Table 1.
Comparison of Different Methods Using p-Value and R2
| Methods | R2 | p-value | |
|---|---|---|---|
| Wang et al. (2005) data set | Gene-based SPCA | 0.284 | 4.94 × 10−7 |
| Pathway-based SPCA | 0.304 | 1.33 × 10−7 | |
| Gene-based Lasso | 0.060 | 1.51 × 10−2 | |
| Pathway-based Lasso | 0.146 | 9.83 × 10−5 | |
| Miller et al. (2005) data set | Gene-based SPCA | 0.115 | 2.09 × 10−2 |
| Pathway-based SPCA | 0.160 | 2.97 × 10−3 | |
| Gene-based Lasso | 0.003 | 0.646 | |
| Pathway-based Lasso | 0.093 | 5.26 × 10−2 |
The p-values were calculated from likelihood ratio test comparing null model (intercept only) with a model with intercept and the linear predictor (used in last step of SPCA or Lasso model to predict survival outcome). The R2 statistic, which ranges from 0 to 1, represents the proportion of variation in outcome explained by prediction model.