Table 2.
Activities of daily living classifier performance. Italic values represent the optimal performance in each data set.
| Model | AUROCa (95% CI) | AUPRCb (95% CI) | ||
| Filtered cohort | ||||
| Training set | ||||
| Deep learning | 0.952 (0.945-0.960) | 0.864 (0.844-0.881) | ||
| Bio+Clinical BERTc | 0.870 (0.842-0.897) | 0.826 (0.789-0.862) | ||
| Logistic regression | 0.955 (0.949-0.962) | 0.855 (0.837-0.872) | ||
| LASSOd | 0.958 (0.952-0.965) | 0.865 (0.846-0.883) | ||
| Random forest | 0.953 (0.946-0.960) | 0.857 (0.838-0.875) | ||
| SVMe | 0.954 (0.946-0.960) | 0.854 (0.835-0.872) | ||
| XGBoost | 0.955 (0.948-0.962) | 0.848 (0.826-0.869) | ||
| Validation set | ||||
| Deep learning | 0.961 (0.951-0.971) | 0.880 (0.852-0.906) | ||
| Bio+Clinical BERT | 0.873 (0.852-0.891) | 0.847 (0.823-0.869) | ||
| Logistic regression | 0.963 (0.954-0.971) | 0.871 (0.841-0.896) | ||
| LASSO | 0.962 (0.954-0.970) | 0.870 (0.841-0.896) | ||
| Random forest | 0.971 (0.964-0.977) | 0.887 (0.859-0.913) | ||
| SVM | 0.963 (0.954-0.971) | 0.877 (0.849-0.902) | ||
| XGBoost | 0.961 (0.951-0.969) | 0.873 (0.846-0.898) | ||
| Unfiltered validation cohort | ||||
| Deep learning | 0.991 (0.986-0.994) | 0.817 (0.746-0.882) | ||
| Bio+Clinical BERT | 0.785 (0.582-0.999) | 0.621 (0.227-0.901) | ||
| Logistic regression | 0.981 (0.971-0.990) | 0.737 (0.644-0.817) | ||
| LASSO | 0.969 (0.954-0.983) | 0.675 (0.573-0.769) | ||
| Random forest | 0.990 (0.984-0.995) | 0.806 (0.723-0.880) | ||
| SVM | 0.986 (0.975-0.994) | 0.822 (0.748-0.887) | ||
| XGBoost | 0.978 (0.959-0.992) | 0.771 (0.680-0.846) | ||
aAUROC: area under the receiver operating characteristic curve.
bAUPRC: area under the precision-recall curve.
cBERT: bidirectional encoder representations from transformers.
dLASSO: least absolute shrinkage and selection operator.
eSVM: support vector machine.