Table 1.
The differences between values of AHI and ODI determined with manual scoring of polygraphic recordings and automatic artificial neural network analyses in the primary test set and in the Embletta test set.
| Error parameter | Primary test set (N = 198) | Embletta test set (N = 1959) | ||
|---|---|---|---|---|
| AHI | ODI | AHI | ODI | |
| mean absolute error (events/hour)* | 1.41 | 1.17 | 2.23 | 1.34 |
| median absolute error (events/hour)* | 0.78 | 0.68 | 1.35 | 0.76 |
| min error (events/hour)* | 0 | 0 | 0 | 0 |
| max error (events/hour)* | 9.45 | 8.36 | 40.1 | 34.5 |
| median % error* | 15.0 | 14.5 | 25.6 | 10.1 |
| misclassified: mean absolute error (events/hour) | 1.72 | 1.18 | 3.40 | 1.86 |
| misclassified: median absolute error (events/hour) | 1.10 | 0.67 | 1.83 | 0.95 |
| misclassified: median % error | 12.2 | 11.0 | 12.1 | 8.10 |
Mean absolute error, median absolute error, and median % error were also calculated separately for those patients who were misclassified to a wrong OSA severity category when using the neural network- estimated AHI and ODI. *Denotes that the error was calculated for the whole test set.