Skip to main content

Advertisement

Table 3 Performance measures of seven-gene Heterogeneous Mixture Discriminant Analysis (MDAhet) classifier

From: A robust classifier of high predictive value to identify good prognosis patients in ER-negative breast cancer

  Training set    Test Sets   
Cohort NKI2+EMC+NCH UPP JRH-2 CAL Kreike UNC248 Loi
Cohort size 186 34 24 46 97 85 40
Annotated 183 31 24 46 71 80 34
Good prognosis (%) 59 81 75 67 76 74 76
Poor prognosis (%) 41 19 25 33 24 26 24
Chemotherapy (%) 0 0 0 67 0 66 0
MDA        
NPV (%) 74 92 93 69 83 74 100
PPV (%) 55 28 56 35 29 27 40
SE (%) 69 83 83 53 71 38 100
SP (%) 61 48 78 52 44 63 54
MDAhet        
NPV (%) 80 100 100 100 85 92 100
PPV (%) 51 30 37 45 29 36 35
SE (%) 84 100 100 100 76 90 100
SP (%) 44 44 44 42 41 42 42
NPV at 4 years (%) 83 100 100 100 88 93 100
PPV at 4 years (%) 42 24 33 35 25 45 35
SE at 4 years (%) 83 100 100 100 79 88 100
SP at 4 years (%) 44 42 43 37 40 45 43
LN        
NPV (%) 61 84 NA 85 NA 85 76
PPV (%) 50 30 NA 46 NA 37 0a
SE (%) 27 50 NA 80 NA 71 0a
SP (%) 81 70 NA 55 NA 58 100a
NPV at 4 years (%) 67 88 NA 90 NA 82 77
PPV at 4 years (%) 39 37 NA 38 NA 47 0a
SE at 4 years (%) 25 84 NA 85 NA 69 0a
SP at 4 years (%) 80 74 NA 53 NA 60 100a
  1. aLoi's cohort consists only of LN- samples. The table summarises performance indicators of the seven-gene MDAhet classifier and lymph node status (LN) across oestrogen receptor negative (ER-) training and test sets. For each cohort, we also give the number of tumours (cohort size), number of clinically annotated tumours (annotated), the percentage of good and poor prognosis patients (as defined by disease-specific death or distant metastasis event) and the percentage of patients treated with chemotherapy. NPV, PPV, SE and SP are evaluated at four years and at end of study. NPV, negative predictive value (precision for good prognosis); PPV, positive predictive value (precision for poor prognosis); SE, sensitivity; SP, specificity.