TABLE 4.

Comparison of the severity classification performance for different event indices derived from home sleep apnea tests.

Metric	REIHSAT	AHICReSS	P	AHICReSS+AutAr	P
Accuracy	0.702 (0.64, 0.75)	0.771 (0.71, 0.82)	0.019	0.804 (0.75, 0.85)	<0.001
Underestimated	0.188 (0.14, 0.24)	0.131 (0.09, 0.18)	0.022	0.073 (0.05, 0.11)	<0.001
Overestimated	0.110 (0.08, 0.15)	0.098 (0.07, 0.14)	0.624	0.122 (0.09, 0.17)	0.596
Cohen’s κ	0.580 (0.50, 0.66)	0.674 (0.60, 0.75)	0.014	0.716 (0.64, 0.79)	<0.001

Diagnostic performance for the four-class classification task of categorizing sleep disordered breathing into no, mild, moderate, and severe using thresholds of 5, 15, and 30. Accuracy, the fraction of subjects where the severity was underestimated, the fraction of subjects where the severity was overestimated, and Cohen’s κ for AHI, surrogates derived from HSAT, are listed. Point estimates and two-sided 95% confidence intervals are compared in three columns for the number of apneas and hypopneas per hour of monitoring time (REI_HSAT), per hour of total sleep time derived from cardio-respiratory signals (AHI_CReSS), and per hour of total sleep time derived from cardio-respiratory signals where hypopneas were also confirmed with autonomic arousals (AHI_CReSS+AutAr). Metrics for the AHI_CReSS, and AHI_CReSS+AutAr were tested for statistically significant differences to the REI, and corresponding p-values are provided in the columns next to the point estimates and confidence intervals. Statistically significant results are highlighted in grey.