Prognostic survival model for people diagnosed with invasive cutaneous melanoma

BMC Cancer

Table 2 Effect of removing and adding variables from the prognostic model (Probit model with 3df) on discrimination (D) and explained variation ( \( {\boldsymbol{R}}_{\boldsymbol{D}}^{\mathbf{2}} \) )

	Removing variable					Adding variable
Variable	Single^a		Cumulative^b			Single variable only			Added to thickness
	D	\( {\boldsymbol{R}}_{\boldsymbol{D}}^{\mathbf{2}} \)	Order	D	\( {\boldsymbol{R}}_{\boldsymbol{D}}^{\mathbf{2}} \)	D	\( {\boldsymbol{R}}_{\boldsymbol{D}}^{\mathbf{2}} \)	Harrell’s C	D^d	\( {\boldsymbol{R}}_{\boldsymbol{D}}^{\mathbf{2}} \) ^d
<full model>	1.526	0.478
Clark’s level	1.521	0.476	1	1.521	0.476	1.108	0.325	0.815	0.006	0.002
Morphology	1.519	0.475	2	1.513	0.473	0.643	0.140	0.699	0.017	0.006
Gender	1.513	0.473	3	1.500	0.469	0.277	0.029	0.569	0.038	0.014
Age^c	1.504	0.470	4	1.467	0.458	0.398	0.058	0.646	0.042	0.015
Positive lymph nodes	1.504	0.470	5	1.457	0.451	0.990	0.278	0.541	0.038	0.014
Ulceration	1.503	0.470	6	1.409	0.438	0.998	0.281	0.715	0.044	0.016
metastasis	1.494	0.467	7	1.358	0.420	1.068	0.309	0.530	0.053	0.019
Body site	1.490	0.466	8	1.307	0.401	0.303	0.035	0.600	0.052	0.019
Thickness^c	1.354	0.419	9	-	-	1.307	0.401	0.866

^aBased on the prognostic regression model after removing only the given variable.
^bBased on the prognostic regression model after sequentially removing the variables in the stated order (1 = first, 2 = second, ..).
^cTreated as transformed continuous variables (see text for details). The remaining variables are categorical.
^dValues represent the difference in fit statistics between the model with thickness only ^cand the model including thickness ^cand the shown variable.

ISSN: 1471-2407