Table 3.

Performance of repeated five-fold cross-validated LASSO-regularized Cox regression-based classifier using 14 concentration-normalized biomarkers, stratified by cohort and subsets thereof.

Classifier prediction	Events/N	Median OS (95% CI)	HR (95% CI)	P-value
Discovery cohort (n=202)
Likely to benefit	92/179	54.3 (37.9, NR)	Reference
Unlikely to benefit	21/23	3.7 (2.4, 10.8)	5.1 (3.1, 8.4)	7.6×10-11
Discovery cohort: training set (n=79)
Likely to benefit	31/67	55.2 (42.6, NR)	Reference
Unlikely to benefit	11/12	2.5 (1.2, NR)	10.3 (4.7, 22.6)	4.5×10-9
Discovery cohort: validation set (n=59)
Likely to benefit	28/55	54.8 (16.3, NR)	Reference
Unlikely to benefit	4/4	5.8 (2.9, NR)	3.9 (1.4, 11.4)	0.012
Discovery cohort: test set (n=64)
Likely to benefit	33/57	30.2 (16.4, NR)	Reference
Unlikely to benefit	6/7	6.0 (2.4, NR)	2.7 (1.1, 6.6)	0.026
Discovery cohort: validation and test set (n=123)
Likely to benefit	61/112	40.6 (24.8, NR)	Reference
Unlikely to benefit	10/11	6.0 (3.3, NR)	3.2 (1.6, 6.3)	8.2×10-4
Independent validation cohort (n=27)
Likely to benefit	6/23	NR (15.8, NR)	Reference
Unlikely to benefit	3/4	6.0 (2.4, NR)	5.6 (1.2, 25.5)	0.027

NR, not reached.