Table 4. Comparison of methods for subgroup detection with respect to AUC values, for group size and subgroup size , , corresponding to Figure 1.
Criterion | AUC(LR)≥0.95 | AUC(Method) = 0.95 | ||||
Scenario | I | II | III | I | II | III |
LR (optimum) | 0.96 | 0.95 | 0.96 | 2.0 | 3.63 | 2.93 |
FisherSum | 0.89 | 0.87 | 0.72 | 2.6 | 5.28 | >6 |
ORT | 0.79 | 0.77 | 0.68 | ∞ | ∞ | ∞ |
OS | 0.70 | 0.72 | 0.66 | ∞ | ∞ | ∞ |
t-test | 0.79 | 0.77 | 0.49 | 4.2 | 9.24 | ∞ |
PADGE | 0.75 | 0.74 | 0.55 | 6.6 | >10 | ∞ |
Kurtosis | 0.60 | 0.62 | 0.66 | 4.6 | >10 | >15 |
Bartlett | 0.70 | 0.72 | 0.80 | 3.2 | 6.60 | 5.33 |
Left three columns contain AUC values when tests are compared at value with AUC for LR. Right three columns correspond to values with AUC for each test. Best values plotted in bold per simulation scenario.