Table 2 Overview of annotated events, Polysomnography’s (PSG), and demographics of cohorts used to train and evaluate ABED

From: Expert-level probabilistic breathing event detector informs phenotyping of sleep apnea

 

EVENT

PSG

DEMOGRAPHIC

COHORT

OA

CA

HYPO

IRE

Train/Val

Test

Total

Age

BMI

(F - M)

SINGLE SCORED

MESA *

32,249

3865

145,119

131,600

1175

249

1424

69.2 ± 9.0

28.8 ± 5.3

54–46

MROS *

97,544

26,919

214,605

198,641

2337

500

2837

76.4 ± 5.5

27.2 ± 3.8

0–100

WSC°

28,964

4753

155,517

0

1317

250

1567

57.3 ± 8.0

31.8 ± 7.3

46–54

CFS *

16,245

2384

74,091

0

627

100

727

41.5 ± 19.4

32.4 ± 9.5

55–45

TOTAL

175,002

37,921

589,332

330,241

5456

1099

6555

67.1 ± 14.6

29.1 ± 2.5

29–71

CONSENSUS SCORED

ALLIANCE ‘’

1521

228

6050

0

0

58

58

51.1 ± 4.2

32.9 ± 9.2

100–0

DREEM *

2166

501

4,026

0

0

55

55

45.6 ± 16.5

29.6 ± 6.4

36–64

TOTAL

3687

729

10,076

0

0

113

113

48.4 ± 11.9

31.3 ± 8.0

69–31

  1. The number of events in ALLIANCE and DREEM are derived as the maximum annotations of the consensus where at least two experts agree on an event.
  2. All single-scored cohorts had one expert scoring per PSG. DREEM had five experts scoring per PSG, and Alliance had four. Criteria used for hypopnea (HYPO) scoring are the recommended, alternative and 4% desaturation HYPO criteria: * HYPOrec, ‘’ HYPOalt, ° HYPO4%, respectively.