Table 2. Quantitative Analysis of Model Performance .
| Variable | SHC Data | Duke Data | ||
|---|---|---|---|---|
| AUROC | P Value | AUROC | P Value | |
| Holdout Testing on Internal and External Data Set | ||||
| ElasticNet model | 0.93 | .01 | 0.70 | .17 |
| PE neural model | 0.85 | 0.72 | ||
| Comparison With Clinical Scoring Systems on Consecutive Outpatients | ||||
| Machine learning models | ||||
| ElasticNet model | 0.73 | .42 | 0.74 | .01 |
| PE neural model | 0.81 | 0.81 | ||
| Clinical scoring | ||||
| Wells | 0.48 | NA | 0.51 | NA |
| PERC | 0.51 | 0.60 | ||
| rGeneva | 0.53 | 0.47 | ||
Abbreviations: AUROC, area under the receiver operating characteristic curve; Duke, Duke University Medical Center; NA, not applicable; PE, pulmonary embolism; PERC, Pulmonary Embolism Rule-out Criteria; rGeneva, revised Geneva; SHC, Stanford University hospital and clinics.