Table 2 Comparison of the models for OS and PFS based on held out validation sets.

From: Machine learning models for identifying predictors of clinical outcomes with first-line immune checkpoint inhibitor therapy in advanced non-small cell lung cancer

(a) Overall survival

Model

Hinge Loss

Margin Loss

Hinge Loss—1 Year

Hinge Loss—2 Year

Hinge Loss—Uncensored

Hinge Loss—Censored

Concordance Index Train

Concordance Index Test

CPH

209.0 (204.2, 213.9)

386.1 (376.6, 395.7)

149.3 (139.5, 159.1)

155.3 (146.2, 164.3)

227.5 (219.5, 235.5)

178.3 (160.5, 196.1)

0.687 (0.685, 0.689)

0.671 (0.660, 0.681)

LogLogisticAFT

212.7 (208.3, 217.1)

387.6 (377.7, 397.5)

153.5 (144.3, 162.8)

161.5 (153.7, 169.4)

234.6 (227.7, 241.5)

176.4 (160.2, 192.7)

0.691 (0.689, 0.693)

0.671 (0.662, 0.679)

GBDT (CPH)

206.8 (199.1, 214.5)

384.3 (371.7, 396.9)

142.7 (135.6, 149.8)

149.3 (142.4, 156.2)

219.8 (211.4, 228.1)

185.2 (170.1, 200.4)

0.715 (0.712, 0.719)

0.672 (0.654, 0.689)

SSVM

349.5 (338.1, 360.9)

527.6 (508.3, 546.9)

146.2 (142.5, 149.9)

236.0 (234.2, 237.7)

255.4 (242.3, 268.5)

502.5 (486.8, 518.2)

0.692 (0.690, 0.694)

0.671 (0.661, 0.680)

DeepSurv

209.1 (203.3, 215.0)

386.3 (377.0, 395.5)

146.9 (139.1, 154.8)

154.8 (146.0, 163.6)

224.9 (214.6, 235.3)

182.9 (166.1, 199.7)

0.698 (0.691, 0.706)

0.669 (0.656, 0.681)

(b) Progression-free survival

Model

Hinge Loss

Margin Loss

Hinge Loss—1 Year

Hinge Loss—2 Year

Hinge Loss—Uncensored

Hinge Loss—Censored

Concordance Index Train

Concordance Index Test

CPH

176.7 (171.6, 181.8)

263.8 (253.4, 274.1)

81.5 (79.1, 83.9)

117.3 (114.8, 119.8)

133.0 (128.2, 137.9)

356.1 (327.4, 384.7)

0.635 (0.633, 0.638)

0.611 (0.598, 0.624)

LogLogisticAFT

178.0 (173.7, 182.4)

265.3 (255.7, 274.8)

88.8 (85.0, 92.7)

120.9 (118.0, 123.7)

137.8 (132.7, 142.8)

343.3 (314.6, 372.1)

0.642 (0.640, 0.645)

0.612 (0.597, 0.626)

GBDT (CPH)

176.0 (169.9, 182.1)

262.9 (252.0, 273.9)

77.6 (76.1, 79.1)

115.2 (113.2, 117.2)

130.6 (127.8, 133.3)

362.7 (333.7, 391.7)

0.666 (0.663, 0.668)

0.612 (0.602, 0.622)

SSVM

259.3 (251.3, 267.2)

346.3 (333.4, 359.2)

128.7 (126.1, 131.3)

187.8 (182.4, 193.3)

196.6 (190.3, 202.8)

517.1 (484.1, 550.1)

0.643 (0.640, 0.645)

0.612 (0.599, 0.625)

DeepSurv

176.3 (170.7, 182.0)

263.4 (252.7, 274.1)

79.0 (75.6, 82.4)

115.6 (112.8, 118.4)

131.0 (127.3, 134.7)

362.5 (330.8, 394.2)

0.652 (0.650, 0.653)

0.612 (0.598, 0.625)

  1. AFT, accelerated failure time; CPH, Cox proportional hazard; GBDT, gradient-boosted decision tree; SSVM, survival support vector machine.
Back to article page