Computational and Mathematical Methods in Medicine

Research Article

A Gradient Boosting Algorithm for Survival Analysis via Direct Optimization of Concordance Index

Table 2

Numerical statistics of predictive concordance indices of GBM models and the Cox model on the breast cancer dataset. The five feature representations are explained in Table 1. “gbmsci”-I and “gbmcox”-I run without subsampling , while “gbmsci”-II and “gbmcox”-II run with subsampling . The numerics in each entry show the average C-index and the standard deviation (in parentheses) over 50 random runs. The best performance in each column is highlighted by the bold font.
Model
Feature Representation
cl clge ge mt mi
“gbmsci”-I 0.7107 (0.0015) 0.7287 (0.0005) 0.6599 (0.0004) 0.7145 (0.0004) 0.7416 (0.0010)
“gbmcox”-I 0.7039 (0.0008) 0.7268 (0.0013) 0.6523 (0.0007) 0.7110 (0.0014) 0.7222 (0.0003)
“gbmsci”-II 0.7063 (0.0011) 0.7341 (0.0014) 0.6617 (0.0020) 0.7169 (0.0017) 0.7405 (0.0015)
“gbmcox”-II 0.6983 (0.0009) 0.7298 (0.0008) 0.6549 (0.0014) 0.7173 (0.0010) 0.7306 (0.0008)
“cox” 0.7042 0.7140 0.6590 0.6659 0.7299