Skip to main content

Table 3 Performance metrics of models predicting days on ventilator using pCXR alone, non-imaging data alone and their combination

From: Deep learning of longitudinal chest X-ray and clinical variables predicts duration on ventilator and mortality in COVID-19 patients

  

Slope

Intercept

R2

p

MAE

Day 1

CXR

0.19 (0.08)

8.5 (1.1)

0.18 (0.09)

0.059

5.30 (0.42)

Non-imaging variables

0.27 (0.10)

11.0 (1.3)

0.50 (0.14)

 < 0.001

4.67 (0.33)

CXR + non-imaging variables

0.07 (0.02)

11.0 (2.1)

0.11 (0.04)

0.120

4.54 (0.36)

Day 5

CXR

0.32 (0.08)

8.03 (1.9)

0.32 (0.11)

0.002

5.01 (0.44)

Non-imaging variables

0.40 (0.12)

7.94 (1.7)

0.25 (0.19)

 < 0.001

4.88 (0.38)

CXR + non-imaging variables

0.41 (0.15)

6.73 (2.1)

0.37 (0.20)

0.008

4.21 (0.56)

Day 1–3

CXRs

0.51 (0.13)

6.49 (1.3)

0.58 (0.11)

 < 0.001

3.41 (0.32)

Non-imaging variables

0.62 (0.18)

3.76 (1.1)

0.54 (0.12)

 < 0.001

3.13 (0.35)

CXR + non-imaging variables

0.47 (0.12)

7.69 (1.2)

0.53 (0.11)

 < 0.001

2.96 (0.33)

Day 3–5

CXRs

0.57 (0.16)

8.85 (1.6)

0.52 (0.12)

 < 0.001

3.14 (0.57)

Non-imaging variables

0.62 (0.17)

5.43 (1.1)

0.50 (0.15)

 < 0.001

3.11 (0.32)

CXR + non-imaging variables

0.59 (0.17)

6.79 (1.3)

0.51 (0.13)

 < 0.001

3.05 (0.41)

Day 1–5

CXRs

0.60 (0.13)

6.55 (1.1)

0.80 (0.18)

 < 0.001

3.11 (0.25)

Non-imaging variables

0.62 (0.12)

6.26 (1.0)

0.69 (0.18)

 < 0.001

2.88 (0.25)

CXR + non-imaging variables

0.69 (0.10)

3.52 (0.7)

0.66 (0.15)

 < 0.001

2.56 (0.24)

  1. Values in parentheses are standard deviations
  2. MAE mean absolute error