Table 2.
Performance and clinical usefulness of neonatal cholestasis-related mortality (NCM) prediction model. The NCM model was derived from complete-case dataset by logistic regression analysis.
| Prediction model | Dataset | Performance |
Clinical usefulness |
||||||
|---|---|---|---|---|---|---|---|---|---|
| Brier score | AUC | Calibration slope | Threshold | Sensitivity | Specificity | PPV | NPV | ||
| Logistic regression | |||||||||
| Complete-case | 0.072 | 0.916 | 1.04 | 16.1% | 86.9% | 84.6% | 47.3% | 97.6% | |
| Whole-case | 0.052 | 0.937 | 1.05 | 13.7% | 86.9% | 88.4% | 45.1% | 98.4% | |
| Alternative analyses | |||||||||
| CART* | Complete-case | 0.071 | 0.89 | 1.01 | 9% | 87.7% | 78.6% | 39.5% | 98.6% |
| Whole-case* | 0.052 | 0.89 | 1.02 | 9% | 82.4% | 87.8% | 42.6% | 97.8% | |
| CHAID* | Complete-case | 0.071 | 0.916 | 1.06 | 14% | 87.4% | 81% | 42.4% | 97.6% |
| Whole-case* | 0.052 | 0.936 | 1.08 | 7% | 94.2% | 77.8% | 31.7% | 99.2% | |
| Random forest⁎⁎ | Complete-case | 0.067 | 0.921 | 1.03 | 18.9% | 84% | 87.8% | 50.9% | 97.3% |
| Whole-case | 0.046 | 0.954 | 1.09 | 14.5% | 89% | 90.3% | 49.8% | 98.6% | |
| XGBoost⁎⁎ | Complete-case | 0.074 | 0.919 | 0.82 | 5.6% | 84.9% | 85.6% | 47.4% | 97.4% |
| Whole-case | 0.049 | 0.951 | 0.93 | 11% | 86.6% | 90.9% | 51% | 98.4% | |
⁎
Missing values were not imputed
⁎⁎
Values of performances were derived from the test dataset (See Supplementary Figure 7).
AUC; area under the curve, CART; classification and regression tree, CHAID; chi-square automatic interaction detection, ML; machine learning PPV; positive predictive value, NPV; negative predictive value, XGBoost; extreme gradient boost.