This study focuses on the at-home Sleep apnea-hypopnea
syndrome (SAHS) severity estimation. Three percent
oxygen desaturation index ðODI3Þ from nocturnal
pulse-oximetry has been commonly evaluated as simplified
alternative to polysomnography (PSG), the standard
in-hospital diagnostic test. However, ODI3 has shown
limited ability to detect SAHS as it only sums up
information from desaturation events. Other physiological
signs of SAHS can be found in respiratory and cardiac
signals, providing additional helpful data to establish
SAHS and its severity. Pulse rate variability time series
(PRV), also derived from nocturnal oximetry, is considered
a surrogate for heart rate variability, which provides
both cardiac and respiratory information. In this study,
200 oximetric recordings obtained at patients home were
involved, divided into training (50%) and test (50%)
groups. ODI3 and PRV were obtained from them, the
latter being characterized by the extraction of statistical
features in time domain, as well as the spectral entropy
from the commonly used very low (0–0.04 Hz.), low
(0.04–0.15 Hz.), and high (0.15–0.4 Hz.) frequency
bands. The ODI3 and PRV features were joined in a
multi-layer perceptron artificial neural network (MLP),
trained to estimate the apnea-hypopnea index (AHI),
which is the PSG-derived parameter used to diagnose
SAHS. Our results showed that single ODI3 rightly
assigned 62.0% of the subjects from the test group into
one out the four SAHS severity degrees, reaching 0.470
Cohens kappa, and 0.840 intra-class correlation
coefficient (ICC) with the actual AHI (accuracies of
90.0, 88.0 and 82.0% in the increasing AHI cutoffs used
to define SAHS severity). By contrast, our MLP model
rightly assigned 75.0% of the subjects into their corresponding
SAHS severity level, reaching 0.614 j and
0.904 ICC (accuracies of 93.0, 88.0 and 90.0%). These
results suggest that SAHS diagnosis could be accurately
conducted at-patients home by combining ODI3 and PRV
from nocturnal oximetry
