Table III.
Performance statistics for each model in both the training and test datasets
| Statistic | SSI | UTI | Sepsis | Pneumonia | ||||
|---|---|---|---|---|---|---|---|---|
|
|
|
|
|
|
||||
| Data set | Training | Test | Training | Test | Training | Test | Training | Test |
|
| ||||||||
| NSQIP prevalence | 784 (3.7%) | 285 (3.1%) | 322 (1.5%) | 144 (1.6%) | 584 (2.7%) | 184 (2.0%) | 222 (1.0%) | 63 (0.7%) |
| Model prevalence | 3.7% | 3.3% | 1.5% | 1.7% | 2.7% | 2.5% | 1.0% | 0.9% |
| FDR level | 0.175 | NA | 0.1 | NA | 0.125 | NA | 0.125 | NA |
| No. of predictors selected | 4 | NA | 7 | NA | 7 | NA | 8 | NA |
| Threshold for classification* | 2.9 | 2.9 | 1.1 | 1.1 | 4.5 | 4.5 | 1.6 | 1.6 |
| AUC | 0.91 | 0.91 | 0.91 | 0.93 | 0.95 | 0.95 | 0.96 | 0.96 |
| Specificity* | 81 (80, 81) | 82 (81,82) | 90 (90,91) | 90 (89,91) | 89 (89, 90) | 89 (88, 90) | 95 (94, 95) | 95 (94, 95) |
| Sensitivity* | 89 (87, 91) | 89 (85, 93) | 82 (78, 86) | 90 (85, 94) | 89 (87, 92) | 93 (90, 97) | 88, (84, 92) | 87 (79, 95) |
| Accuracy* | 81 (81, 82) | 82 (81,83) | 90 (89,91) | 90 (89,91) | 89 (89, 90) | 89 (88, 90) | 94 (94, 95) | 95 (94, 95) |
| NPV* | 99 (99, 99) | 99 (99, 99) | 99 (99, 99) | 99 (99, 99) | 99 (99, 99) | 99 (99, 99) | 99 (99, 99) | 99 (99, 99) |
| PPV* | 15 (15, 15) | 13 (13, 14) | 11 (10, 11) | 12 (11, 13) | 19(18, 20) | 15 (14, 16) | 14(13, 15) | 10(9, 11) |
AUC, area under the curve; FDR, false discovery rate; NA, not applicable; NPV, negative predictive value; NSQIP, National Surgical Quality Improvement Program; PPV, positive predictive value; SSI, surgical site infection; UTI, urinary tract infection.
*
Values multiplied by 100.