Table 3.
Diagnostic properties of logistic regression models
| Model | Severity of detected PD | Reference standard: dentist’s assessment | Bias-corrected AUC | AUC of the original model applied on radiograph scoring data | ||
|---|---|---|---|---|---|---|
| Sensitivity (%) | Specificity (%) | AUC | ||||
| Age + number of teeth | Mild, moderate or severe versus no | 86.0 | 48.6 | 0.73 | 0.73 | 0.82 |
| Moderate or severe versus no or mild | 80.3 | 64.1 | 0.78 | 0.77 | 0.72 | |
| Severe versus no, mild or moderate | 86.1 | 78.1 | 0.86 | 0.85 | 0.66 | |
| Age + 6 patient-reported items | Mild, moderate or severe versus no | 64.2 | 88.5 | 0.82 | 0.81 | 0.88 |
| Moderate or severe versus no or mild | 72.8 | 80.7 | 0.85 | 0.83 | 0.83 | |
| Severe versus no, mild or moderate | 96.6 | 81.5 | 0.92 | 0.90 | 0.77 | |
Sensitivities, specificities and AUCs to detect different levels of severity of PD in the simple model and in the model including six questionnaire items. The table also shows the AUCs of these models after correction for overoptimism with bootstrap methods and the AUCs of the models if the independent assessment of PD with radiographs is used as a reference standard