Table 3.

Predictive performance of radiomics models, clinical model, and three joint models in different clinical stages

Model	Stage I–II n = 233	Stage III–IV n = 282	p for interaction
	AUC (95% CI)	AUC (95% CI)
Training set
Clinical model	0.728 (0.650–0.806)	0.729 (0.661–0.797)	0.856
CT_RF	0.791 (0.727–0.856)	0.576 (0.499–0.653)	< 0.001
CT joint model	0.838 (0.779–0.896)	0.699 (0.628–0.769)	0.001
PET_RF	0.711 (0.632–0.789)	0.589 (0.511–0.666)	0.005
PET joint model	0.753 (0.680–0.826)	0.720 (0.652–0.788)	0.251
PET/CT_RF	0.802 (0.737–0.867)	0.612 (0.534–0.689)	< 0.001
PET/CT joint model	0.794 (0.727–0.860)	0.722 (0.653–0.790)	0.032
Testing set
Clinical model	0.655 (0.446–0.865)	0.675 (0.547–0.804)	0.810
CT_RF	0.797 (0.634–0.961)	0.665 (0.537–0.794)	0.067
CT joint model	0.782 (0.634–0.929)	0.682 (0.556–0.808)	0.520
PET_RF	0.625 (0.439–0.811)	0.707 (0.577–0.836)	0.316
PET joint model	0.675 (0.481–0.870)	0.707 (0.580–0.835)	0.869
PET/CT_RF	0.744 (0.582–0.961)	0.702 (0.573–0.830)	0.784
PET/CT joint model	0.750 (0.589–0.911)	0.723 (0.599–0.846)	0.837

AUC area under the curve; CI confidence interval; RF random forest