Table 8.
Mortality model results for six algorithms using different performance metrics.
| Method and algorithm | Training set accuracy, mean (SD) | Test set accuracy (95% CI) | Test set sensitivity (95% CI) | Test set specificity (95% CI) | Test set NPVa (95% CI) | Test set PPVb (95% CI) | |
| Baseline approach | |||||||
|
|
LRc | 0.8826 (0.0058) | 0.8806 (0.874-0.890) | 0.0331 (0.033-0.034) | 0.9979 (0.991-1.009) | 0.8817 (0.875-0.891) | 0.6923 (0.688-0.700) |
|
|
LDAd | 0.8817 (0.0058) | 0.8788 (0.873-0.888) | 0.0523 (0.052-0.053) | 0.9932 (0.986-1.004) | 0.8833 (0.877-0.893) | 0.5182 (0.515-0.524) |
|
|
RFe | 0.8846 (0.0061) | 0.8854 (0.879-0.895) | 0.1127 (0.112-0.114) | 0.9923 (0.985-1.003) | 0.8898 (0.884-0.899) | 0.6710 (0.666-0.679) |
|
|
kNNf | 0.8765 (0.0054) | 0.8760 (0.870-0.886) | 0.0854 (0.085-0.087) | 0.9855 (0.978-0.996) | 0.8861 (0.880-0.896) | 0.4496 (0.447-0.455) |
|
|
SVMg | 0.8837 (0.0058) | 0.8808 (0.875-0.890) | 0.0272 (0.027-0.028) | 0.9989 (0.992-1.010) | 0.8811 (0.875-0.891) | 0.7872 (0.782-0.796) |
|
|
XGBh | 0.8842 (0.0061) | 0.8815 (0.875-0.891) | 0.1429 (0.142-0.145) | 0.9837 (0.977-0.994) | 0.8923 (0.886-0.902) | 0.5495 (0.546-0.556) |
| Quantiles approach | |||||||
|
|
LR | 0.8838 (0.0063) | 0.8815 (0.875-0.891) | 0.0545 (0.054-0.055) | 0.9960 (0.989-1.007) | 0.8838 (0.878-0.893) | 0.6548 (0.650-0.662) |
|
|
LDA | 0.8821 (0.0067) | 0.8814 (0.875-0.891) | 0.0935 (0.093-0.095) | 0.9905 (0.983-1.001) | 0.8875 (0.881-0.897) | 0.5772 (0.573-0.584) |
|
|
RF | 0.8859 (0.0064) | 0.8861 (0.880-0.896) | 0.0891 (0.089-0.090) | 0.9964 (0.989-1.007) | 0.8876 (0.881-0.897) | 0.7756 (0.770-0.784) |
|
|
KNN | 0.8802 (0.0060) | 0.8764 (0.870-0.886) | 0.0589 (0.059-0.060) | 0.9895 (0.982-1.000) | 0.8836 (0.877-0.893) | 0.4395 (0.437-0.445) |
|
|
SVM | 0.8851 (0.0058) | 0.8820 (0.876-0.892) | 0.0449 (0.045-0.046) | 0.9816 (0.991-1.009) | 0.8829 (0.877-0.893) | 0.7439 (0.739-0.752) |
|
|
XGB | 0.8844 (0.0061) | 0.8822 (0.875-0.891) | 0.1643 (0.164-0.167) | 0.9816 (0.975-0.992) | 0.8945 (0.888-0.904) | 0.5533 (0.550-0.560) |
aNPV: negative predictive value.
bPPV: positive predictive value.
cLR: logistic regression.
dLDA: linear discriminant analysis.
eRF: random forest.
fkNN: k-nearest neighbor.
gSVM: support vector machine.
hXGB: extreme gradient boosting.