Table 4. Performance characteristics for the SVM models.
Train (n = 147,799; 64%) |
Test (n = 36,950; 16%) |
|||||||
---|---|---|---|---|---|---|---|---|
AUC | PPV | Sensitivity | Specificity | AUC | PPV | Sensitivity | Specificity | |
Vitals | 0.67 | 0.22 | 0.56 | 0.67 | 0.67 | 0.22 | 0.56 | 0.68 |
CC | 0.84 | 0.34 | 0.78 | 0.75 | 0.83 | 0.32 | 0.75 | 0.75 |
BoW | 0.89 | 0.40 | 0.83 | 0.80 | 0.86 | 0.38 | 0.78 | 0.79 |
Topics | 0.86 | 0.34 | 0.81 | 0.75 | 0.85 | 0.34 | 0.80 | 0.75 |
Vitals—Age, Gender, Severity, Temperature, Heart Rate, Respiratory Rate, Oxygen Saturation, Systolic Blood Pressure, Diastolic Blood Pressure, Pain Scale CC—Chief Complaint + Vitals BoW—Bag of Words model using Vitals + Chief Complaint + Triage Assessment Topics—Topic Model using Vitals + Chief Complaint + Triage Assessment * All Test AUCs have a 95% CI of +-0.02. A validation data set (n = 46,187; 20%) was also used as an intermediary data set between train and test to select regularization parameters. We do not show these here, for brevity.