Table 4 Diagnostic performance of M3-score for training set (Wave 1) and test set (Wave 2–3) at different M3-score cutoffs and multiple result intervals.

From: AI-derived CT biomarker score for robust COVID-19 mortality prediction across multiple waves and regions using machine learning

Patient cohort

M3-score

Survival

Death

LR (95%CI)

PPOST (%)

Training set (UZ Brussel)

(COVID-19 wave 1)

 n = 252

 

213

39

  

  PPRE (%) = 15.5

     
 

0.00–0.10

99

0

0.00 (0.00-0.44)

0.0

0.10–0.25

43

1

0.13 (0.02–0.90)

2.3

0.25–0.50

32

5

0.85 (0.35–2.05)

13.5

0.50–0.75

23

9

2.14 (1.07–4.27)

28.2

0.75–1.00

16

24

8.19 (4.81-14.00)

60.0

Test set (AZ Delta)

(COVID-19 wave 2–3)

 n = 175

 

152

23

  

  PPRE (%) = 13.1

     
 

0.00–0.10

69

1

0.10 (0.01–0.66) a, b

1.4

0.10–0.25

31

3

0.64 (0.21–1.92) a, c

8.8

0.25–0.50

30

6

1.32 (0.62–2.82) a, c

16.6

0.50–0.75

17

7

2.72 (1.27–5.84) a, c

29.1

0.75–1.00

5

6

7.93 (2.63–23.89) a, c

54.5

  1. LR, Likelihood Ratio; CI, Confidence Interval; PPOST, Posttest Probability; PPRE, Pretest Probability.
  2. a No significant difference compared to the training set (proportional overlap > 0.50 indicating P > 0.05, Cumming G et al.26, see Appendix. Supplementary Materials and Methods.)
  3. b Statistically significant difference (P < 0.05) compared to the training set according to delta method (see Appendix. Supplementary Materials and Methods.)
  4. c No statistically significant difference (P > 0.05) compared to the training set according to delta method (see Appendix. Supplementary Materials and Methods.)