Table 3.

Multivariate cox regression analysis results of the derivation cohort.

Factors	Comparison	Model Aa		Model Bb		Model Cc
		HR (95% CI)	P	HR (95% CI)	P	HR (95% CI)	P
Clinicopathological factors
ENE	Yes vs. No	3.23 (1.55, 6.70)	0.002	2.35 (1.28, 4.32)	0.006	–
Responsed	Poor vs. Good	6.7 (3.30, 13.6)	<0.001	7.36 (3.85, 14.1)	<0.001	6.26 (3.33, 11.8)	<0.001
Somatic mutation profiles
TMB, mut/Mb	High (≥3.78) vs. Low	2.29 (1.16, 4.53)	0.017	1.84 (0.98, 3.43)	0.056	1.84 (0.98, 3.44)	0.058
RMS	High (≥1.135) vs. Low	5.73 (2.61, 12.6)	<0.001	5.87 (2.99, 11.5)	<0.001	5.49 (2.79, 10.8)	<0.001
CRMS	High (≥−0.62) vs. Low	8.42 (1.77, 40.1)	0.007	6.13 (1.43, 26.3)	0.015	7.34 (1.73, 31.1)	0.007
Harrel’s C-indexe		0.836 (0.763–0.909)		0.873 (0.811–0.934)		0.879 (0.821–0.938)

^aModel A abstracted the multivariate estimation results of significant factors for relapse risk according to Supplementary Table 3. All somatic mutation profiles were retained in the multivariate model, and clinicopathological factors with p < 0.2 in univariate analysis results were included in the multivariate model.

^bModel B included only the factors that were significant in Model A.

^cModel C included CCRT response and somatic mutation profiles only because the ENE status was lacking in the validation cohort.

^dPoor response indicates patients with PD response after receiving CCRT, and good response indicates patients with either CR, PR or SD response after receiving CCRT.

^eHarrell’s C-index was computed to estimate the model performance between three different multivariate models, and the model with the highest C-index indicated better predictive performance.