Table 3.
Two-by-two table of prediction models’ validity at selected cut-off.
| “Optimum Data Availability” Cut-off: 0.217 | “Scottish Data” Cut-off: 0.226 | |||||||
|---|---|---|---|---|---|---|---|---|
| N = 2118 | True (observed) outcome | N = 2279 | True (observed) outcome | |||||
| Obese | Non-obese | TOTALS | Obese | Non-obese | TOTALS | |||
| Predicted outcome | Obese | 297 | 387 | 784 | Obese | 314 | 388 | 702 |
| Non-obese | 92 | 1342 | 1434 | Non-obese | 98 | 1479 | 1577 | |
| TOTALS | 389 (Obese) | 1729 (Non-Obese) | TOTALS | 412 (Obese) | 1867 (Non-Obese) | |||
| Sensitivity: | 297/389 = | 76.3% | Sensitivity: | 314/412= | 76.2% | |||
| Specificity: | 1342/1729 = | 77.6% | Specificity: | 1479/1867= | 79.2% | |||
| PPV | 297/784= | 37.8% | PPV: | 314/702= | 44.7% | |||
| NPV | 1342/1434= | 93.6% | NPV: | 1479/1577= | 93.8% | |||
| Referral Burden: | 784/2118 = | 37.0% | Referral Burden: | 702/2279 = | 30.8% | |||
Of the 387 “false positives” from the Optimal data model 160 (41.4%) were overweight, so that those unlikely to benefit at all would be 227 or 10.7% of all those screened; of the 388 “false positives” from Scottish data model 161 (41.5%) were overweight, so that those unlikely to benefit at all would be 227 or 10.0% of all those screened.
PPV positive predictive value, NPV negative predictive value.