Table 3.

KM stratification summary with HR [95% CI] between high and low-risk groups across all test cohorts for 2D axial MIL, radiomics and clinical models alongside logrank p-value and cutoff extracted from the training cohort. NAs indicate that stratification into two groups produced only one group with events and was thus not possible to assess separation. HRs smaller than 1 indicate that patients with higher risk scores have a smaller probability of event

Model	Endpoint	Cutoff	HR [95% CI] RADCURE-test	Logrank p-value	HR [95% CI] HN1	Logrank p-value	HR [95% CI] HN-PET-CT	Logrank p-value
2D MIL axial	OS	0.48169	3.98 [2.63–6.00]	< 0.001	2.45 [1.54–3.91]	< 0.001	4.09 [1.85–9.04]	< 0.001
2D MIL axial	LRC	0.51809	4.77 [2.84–8.02]	< 0.001	3.28 [1.52–7.06]	0.001	2.14 [1.02–4.48]	0.039
2D MIL axial	FFDM	0.46167	4.24 [2.28–7.88]	< 0.001	2.11 [0.50–8.86]	0.30	2.51 [1.05–6.00]	0.032
Radiomics	OS	0.43367	4.26 [2.79–6.49]	< 0.001	2.01 [1.27–3.27]	0.003	1.71 [0.73–4.01]	0.20
Radiomics	LRC	0.44668	2.52 [1.57–4.05]	< 0.001	3.66 [1.76–7.64]	< 0.001	1.99 [1.04–3.80]	0.033
Radiomics	FFDM	0.41091	3.43 [1.97–5.67]	< 0.001	5.04 [1.02–25.00]	0.028	3.13 [0.75–13.00]	0.098
Clinical	OS	0.37451	5.43 [3.42–8.61]	< 0.001	5.31 [2.85–9.91]	< 0.001	8.98 [3.84–21.08]	< 0.001
Clinical	LRC	0.40972	4.98 [3.03–8.18]	< 0.001	20.61 [2.80–150.55]	< 0.001	3.61 [1.51–8.59]	0.002
Clinical	FFDM	0.39670	4.85 [2.55–9.21]	< 0.001	NA	NA	5.73 [1.76–18.63]	0.001