A whole slide image-based machine learning approach to predict ductal carcinoma in situ (DCIS) recurrence risk

Breast Cancer Research

Table 2 Model performance

Model and Clinicopathologic Variables 2x2 Performance Metrics
	Training Cohort					Validation Cohort
Variable	Rec. Status at 10 Years			Metrics		Rec. Status at 10 Years			Metrics
Model		Censored	Recurred	Acc: 0.87	PPV: 0.65		Censored	Recurred	Acc: 0.85	PPV: 0.46
	Low Risk	116	9	Sn: 0.71	NPV: 0.93	Low Risk	144	13	Sn: 0.50	NPV: 0.92
	High Risk	12	22	Sp: 0.91	OR: 23.6	High Risk	15	13	Sp: 0.91	OR: 9.60
Necrosis		Censored	Recurred	Acc: 0.41	PPV: 0.18		Censored	Recurred	Acc: 0.26	PPV: 0.13
	No	47	13	Sn: 0.58	NPV: 0.78	No	28	6	Sn: 0.77	NPV: 0.82
	Yes	81	18	Sp: 0.37	OR: 0.80	Yes	131	20	Sp: 0.18	OR: 0.71
Size		Censored	Recurred	Acc: 0.50	PPV: 0.15		Censored	Recurred	Acc: 0.49	PPV: 0.10
	Below	68	20	Sn: 0.33	NPV: 0.77	Below	83	18	Sn: 0.31	NPV: 0.82
	Above	58	10	Sp: 0.54	OR: 0.59	Above	76	8	Sp: 0.52	OR: 0.49
Age		Censored	Recurred	Acc: 0.27	PPV: 0.18		Censored	Recurred	Acc: 0.24	PPV: 0.15
	Below	19	7	Sn: 0.77	NPV: 0.73	Below	21	2	Sn: 0.92	NPV: 0.91
	Above	109	24	Sp: 0.15	OR: 0.60	Above	138	24	Sp: 0.13	OR: 1.83
Radiotherapy		Censored	Recurred	Acc: 0.64	PPV: 0.19		Censored	Recurred	Acc: 0.66	PPV: 0.05
	No	94	23	Sn: 0.26	NPV: 0.80	No	121	24	Sn: 0.08	NPV: 0.83
	Yes	34	8	Sp: 0.73	OR: 0.96	Yes	38	2	Sp: 0.76	OR: 0.27
Grade		Censored	Recurred	Acc: 0.41	PPV: 0.21	*
	I/II	41	8	Sn: 0.74	NPV: 0.84
	III	86	23	Sp: 0.32	OR: 1.37

The 2 × 2 confusion matrix and performance metrics for the 8-feature model and common clinopathological variables in the training and validation cohorts. For each variable, the positive condition was recurrence within 10 years. A 2 × 2 matrix for grade in the validation cohort was omitted due to all patients belonging to grade III. Margin status was not shown for either cohort due to almost all patients having negative margins. The threshold used for patient age was 50 years, and the threshold for size was 2 cm

ISSN: 1465-542X