Skip to main content

Table 2 Evaluation of different deep learning models, using KFRE as comparison

From: Deep learning algorithms for predicting renal replacement therapy initiation in CKD patients: a retrospective cohort study

 

ROC AUC

F1 score

Sensitivity

Specificity

PPV

NPV

CNN

0.91 (0.907—0.914)

0.79 (0.781—0.795)

0.47 (0.465—0.474)

0.97 (0.973—0.975)

0.90 (0.895—0.906)

0.79 (0.784—0.787)

CNN + LSTM + ANN

0.90 (0.896—0.902)

0.76 (0.758—0.769)

0.74 (0.733—0.744)

0.88 (0.879—0.885)

0.76 (0.752—0.764)

0.87 (0.868—0.874)

ANN

0.88 (0.876—0.882)

0.76 (0.756—0.767)

0.71 (0.706—0.719)

0.87 (0.870—0.876)

0.74 (0.732—0.743)

0.86 (0.856—0.862)

ConvLSTM + ANN

0.88 (0.875—0.881)

0.76 (0.755—0.763)

0.71 (0.707—0.715)

0.87 (0.870—0.874)

0.73 (0.732—0.740)

0.86 (0.856—0.860)

LSTM

0.85 (0.844—0.852)

0.68 (0.673—0.684)

0.50 (0.493—0.505)

0.88 (0.875—0.881)

0.67 (0.664—0.679)

0.78 (0.776—0.781)

KFRE (4 variable)

0.84 (0.841—0.842)

0.32 (0.313—0.319)

0.40 (0.395—0.405)

0.75 (0.747—0.754)

0.42 (0.410—0.429)

0.88 (0.870—0.882)

KFRE (8 variable)

0.84 (0.836—0.837)

0.40 (0.394—0.406)

0.40 (0.395—0.405)

0.75 (0.747—0.754)

0.42 (0.410—0.429)

0.88 (0.870—0.882)