Combining techniques for screening and evaluating interaction terms on high-dimensional time-to-event data

BMC Bioinformatics

Table 4 Results of the simulation study for rsf-VIF-res in scenarios with correlated variables

	IntScreen	IntSensiA	VarsTotal	MainSensi	IntSensi	rIPEC
						CoxBoostM	Final Model
Sim22_corr01	19779.06	0.22 (0.04)	17.76 (13.25)	0.75 (0.04)	0.21 (0.04)	0.11 (0.07)	0.1 (0.11)
Sim22_corr03	17961.56	0.23 (0.04)	12.74 (12.85)	0.46 (0.05)	0.18 (0.04)	0.06 (0.07)	0.03 (0.08)
Sim22_corr05	15953.22	0.15 (0.04)	6.66 (8.6)	0.14 (0.03)	0.06 (0.02)	0.03 (0.06)	0 (0.03)
Sim22_corr07	14174.12	0.15 (0.04)	7.26 (8.49)	0.06 (0.02)	0.04 (0.02)	-0.03 (0.2)	0.01 (0.3)

IntScreen (given as mean) is the number of selected interactions by the corresponding screening method; IntSensiA (given as ‘sensitivity value (sd)’) is the sensitivity related to the availability of true interactions; VarsTotal (given as ‘mean (sd)’) is the number of total variables in the final model; MainSensi (given as ‘sensitivity value (sd)’) is the sensitivity related to the inclusion of true main effects; and IntSensi (given as ‘sensitivity value (sd)’) is the sensitivity related to the inclusion of true interactions. The rIPEC values (given as ‘mean (sd)’) are shown for CoxBoostM and the final model. The scenarios were repeated 50 times.

ISSN: 1471-2105