Table 3.
Decreased false-positive and false-negative rates with increased AUC.a
| AUC | Sensitivity | Specificity | False positive rate, % | False negative rate, % |
|---|---|---|---|---|
| 0.8 | 82% | 75% | 25% | 18% |
| 77% | 77% | 23% | 23% | |
| 70% | 85% | 15% | 30% | |
| 0.85 | 82% | 78% | 22% | 18% |
| 78% | 78% | 22% | 22% | |
| 72% | 86% | 14% | 28% | |
| 0.9 | 94% | 72% | 28% | 6% |
| 88% | 78% | 22% | 12% | |
| 86% | 84% | 16% | 14% | |
| 0.95 | 98% | 74% | 26% | 2% |
| 88% | 88% | 12% | 12% | |
| 79% | 94% | 6% | 21% |
For any AUC, there will be a variety of sensitivities and specificities that depend on the shape of the ROC curve. The values in this table are based on hypothetical data and are meant to illustrate the large improvements in false-positive and false-negative rates that could be observed from small improvements in AUC. Based on this hypothetical data, the difference between a test with an AUC of 0.85 and 0.95 can improve the false-positive and false-negative rate from 22% to 12% for balanced sensitivity and specificity. When used as a screening test (i.e., when the cutoff is moved to improve sensitivity at the expense of specificity), the numbers of false negatives can improve from 18% to 2%.