Difference between automatic and manual analysis of respiratory events in sleep studies

Abstract

Introduction A reliable automatic analysis of respiratory and oxygen desaturation events in sleep studies can decrease the time needed for manual scoring. This study aimed to assess the validity of automatic vs. manual analysis and verify obstructive sleep apnea (OSA) categorization. Materials and methods A general population cohort (n = 27) was assessed twice (’06/’12). The first study was performed with Embla A10 (Natus Medical Inc.) but the latter with T3 device (Nox Medical). All studies were scored in Noxturnal software. Recordings with no cannula flow signal were scored using respiratory inductive plethysmography (RIP) flow. Results The apnea hypopnea index (AHI) ranged from 0 to 20 events/h in the Embla recordings and 0–23 in the T3 by manual scoring. The Embla recordings showed a correlation of r = 0.96 between automatic and manual scoring for AHI and r = 0.99 for oxygen desaturation index (ODI). For T3 recordings the correlation was r = 0.98 and r = 1.00, respectively. For Embla recordings the sensitivity and specificity of the automatic scoring was 0.89 and 0.81 for AH events and 1.00 and 0.88 for OD events. For T3 it was 0.94 and 0.74 for AH events and 1.00 and 0.97 for OD events, respectively. When looking at how subjects were classified by OSA category (non OSA (AHI Conclusion The automatic scoring was acceptable regarding sensitivity and specificity of AH and OD events but its tendency to overestimate respiratory events makes OSA categorizing inaccurate so manual overview is recommended.