Respiratory event index underestimates severity of sleep apnea compared to apnea-hypopnea index

Abstract

Polygraphy (PG) is often used to diagnose obstructive sleep apnea (OSA). However, it does not use electroencephalography (EEG), and therefore cannot estimate sleep time or score arousals and related hypopneas. Consequently, the PG-derived respiratory event index (REI) differs from the polysomnography (PSG) derived apnea-hypopnea index (AHI). In this study, we comprehensively analyzed the differences between AHI and REI. Conventional AHI and REI were calculated based on total sleep time (TST) and total analyzed time (TAT), respectively, from two different PSG datasets (N=1561). Moreover, TAT-based AHI (AHITAT) and TST-based REI (REITST) were calculated. These indices were compared keeping AHI as the gold standard. The REI, AHITAT, and REITST were significantly lower than AHI (p<0.0001, p≤0.002, p≤0.01, respectively). The total classification accuracy of OSA severity based on REI was 42.1% and 72.8% for two datasets. Based on AHITAT the accuracies were 68.4% and 85.9%, and based on REITST they were 65.9% and 88.5% compared to AHI. AHI was most correlated with REITST (r=0.98 and r=0.99 for the datasets) and least with REI (r=0.92 and r=0.97). Compared to AHI, REI had the largest mean absolute error (13.9 and 6.7) and REITST the lowest (5.9 and 1.9). REI had the lowest sensitivity (42.1% and 72.8%) and specificity (80.7% and 90.9%) in both datasets. Based on these present results, REI underestimates AHI. Furthermore, these results indicate that arousal-related hypopneas are an important measure for accurately classifying OSA.