Table 5 Mean (95% CI) integrated Brier scores for competing learning methods, averaged over 40 simulations in nine scenarios where interaction and non-linear is assumed.
Sample size | |||
|---|---|---|---|
N = 500 | N = 1000 | N = 5000 | |
Censoring = 50% | |||
Cox PH | 0.113 (0.098, 0.128) | 0.099 (0.094, 0.104) | 0.083 (0.081, 0.085) |
Penalized Cox PH | 0.104 (0.092, 0.115) | 0.095 (0.091, 0.099) | 0.082 (0.080, 0.083) |
Penalized Cox PH (NLI) | 0.097 (0.090, 0.105) | 0.091 (0.082, 0.099) | 0.079 (0.074, 0.085) |
RSF | 0.109 (0.106, 0.112) | 0.100 (0.097, 0.102) | 0.084 (0.082, 0.085) |
ORSF-fast | 0.102 (0.096, 0.107) | 0.094 (0.092, 0.097) | 0.079 (0.077, 0.080) |
ORSF-cph | 0.101 (0.096, 0.105) | 0.094 (0.092, 0.097) | 0.079 (0.078, 0.080) |
ORSF-net | 0.101 (0.095, 0.106) | 0.094 (0.091, 0.096) | 0.079 (0.078, 0.080) |
Censoring = 70% | |||
Cox PH | 0.150 (0.126, 0.174) | 0.126 (0.114, 0.137) | 0.104 (0.101, 0.106) |
Penalized Cox PH | 0.142 (0.121, 0.163) | 0.120 (0.110, 0.129) | 0.102 (0.100, 0.104) |
Penalized Cox PH (NLI) | 0.127 (0.110, 0.143) | 0.112 (0.104, 0.119) | 0.110 (0.095, 0.125) |
RSF | 0.146 (0.133, 0.159) | 0.133 (0.125, 0.141) | 0.115 (0.113, 0.117) |
ORSF-fast | 0.136 (0.121, 0.151) | 0.128 (0.119, 0.138) | 0.108 (0.106, 0.110) |
ORSF-cph | 0.137 (0.121, 0.152) | 0.128 (0.118, 0.137) | 0.108 (0.106, 0.110) |
ORSF-net | 0.137 (0.120, 0.155) | 0.128 (0.119, 0.138) | 0.108 (0.106, 0.110) |
Censoring = 90% | |||
Cox PH | 0.123 (0.104, 0.142) | 0.097 (0.091, 0.104) | 0.097 (0.089, 0.106) |
Penalized Cox PH | 0.125 (0.110, 0.139) | 0.105 (0.096, 0.113) | 0.097 (0.089, 0.105) |
Penalized Cox PH (NLI) | 0.135 (0.114, 0.156) | 0.107 (0.099, 0.114) | 0.095 (0.089, 0.101) |
RSF | 0.145 (0.125, 0.166) | 0.113 (0.108, 0.119) | 0.117 (0.110, 0.124) |
ORSF-fast | 0.170 (0.145, 0.195) | 0.130 (0.125, 0.135) | 0.113 (0.107, 0.119) |
ORSF-cph | 0.168 (0.141, 0.195) | 0.123 (0.118, 0.128) | 0.113 (0.107, 0.119) |
ORSF-net | 0.163 (0.137, 0.189) | 0.123 (0.119, 0.128) | 0.112 (0.106, 0.118) |
