Table 4.
Ten-fold cross-validation results for logistic regression (LR), random forest (RF), and generalized boosted regression (GBM) models for predicting diagnosis, management, and severity.
| Classifier | Diagnosis | Management | Severity | |||
|---|---|---|---|---|---|---|
| Sens. (±SD) | Spec. (±SD) | Sens. (±SD) | Spec. (±SD) | Sens. (±SD) | Spec. (±SD) | |
| Random | 0.57 | 0.43 | 0.62 | 0.38 | 0.88 | 0.12 |
| AS or PAS ≥ 4 and appendix diameter ≥ 6 mm | 0.91 | 0.73 | — | — | — | — |
| Suspected diagnosis | 1.00 | 0.46 | — | — | — | — |
| LR (full) | 0.88 (±0.06) | 0.76 (±0.11) | 0.85 (±0.09) | 0.82 (±0.09) | 0.93 (±0.05) | 0.42 (±0.32) |
| LR (w/o US) | 0.75 (±0.06) | 0.72 (±0.09) | 0.92 (±0.07) | 0.85 (±0.05) | 0.95 (±0.04) | 0.52 (±0.29) |
| LR (w/o peritonitis/abdominal guarding) | 0.87 (±0.07) | 0.76 (±0.12) | 0.84 (±0.10) | 0.68 (±0.15) | 0.94 (±0.05) | 0.40 (±0.36) |
| LR (w/o US and peritonitis/abdominal guarding) | 0.77 (±0.06) | 0.67 (±0.11) | 0.82 (±0.06) | 0.63 (±0.07) | 0.97 (±0.05) | 0.44 (±0.34) |
| RF (full) | 0.91 (±0.03) | 0.86 (±0.08) | 0.94 (±0.07) | 0.80 (±0.09) | 0.98 (±0.02) | 0.45 (±0.16) |
| RF (w/o US) | 0.81 (±0.07) | 0.71 (±0.07) | 0.93 (±0.07) | 0.82 (±0.07) | 0.97 (±0.02) | 0.44 (±0.13) |
| RF (w/o peritonitis/abdominal guarding) | 0.91 (±0.04) | 0.90 (±0.06) | 0.86 (±0.07) | 0.65 (±0.18) | 0.98 (±0.02) | 0.37 (±0.17) |
| RF (w/o US and peritonitis/abdominal guarding) | 0.79 (±0.06) | 0.64 (±0.11) | 0.81 (±0.06) | 0.56 (±0.06) | 0.98 (±0.02) | 0.40 (±0.15) |
| GBM (full) | 0.93 (±0.02) | 0.86 (±0.07) | 0.93 (±0.07) | 0.86 (±0.07) | 0.97 (±0.02) | 0.46 (±0.18) |
| GBM (w/o US) | 0.80 (±0.07) | 0.74 (±0.11) | 0.91 (±0.08) | 0.85 (±0.05) | 0.97 (±0.03) | 0.44 (±0.16) |
| GBM (w/o peritonitis/abdominal guarding) | 0.92 (±0.04) | 0.83 (±0.09) | 0.88 (±0.04) | 0.66 (±0.11) | 0.97 (±0.03) | 0.47 (±0.20) |
| GBM (w/o US and peritonitis/abdominal guarding) | 0.80 (±0.06) | 0.61 (±0.10) | 0.82 (±0.07) | 0.59 (±0.09) | 0.97 (±0.03) | 0.47 (±0.19) |
Results are given by average sensitivities (sens.) and specificities (spec.) with standard deviations across 10 folds. “Full” models use all predictors; models “w/o US” were trained without ultrasonographic findings; models “w/o peritonitis/abdominal guarding” were trained without the “peritonitis/abdominal guarding” predictor; and models “w/o US and peritonitis/abdominal guarding” were trained without ultrasonographic findings or the “peritonitis/abdominal guarding” predictor. For all classifiers, a probability threshold of 0.5 was used to differentiate between classes. “Random” corresponds to a random guess and serves as a naïve baseline. Bold values correspond to the best average performances achieved across all models.