Radiomics for therapy-specific head and neck squamous cell carcinoma survival prognostication (part I)

BMC Medical Imaging

Table 2 Model performance for the EN, RSF, benchmark and combined (quantitative imaging features and benchmark) models

Cohort	AUC test	95% CI	AUC train	95% CI	p(test) vs RSF	p(test) vs Benchmark or combined
EN
Complete	0.711	0.695–0.728	0.799	0.793–0.805	0.002	< 0.001
Surgery	0.672	0.627–0.718	0.778	0.751–0.804	0.938	< 0.001
RCTX	0.560	0.527–0.593	0.827	0.808–0.846	< 0.001	0.420
RT	0.585	0.516–0.653	0.778	0.748–0.808	0.158	0.434
EN Benchmark
Complete	0.632	0.613–0.650	0.745	0.730–0.760
Surgery	0.445	0.404–0.486	0.772	0.741–0.804
RCTX	0.578	0.546–0.611	0.789	0.771–0.808
RT	0.621	0.560–0.681	0.776	0.751–0.801
RSF
Complete	0.746	0.731–0.760	0.948	0.946–0.949		< 0.001
Surgery	0.675	0.629–0.721	0.950	0.946–0.954		0.006
RCTX	0.652	0.622–0.681	0.938	0.935–0.942		0.002
RT	0.507	0.422–0.591	0.897	0.887–0.907		< 0.001
RSF Benchmark
Complete	0.669	0.651–0.686	0.827	0.823–0.830
Surgery	0.578	0.526–0.630	0.809	0.797–0.821
RCTX	0.575	0.538–0.613	0.858	0.852–0.865
RT	0.788	0.734–0.842	0.818	0.803–0.833
EN combined
Complete	0.799	0.793–0.805	0.798	0.792–0.805	< 0.001	0.065
Surgery	0.645	0.604–0.685	0.795	0.770–0.819	0.335	0.369
RCTX	0.555	0.525–0.586	0.842	0.820–0.864	< 0.001	0.853
RT	0.595	0.527–0.664	0.833	0.803–0.864	0.601	0.828
RSF combined
Complete	0.759	0.745–0.772	0.949	0.948–0.951		0.215
Surgery	0.673	0.632–0.714	0.953	0.947–0.958		0.944
RCTX	0.650	0.615–0.684	0.943	0.939–0.956		0.924
RT	0.568	0.488–0.647	0.904	0.895–0.912		0.298

The Cox-Survival (Harrel's) C (AUC) with the respective confidence interval is shown for each model and subgroup for the training and test dataset. For statistical analysis two-sided, independent Student’s t-test was performed comparing the iterated AUC values of the test dataset
AUC Cox-Survival (Harrel’s) C, EN Elastic net, RCTX Radiochemotherapy, RT Radiotherapy, RSF Random survival forest

ISSN: 1471-2342