Table 2 Confusion matrices for each stage of validation

From: Development and validation of a clinical wearable deep learning based continuous inhospital deterioration prediction model

A) Device #1 Retrospective

Predicted Label

 
 

Thresh = 0.0303

No Alert − 0

Alert − 1

 

True Label

No Alert − 0

14308

11889

Specificity: 0.55

Alert − 1

213

3317

Sensitivity: 0.94

  

NPV: 0.99

PPV: 0.22

 

B) Device #1 Prospective

Predicted Label

 
 

Thresh = 0.0352

No Alert − 0

Alert − 1

 

True Label

No Alert − 0

2653

2128

Specificity: 0.55

Alert − 1

14

480

Sensitivity: 0.97

  

NPV: 0.99

PPV: 0.18

 

C) Alternate Device #2

Predicted Label

 
 

Thresh = 0.0291

No Alert − 0

Alert − 1

 

True Label

No Alert − 0

33957

4633

Specificity: 0.88

Alert − 1

280

597

Sensitivity: 0.68

  

NPV: 0.99

PPV: 0.11

 
  1. NPV negative predictive value, PPV positive predictive value.