Table 1 Mean absolute error of age estimation models.

From: Age estimation from sleep studies using deep learning predicts life expectancy

 

MAE

   

Model

Train set n = 2500

Val set n = 200

Test set n = 10,509

HomePAP* n = 190

Basic sleep measures

14.9 ± 6.08

14.9 ± 6.53

14.6 ± 5.91

12.5 ± 4.06

(a) Central EEG

5.43 ± 1.25

6.52 ± 2.48

6.77 ± 2.2

7.65 ± 2.7

(b) EEG+EOG+EMG

5.35 ± 0.96

5.88 ± 2.09

6.81 ± 1.84

8.62 ± 2.92

(c) ECG

9.11 ± 1.89

11 ± 4.05

10.4 ± 2.23

13.9 ± 6.74

(d) Respiratory

8.87 ± 2.2

9.31 ± 2.39

8.09 ± 1.89

13.7 ± 6.05

(e) Ensemble–Avg.

5.4 ± 1.01

6.11 ± 1.84

5.8 ± 1.16

8.16 ± 3.75

  1. The MAE is reported as mean ± standard deviation and was averaged across age intervals ([20, 25], [25, 30], …, [85–89]), which are reported for the test and HomePAP set in Supplementary Tables 2 and 3. *The training and validation set includes no PSGs from the HomePAP study, thus it represents expected performance in a new unseen cohort with a different technical setup. Basic sleep measures denote a linear regression model with the following predictive variables: arousal index, apnea-hypopnea index, total sleep time, wake after sleep onset, and percentage of N1, N2, N3, and REM sleep. MAE: mean absolute error.
Back to article page