967 (0.944-0.982)	0.895 (0.845-0.921)	0.797 (0.738-0.839)	0.708 (0.659-0.754)	< 0.0001
0.8	0.999 (0.996-1.000)	0.965 (0.933-0.986)	0.872 (0.810-0.909)	0.746 (0.692-0.784)	0.653 (0.619-0.689)	< 0.0001
0.9**	0.999 (0.994-1.000)	0.937 (0.883-0.968)	0.754 (0.692-0.820)	0.631 (0.581-0.686)	0.573 (0.527-0.612)	< 0.0001
0.95	0.998 (0.990-1.000)	0.808 (0.703-0.886)	0.610 (0.550-0.685)	0.550 (0.502-0.606)	0.526 (0.463-0.583)	< 0.0001

Proportion of resistant cases	Strong signature (2.8-fold)					p-value for trend
	Number of resistance mechanisms
	1	2	3	4	5
0.5*	1.000 (1.000-1.000)	1.000 (1.000-1.000)	1.000 (0.998-1.000)	0.994 (0.984-0.998)	0.985 (0.970-0.994)	0.002
0.6	1.000 (1.000-1.000)	1.000 (1.000-1.000)	0.999 (0.996-1.000)	0.995 (0.987-0.998)	0.982 (0.963-0.991)	0.0006
0.7	1.000 (1.000-1.000)	1.000 (1.000-1.000)	0.999 (0.998-1.000)	0.994 (0.983-0.998)	0.979 (0.958-0.989)	0.0006
0.8	1.000 (1.000-1.000)	1.000 (1.000-1.000)	0.999 (0.997-1.000)	0.993 (0.976-0.997)	0.977 (0.953-0.989)	< 0.0001
0.9**	1.000 (1.000-1.000)	1.000 (1.000-1.000)	0.999 (0.996-1.000)	0.990 (0.975-0.997)	0.966 (0.925-0.985)	0.0006
0.95	1.000 (1.000-1.000)	1.000 (1.000-1.000)	0.997 (0.991-1.000)	0.959 (0.893-0.988)	0.850 (0.757-0.929)	< 0.0001

Perturbed datasets in which s% (s%=5%, 10%, 20%, 30%, 40% or 50%) of the cases were designated to be therapy sensitive were generated. Within the 1-s% resistant cases, we allocated the cases randomly into n (n=1, 2, 3, 4, 5) equally-sized groups of resistance mechanisms. For each nth resistance mechanism, 100 genes were randomly selected as the “true” gene expression changes and were spiked-in by v (v=0.5,1,1.5). Classification was performed using diagonal linear discriminant analysis (DLDA). For each combination of s, n and v, we repeated the spiking and classification 100 times. The mean value with the 95% confidence intervals in parentheses of the AUC of ROCs for each combination of s, n and v are shown. For v=1, 0.5, 1.5, the sections are labeled “Optimal signature (2-fold)”, “Weak signature (1.4-fold)” and “Strong signature (2.8-fold)” respectively. The last column depicts the p-values for the trend tests as the number of resistance mechanisms is increased from 1 to 5 for a given s%.

^*ideal setting;

^**clinically-realistic estimate