Table 3.
Comparison of methicillin-resistant Staphylococcus aureus (MRSA) screening strategies using different predictive models
| Cut-off for screening | No. of patients to be screened (%) | No. of patients MRSA positive | Sensitivity (%) | Specificity (%) | PPV (%) | NPV (%) |
|---|---|---|---|---|---|---|
| Universal screening | 1,451 (100) | 49 | 100 | 0 | 3.4 | - |
| Predicted probability ≥ 2% | ||||||
| Stepwise model | 1,213 (83.6) | 46 | 93.9 | 16.8 | 3.8 | 98.7 |
| Best BMA model | 1,451 (100) | 49 | 100 | 0 | 3.4 | - |
| BMA model | 1,437 (99.0) | 49 | 100 | 1.0 | 3.4 | 100 |
| Simple model | 1,201 (82.8) | 47 | 95.9 | 17.7 | 3.9 | 99.2 |
| Predicted probability ≥ 3% | ||||||
| Stepwise model | 813 (56.0) | 37 | 75.5 | 44.7 | 4.6 | 98.1 |
| Best BMA model | 528 (36.4) | 34 | 69.4 | 64.8 | 6.4 | 98.4 |
| BMA model | 1,137 (78.4) | 44 | 89.8 | 22.0 | 3.9 | 98.4 |
| Simple model | 776 (53.5) | 39 | 79.6 | 47.4 | 5.0 | 98.5 |
| Predicted probability ≥ 4% | ||||||
| Stepwise model | 482 (33.2) | 30 | 61.2 | 67.8 | 6.2 | 98.0 |
| Best BMA model | 528 (36.4) | 34 | 69.4 | 64.8 | 6.4 | 98.4 |
| BMA model | 463 (31.9) | 31 | 63.3 | 69.2 | 6.7 | 98.2 |
| Simple model | 479 (33.0) | 33 | 67.3 | 68.2 | 6.9 | 98.4 |
| Predicted probability ≥ 5% | ||||||
| Stepwise model | 354 (24.4) | 26 | 53.1 | 76.6 | 7.3 | 97.9 |
| Best BMA model | 167 (11.5) | 19 | 38.8 | 89.4 | 11.4 | 97.7 |
| BMA model | 133 (9.2) | 16 | 32.7 | 91.7 | 12.0 | 97.5 |
| Simple model | 336 (23.2) | 27 | 55.1 | 78.0 | 8.0 | 98.0 |
| Predicted probability ≥ 6% | ||||||
| Stepwise model | 229 (15.8) | 18 | 36.7 | 85.0 | 7.9 | 97.5 |
| Best BMA model | 126 (8.7) | 17 | 34.7 | 92.2 | 13.5 | 97.6 |
| BMA model | 64 (4.4) | 6 | 12.2 | 95.9 | 9.4 | 96.9 |
| Simple model | 220 (15.2) | 20 | 40.8 | 85.7 | 9.1 | 97.6 |
Note. The table shows the results when a random sample of 50% of the full cohort was used as the derivation dataset with the remaining data used as the validation dataset. MRSA, methicillin-resistant Staphylococcus aureus; NPV, negative predictive value; PPV, positive predictive value.