Table 4. Test characteristics of UTI prediction models on validation data*.
| Models | AUC (95%CI) | Sensitivity (95% CI) | Specificity (95% CI) |
+LR (95% CI) | -LR (95% CI) | Accuracy (95% CI) | P–value |
|---|---|---|---|---|---|---|---|
| XGBoost | .904(.898-.910) | 61.7(60.0–63.3) | 94.9 (94.5–95.3) | 12.0(11.1–13.0) | .404(.387-.421) | 87.5 (87.0–88.0) | NA |
| Random Forest | .902(.896-.908) | 57.3(55.6–58.9) | 96.0 (95.6–96.3) | 14.3(13.0–15.6) | .445(.428-.462) | 87.4 (86.9–87.9) | 0.58 |
| Adaboost | .880(.874-.887) | 62.2(60.6–63.8) | 92.3(91.8–92.7) | 8.06(7.54–8.61) | .409(.392-.427) | 85.6(85.1–86.2) | < .001 |
| Support Vector Machine | .878(.871-.884) | 49.6(47.9–51.2) | 96.8(96.4–97.1) | 15.3(13.8–16.9) | .521(.504-.538) | 86.3(85.7–86.8) | < .001 |
| ElasticNet | .892(.885-.898) | 56.8(55.2–58.4) | 94.9(94.5–95.2) | 11.1(10.2–12.0) | .455(.438-.473) | 86.4(85.9–87.0) | < .001 |
| Logistic Regression | .891 (.884-.897) | 57.5(55.8–59.1) | 94.7(94.3–95.1) | 10.9(10.0–11.8) | .449(.432-.466) | 86.4(85.9–87.0) | < .001 |
| Neural Network | .884 (.878-.890) | 54.6(52.9–56.2) | 95.3(95.0–95.7) | 11.7(10.8–12.8) | .476(.460-.494) | 86.3(85.8–86.8) | <001 |
| Reduced XGBoost | .877(.871-.884) | 54.7(53.0–56.3) | 94.7(94.3–95.1) | 10.4(9.6–11.3) | .479(.462-.496) | 85.9(85.3–86.4) | < .001 |
| Reduced Random Forest | .861(.853-.868) | 54.8(53.1–56.4) | 94.3(93.9–94.7) | 9.66(8.94–10.4) | .479(.462-.497) | 85.5(85.0–86.1) | < .001 |
| Reduced Adaboost | .826(.817-.834) | 61.9(60.3–63.5) | 88.8(88.2–89.3) | 5.50(5.21–5.81) | .429(.412-.448) | 82.8(82.2–83.3) | < .001 |
| Reduced Support Vector Machine | .822(.813-.832) | 49.4(47.8–51.1) | 95.8(95.4–96.1) | 11.7(10.7–12.9) | .528(.511-.546) | 85.5(84.9–86.0) | < .001 |
| Reduced Elastic Net | .870(.863-.877) | 52.4(50.7–54.1) | 95.2(94.8–95.5) | 10.9(9.99–11.8) | .500(.482-.571) | 85.7(85.1–86.2) | < .001 |
| ReducedLogistic Regression | .870(.863-.877) | 53.3(51.6–54.9) | 94.8(94.4–95.2) | 10.3(9.52–11.2) | .492(.476-.510) | 85.6(85.0–86.2) | < .001 |
| Reduced Neural Network | .873(.867-.881) | 54.0(52.3–55.6) | 95.0(94.6–95.4) | 10.9(10.0–11.8) | .485(.468-.502) | 85.9(85.4–86.5) | < .001 |
* Test Characteristics determined at optimal AUC threshold
Full models were developed on 212 variables, while the reduced models were developed on 10 variables.
P-values obtained by AUC comparison to best performing model