Table 4 Final multivariate stepwise binary logistic regression models elaborated with either radiological or clinical-radiological variables.

From: Practical clinical and radiological models to diagnose COVID-19 based on a multicentric teleradiological emergency chest CT cohort

Variables§

Coefficients (βi)

P-value

Multivariate OR (95%CI)

Examples§§

A

B

C

Radiological model

0. (Intercept)

− 2.758126351

–

0.06 (0.03–0.12)

1

1

1

1. Presence of GGO

1.089578460

0.04*

2.97 (1.03–8.56)

0

1

1

2. Fibrotic band

1.5310410593

 < 0.001***

4.62 (2.11–10.77)

1

0

1

3. GGO predominant pattern

1.5366549953

 < 0.001***

4.65 (1.96–11.45)

0

1

1

4. Subpleural predominant distribution

0.9053515712

0.004**

2.47 (1.33–4.63)

0

1

1

5. Diffuse lesions

0.8414110324

0.02*

2.32 (1.16–4.61)

0

0

1

6. Intralobular reticulations

0.7052466524

0.03*

2.02 (1.05–3.91)

0

0

1

7. Bronchial wall thickening

− 1.758954141

 < 0.001***

0.17 (0.07–0.4)

1

0

0

Probablity for RT-PCR + §§

4.8%

68.4%

97.9%

Clinical-radiological model

0. (Intercept)

− 4.805582918

–

–

1

1

1

1. Fever

1.982642859

 < 0.001***

7.26 (2.82–20.41)

1

1

1

2. Myalgia

0.899937144

0.08

2.46 (0.82–6.94)

1

0

0

3. Presence of GGO

1.181705996

0.08

3.26 (0.86–12.54)

1

1

0

4. Presence of fibrotic band

1.322814497

0.02*

3.75 (1.34–11.47)

1

0

1

5. GGO predominant pattern

1.504660153

0.008*

4.5 (1.52–14.24)

1

1

0

6. Peripheral predominant distribution

0.930711612

0.03*

2.54 (1.12–5.83)

1

1

0

7. Diffuse lesions

1.413720338

0.003**

4.11 (1.66–10.58)

1

0

0

8. Intralobular reticulations

1.028166435

0.02*

2.8 (1.2–6.66)

1

0

0

9. Bronchial wall thickening

− 1.716180719

0.002**

0.18 (0.06–0.51)

0

0

1

Probablity for RT-PCR + §§§

99.6%

68.9%

3.9%

  1. Significant results are highlighted in bold.
  2. 95%CI 95% confidence interval; GGO ground glass opacity; OR odds ratio.
  3. §The variables correspond to those in the final model after the stepwise backward-forward selection.
  4. §§Examples correspond to 6 distinct clinical cases. Each variable has 2 levels: “1” if the variable Xi is present (for instance fever), and “0” if the variable Xi is absent (for instance lack of fever).
  5. §§§The probability for RT-PCR + are calculated as follows:
  6. P(RT-PCR +) = \( 1/1 + {\text{~exp}}[ - (\beta _{{0~}} + \mathop \sum \nolimits_{{i = 1}}^{9} \beta _{i} \times X_{i} )] \).
  7. *P < 0.05, **P < 0.005, ***P < 0.001.
Back to article page