Table 3.
Concordance between classifiers and association with clinical parameters*
| Model | SCMGENE | SCMOD2 | SCMOD1 | PAM50 | SSP2006 | SSP2003 |
| Concordance between pairs of subtype classifications, Cohen’s κ (95% CI), %† | ||||||
| SCMOD2 | 0.70 (0.69 to 0.72), 78 | |||||
| SCMOD1 | 0.65 (0.64 to 0.67), 75 | 0.81 (0.80 to 0.82), 86 | ||||
| PAM50 | 0.59 (0.58 to 0.61] 70 | 0.68 (0.67 to 0.70), 76 | 0.67 (0.65 to 0.68), 76 | |||
| SSP2006 | 0.47 (0.45 to 0.48), 59 | 0.51 (0.50 to 0.53), 63 | 0.50 (0.49 to 0.52), 62 | 0.58 (0.56 to 0.59), 68 | ||
| SSP2003 | 0.34 (0.32 to 0.36), 49 | 0.38 (0.37 to 0.40), 53 | 0.36 (0.34 to 0.38), 51 | 0.45 (0.43 to 0.47), 58 | 0.55 (0.53 to 0.56), 67 | |
| Concordance between subtype classifications and risk predictions, Cohen's κ (95% CI), %‡ | ||||||
| MAMMAPRINT | 0.49 (0.46 to 0.51), 81 | 0.56 (0.53 to 0.58), 82 | 0.55 (0.52 to 0.57), 82 | 0.51 (0.49 to 0.54), 79 | 0.44 (0.42 to 0.46), 72 | 0.38 (0.36 to 0.40), 68 |
| ONCOTYPE | 0.61 (0.59 to 0.64), 83 | 0.62 (0.59 to 0.64), 83 | 0.56 (0.53 to 0.58), 81 | 0.61 (0.58 to 0.63), 82 | 0.60 (0.57 to 0.62), 80 | 0.56 (0.54 to 0.58), 77 |
| GGI | 0.58 (0.56 to 0.61), 80 | 0.67 (0.65 to 0.69), 84 | 0.61 (0.59 to 0.64), 81 | 0.73 (0.70 to 0.75), 87 | 0.70 (0.68 to 0.72), 85 | 0.58 (0.55 to 0.60), 79 |
| Association between subtype classifications and clinical parameters, Cramer's V (95% CI)§ | ||||||
| ER | 0.64 (0.61 to 0.67) | 0.71 (0.68 to 0.74) | 0.69 (0.66 to 0.72) | 0.71 (0.68 to 0.74) | 0.69 (0.66 to 0.72) | 0.69 (0.66 to 0.72) |
| PGR | 0.46 (0.42 to 0.5) | 0.54 (0.5 to 0.58) | 0.54 (0.5 to 0.58) | 0.54 (0.50 to 0.58) | 0.53 (0.49 to 0.57) | 0.52 (0.48 to 0.56) |
| HER2 | 0.52 (0.47 to 0.57) | 0.49 (0.44 to 0.54) | 0.48 (0.42 to 0.53) | 0.41 (0.36 to 0.47) | 0.39 (0.33 to 0.44) | 0.34 (0.29 to 0.39) |
| Histological grade‖ | 0.55 (0.51 to 0.59) | 0.58 (0.54 to 0.62) | 0.58 (0.54 to 0.62) | 0.59 (0.54 to 0.63) | 0.54 (0.5 to 0.58) | 0.51 (0.47 to 0.55) |
| Tumor size (>2 cm) | 0.10 (0.065 to 0.13) | 0.15 (0.11 to 0.18) | 0.18 (0.14 to 0.21) | 0.16 (0.12 to 0.19) | 0.13 (0.089 to 0.16) | 0.14 (0.11 to 0.17) |
| Nodal status | 0.07 (0.04 to 0.09) | 0.08 (0.06 to 0.1) | 0.08 (0.06 to 0.10) | 0.09 (0.07 to 0.11) | 0.08 (0.05 to 0.09) | 0.10 (0.07 to 0.11) |
| Age at diagnosis (>50 y) | 0.13 (0.10 to 0.16) | 0.09 (0.05 to 0.11) | 0.12 (0.09 to 0.15) | 0.12 (0.09 to 0.15) | 0.12 (0.08 to 0.14) | 0.14 (0.10 to 0.17) |
Concordance between subtype classifiers and prognostic gene signatures was estimated using Cohen's κ statistics. Association between subtype classifiers and clinical parameters was estimated using Cramer's V statistic. The statistical significance was calculated using a χ2 test. ER = estrogen receptor; GGI = prognostic gene signature (16); MAMMAPRINT = prognostic gene signature (14); PGR = progesterone receptor; ONCOTYPE = prognostic gene signature (15); PAM50 = single sample predictor (3); SCMGENE = three-gene subtype classification model; SCMOD2 = subtype classification model (8); SCMOD1 = subtype classification model (1); SSP2006 = single sample predictor (2); SSP2003: single sample predictor (6).
Cohen's κ statistic (point estimate and 95% confidence interval in parentheses; all Ps < .001, two-sided) estimating the concordance between each pair of subtype classifications as computed by the models under study (note that the SCMs do not identify normal-like tumors). In addition to the κ statistic, the agreement between classifiers, as estimated by the percentage (%) of identical subtype classifications, is given.
Cohen's κ statistic (point estimate and 95% confidence interval in parentheses; all Ps < .001, two-sided) estimating the concordance between the subtype classifications (where basal-like, HER2-enriched, and luminal B are considered high risk and luminal A and normal-like tumors are considered low risk) and the risk predictions as computed by the prognostic gene signatures (where the intermediate- and low-risk group, as defined by applying the published algorithm of the Oncotype DX prognostic model—ONCOTYPE—are combined). In addition to the κ statistic, the agreement between classifiers, as estimated by the percentage (%) of identical subtype classifications is given.
Cramer's V statistic (point estimate and 95% confidence interval in parentheses; all Ps < .001, two-sided) estimating the association between subtype classifications and widely used clinical parameters.
Intermediate histological grade was not considered for estimating the concordance with subtype classifications because they are present in all the molecular subtypes.