Skip to main content

Table 4 Discrimination, calibration and clinical utility measures of pathologic and pre-operative prediction models

From: Lymph node positivity in different early breast carcinoma phenotypes: a predictive model

  Pathologic model Pre-operative model
Probability of LN positivity
Criteria Parametre Initial set Validation set Initial set Validation set
AUC Est 0.754 0.767 0.681 0.687
  95%CI [0.742–0.765] [0.75–0.783] [0.668–0.693] [0.669–0.705]
AUC (Bootstrap, Est 0.755 0.766 0.682 0.686
B = 2000) 95%CI [0.744–0.767] [0.762–0.769] [0.669–0.694] [0.682–0.69]
Clinical utility < 10% 19 (0%) 7 (0%) 1 (0%) 0 (0%)
10–20% 2255 (32%) 1096 (31%) 579 (7%) 279 (7%)
20–30% 1258 (18%) 586 (17%) 3137 (40%) 1487 (38%)
30–40% 1108 (16%) 559 (16%) 2478 (32%) 1315 (33%)
40–50 391 (5%) 188 (5%) 228 (3%) 125 (3%)
> = 50% 2102 (29%) 1076 (31%) 1429 (18%) 746 (19%)
Calibration p-value 0.158 0.332 0.815 0.200
Probability of macrometastases
AUC (Delong) Est 0.780 0.798 0.718 0.727
95%CI [0.767–0.792] [0.78–0.815] [0.703–0.732] [0.707–0.746]
AUC (Bootstrap, Est 0.780 0.796 0.717 0.725
B = 2000) 95%CI [0.767–0.793] [0.793–0.799] [0.703–0.732] [0.721–0.728]
Clinical utility < 10% 2029 (28%) 1004 (29%) 358 (5%) 184 (5%)
10–20% 2289 (32%) 1075 (31%) 5049 (64%) 2450 (62%)
20–30% 512 (7%) 262 (7%) 829 (11%) 465 (12%)
30–40% 726 (10%) 358 (10%) 307 (4%) 162 (4%)
40–50 644 (9%) 378 (11%) 573 (7%) 306 (8%)
> = 50% 933 (13%) 435 (12%) 736 (9%) 385 (10%)
Calibration p-value 0.427 0.024 0.568 0.174