Table 2.
Bland-Altman statistics for wake after sleep onset (WASO) comparison between PSG and the sleep algorithms applied to wrist-based acceleration.
| ME | RMSE | Correlation | CI-95%+ | CI-95%- | CI-width | |
|---|---|---|---|---|---|---|
| Raw Acceleration | ||||||
| van Hees | 39.3 | 91.1 | 0.786 | 201.9 | −123.4 | 325.3 |
| Random Forest | 1.0 | 93.0 | 0.717 | 184.9 | −183.0 | 367.9 |
| Deep Learning Count | ||||||
| CNN-100 | 11.9 | 99.8 | 0.712 | 207.9 | −184.1 | 392.0 |
| LSTM-50 | 0.7 | 91.6 | 0.743 | 181.9 | −180.5 | 362.4 |
| LSTM-100 | 2.6 | 93.9 | 0.736 | 188.3 | −183.1 | 371.4 |
| Legacy Count | ||||||
| Oakley rsc | −10.4 | 81.7 | 0.795 | 149.9 | −170.7 | 320.6 |
| Sadeh | 33.6 | 93.6 | 0.756 | 206.3 | −139.0 | 345.3 |
| Cole-Kripke | 28.4 | 90.4 | 0.770 | 198.1 | −141.3 | 339.4 |
ME mean error, RMSE root mean squared error, CI confidence interval, rsc rescore.