Skip to main content

Table 4 Real data results. Mean and coefficient of variation (in parenthesis) of precision (when recall = 0.7), PPV (when NPV = 0.95 for Readmit 30 and 0.99 for the others), AUC, NO. (number of features in predictive models) of five prediction models in testing set in 100 repetitions

From: The application of unsupervised deep learning in predictive models using electronic health records

ResponsePrediction
Models
Precision (%)
(recall = 0.7)
PPV (%)
(NPV = 0.95/0.99)
AUCNO.
Readmit30Autoencoder19.04 (0.02)16.88 (0.02)0.707 (0.01)200 (0.00)
LASSO (λmin)19.70 (0.02)17.79 (0.02)0.719 (0.00)162 (0.10)
Random Forest18.48 (0.02)16.50 (0.02)0.707 (0.01)469 (0.00)
Simple Reg18.70 (0.02)16.06 (0.02)0.700 (0.01)25 (0.00)
Enhanced Reg19.69 (0.02)17.68 (0.02)0.717 (0.00)144 (0.10)
COPDAutoencoder55.90 (0.02)42.16 (0.03)0.961 (0.00)200 (0.00)
LASSO (λmin)58.02 (0.02)44.50 (0.02)0.963 (0.00)266 (0.04)
Random Forest56.19 (0.02)40.45 (0.03)0.956 (0.00)469 (0.00)
Simple Reg51.51 (0.02)35.53 (0.03)0.952 (0.00)21 (0.00)
Enhanced Reg57.06 (0.02)43.62 (0.02)0.962 (0.00)161 (0.08)
AMIAutoencoder57.40 (0.04)68.80 (0.03)0.985 (0.00)200 (0.00)
LASSO (λmin)58.57 (0.04)70.10 (0.04)0.986 (0.00)64 (0.59)
Random Forest56.32 (0.03)65.90 (0.03)0.982 (0.00)469 (0.00)
Simple Reg52.24 (0.04)56.43 (0.06)0.984 (0.00)11 (0.00)
Enhanced Reg59.26 (0.04)70.66 (0.03)0.986 (0.00)129 (0.14)
Heart FailureAutoencoder61.48 (0.02)43.94 (0.02)0.961 (0.00)200 (0.00)
LASSO (λmin)63.15 (0.02)45.88 (0.02)0.964 (0.00)195 (0.08)
Random Forest60.67 (0.02)42.56 (0.02)0.958 (0.00)469 (0.00)
Simple Reg57.81 (0.02)38.50 (0.02)0.954 (0.00)18 (0.00)
Enhanced Reg62.37 (0.02)45.09 (0.02)0.962 (0.00)158 (0.10)
PneumoniaAutoencoder40.17 (0.03)34.56 (0.03)0.955 (0.00)200 (0.00)
LASSO (λmin)42.18 (0.03)35.94 (0.02)0.958 (0.00)204 (0.09)
Random Forest38.27 (0.03)32.44 (0.03)0.951 (0.00)469 (0.00)
Simple Reg32.44 (0.02)28.76 (0.02)0.942 (0.00)11 (0.00)
Enhanced Reg41.39 (0.03)35.54 (0.02)0.957 (0.00)173 (0.08)