Table 3.
ROC-Analyses for PQ-16, PCA and PCA/PQ-16 Combined
| Measure | Cut-off | Sensitivity | Specificity | PPV % | NPV % | LR+ | AUC | Standard Error | 95% CI | P |
|---|---|---|---|---|---|---|---|---|---|---|
| PQ-16 | 6 | 0.81 | 0.44 | 29 | 89 | 1.45 | 0.72 | 0.033 | 0.66–0.78 | <.001 |
| PQ-16 | 7 | 0.73 | 0.55 | 32 | 88 | 1.62 | ||||
| PCA | 3 | 0.95 | 0.13 | 25 | 89 | 1.06 | 0.69 | 0.033 | 0.62–0.75 | <.001 |
| PCA | 4 | 0.90 | 0.26 | 27 | 90 | 1.22 | ||||
| PCA | 5 | 0.83 | 0.44 | 31 | 89 | 1.48 | ||||
| Combined | 10 | 0.89 | 0.43 | 42 | 89 | 1.56 | 0.74 | 0.028 | 0.69–0.80 | <.001 |
Note: PPV is the positive predictive—true positive/(true positive + false positives); NPV is the negative predictive value—true negative/(true negative + false negative). The NPV is very high for all these measures ie, a negative score really is likely to be negative. The PPV is relatively low (above threshold is 1/3 chance of being a genuine positive).