Novel computational biology modeling system can accurately forecast response to neoadjuvant therapy in early breast cancer

Table 3 TumorScope Predict outcome metrics, overall and subgroups

	Overall (n = 80)	TNBC (n = 18)	HR−/HER2+ (n = 10)	HR+/HER2− (n = 28)	HR+/HER2+ (n = 24)
AUROC (with volume threshold)	0.910	0.938	0.917	0.75	0.919
AUROC (with response threshold)	0.905	0.900	0.812	0.823	0.944
Accuracy	0.912 (0.828, 0.964)	0.944 (0.727, 0.999)	0.900 (0.555, 0.998)	0.893 (0.718, 0.977)	0.917 (0.730, 0.989)
Sensitivity	0.889 (0.708, 0.977)	0.900 (0.555, 0.998)	1.000 (0.398, 1.000)	0.500 (0.068, 0.932)	1.000 (0.664, 1.000)
Specificity	0.925 (0.818, 0.979)	1.000 (0.631, 1.000)	0.833 (0.359, 0.996)	0.958 (0.789, 0.999)	0.867 (0.595, 0.983)
PPV	0.857 (0.615, 0.958)	1.000 (NA, NA)	0.800 (0.309, 0.973)	0.667 (0.018, 0.996)	0.818 (0.460, 0.960)
NPV	0.942 (0.883, 0.972)	0.889 (0.500, 0.985)	1.000 (NA, NA)	0.920 (0.880, 0.947)	1.000 (NA, NA)

Accuracy represented as area under the receiver operating characteristic curve (AUROC) for the predicted pCR, computed either from a cutoff in the final volume or a cutoff in the response percentage (95% confidence intervals shown in parentheses; NA indicates confidence interval is undefined)

ISSN: 1465-542X