Table 3.

Accuracy and kappa metrics for each surrogate classifier and dataset

	Cohen kappa unweighted (95% CI)		Overall accuracy (95% CI)
Surrogate	STO-3	Clinseq	STO-3	Clinseq
IHC3	0.41 (0.35 to 0.47)	0.43 (0.34 to 0.52)	0.69 (0.65 to 0.73)	0.68 (0.62 to 0.74)
St. Gallen14	0.38 (0.32 to 0.45)	0.36 (0.27 to 0.44)	0.60 (0.55 to 0.64)	0.54 (0.47 to 0.60)
St. Gallen20	0.40 (0.33 to 0.46)	0.43 (0.34 to 0.51)	0.61 (0.57 to 0.65)	0.61 (0.54 to 0.67)
St. GallenFree	0.39 (0.32 to 0.45)	0.52 (0.43 to 0-61)	0.62 (0.58 to 0.66)	0.69 (0.63 to 0.75)
Prolif14	0.48 (0.41 to 0.54)	0.42 (0.34 to 0.51)	0.70 (0.66 to 0.74)	0.60 (0.53 to 0.66)
Prolif20	0.48 (0.41 to 0.54)	0.49 (0.40 to 0.58)	0.71 (0.67 to 0.75)	0.67 (0.60 to 0.73)
ProlifFree	0.48 (0.41 to 0.54)	0.57 (0.48 to 0.66)	0.71 (0.67 to 0.75)	0.75 (0.69 to 0.80)
3-class simplea: luminal vs HER2 vs basal-like	0.71 (0.65 to 0.78)	0.69 (0.58 to 0.80)	0.91 (0.88 to 0.93)	0.90 (0.85 to 0.93)

^a3-class simple: collapsing luminal A and B into 1 class. CI = confidence interval; IHC = immunohistochemistry.