Table 3.

Comparison of LD and HD models for gene expression data with L2 (top) and L1 (bottom) penalties

	Predictors
Model type	Overall	Selected	Log-likelihood	c-Statistic
Cox
LD	5	5	−71.13	0.71	26.35
HD	24 496	24 026	−72.10	0.76	12.75
Exponential
LD	5	5	−86.26	0.71	16.83
HD	24 496	24 344	−98.72	0.75	112.69
Weibull
LD	5	5	−85.64	0.71	12.79
HD	24 496	22 299	−85.56	0.70	9.34
Log-logistic
LD	5	5	−85.65	0.70	14.74
HD	24 496	22 090	−86.14	0.70	5.37
Log-normal
LD	5	5	−52.10	0.70	16.02
HD	24 496	24 154	−65.14	0.66	23.61
Cox
LD	5	4	−71.10	0.71	30.19
HD	24 496	20	−72.64	0.71	7.49
Exponential
LD	5	5	−86.27	0.71	17.14
HD	24 496	13	−87.51	0.67	2.74
Weibull
LD	5	5	−85.63	0.71	12.79
HD	24 496	13	−86.80	0.66	3.75
Log-logistic
LD	5	5	−85.65	0.70	14.71
HD	24 496	9	−86.20	0.68	3.66
Log-normal
LD	5	5	−52.10	0.70	15.98
HD	24 496	9	−53.46	0.66	8.57

For L2 penalization, the number of selected predictors refers to the number of significant predictors . In both cases, bold emphases represent superior performance of one type of Cox model over the other. For a more comprehensive comparison, the results obtained using four parametric models (Mittal and others, 2013) are also presented in the lower parts of the table.