A Broad Learning System to Predict the 28-Day Mortality of Patients Hospitalized with Community-Acquired Pneumonia: A Case-Control Study

Table 2

The predictive performance of the models for the 28-day mortality of CAP patients.

Prediction models	Accuracy	Sensitivity	Specificity	PPV	NPV
Training set
BLS-RF	0.912 (0.891-0.932)	0.970 (0.929-1.000)	0.906 (0.884-0.928)	0.512 (0.425-0.599)	0.997 (0.992-1.000)
BLS-XGB	0.959 (0.944-0.973)	0.881 (0.803-0.958)	0.967 (0.953-0.980)	0.728 (0.632-0.825)	0.988 (0.979-0.996)
DNN	0.941 (0.924-0.958)	0.926 (0.864-0.989)	0.942 (0.924-0.960)	0.624 (0.529-0.718)	0.992 (0.985-0.999)
CNN	0.930 (0.911-0.948)	0.941 (0.885-0.997)	0.929 (0.909-0.948)	0.577 (0.485-0.668)	0.993 (0.987-1.000)
Logistic	0.789 (0.758-0.818)	0.941 (0.885-0.997)	0.774 (0.742-0.806)	0.300 (0.239-0.362)	0.992 (0.985-1.000)
RF	0.791 (0.761-0.820)	0.853 (0.769-0.937)	0.784 (0.753-0.816)	0.290 (0.227-0.353)	0.981 (0.969-0.993)
Testing set
BLS-RF	0.872 (0.842-0.902)	0.925 (0.853-0.996)	0.865 (0.833-0.898)	0.458 (0.364-0.552)	0.989 (0.979-1.000)
BLS-XGB	0.932 (0.909-0.954)	0.717 (0.596-0.838)	0.958 (0.939-0.977)	0.679 (0.565-0.801)	0.965 (0.948-0.982)
DNN	0.903 (0.877-0.929)	0.808 (0.701-0.915)	0.914 (0.888-0.941)	0.532 (0.422-0.642)	0.975 (0.960-0.990)
CNN	0.897 (0.890-0.924)	0.769 (0.655-0.884)	0.912 (0.885-0.939)	0.513 (0.402-0.624)	0.938 (0.910-0.966)
Logistic	0.779 (0.739-0.815)	0.885 (0.798-0.971)	0.766 (0.726-0.806)	0.313 (0.238-0.388)	0.982 (0.968-0.996)
RF	0.756 (0.718-0.794)	0.615 (0.483-0.748)	0.773 (0.734-0.813)	0.246 (0.172-0.320)	0.944 (0.919-0.968)

CAP: community-acquired pneumonia; BLS: broad learning system; RF: random forest; XGB: eXtreme Gradient Boosting; DNN: deep neural network; CNN: convolutional neural network; PPV: positive predictive value; NPV: negative predictive value.