Diagnostic accuracy of respiratory polygraphy interpretation for pediatric obstructive sleep apnea using a built-in algorithm

Respiratory polygraphy (RP) underestimates the obstructive apnea-hypopnea index (OAHI) compared with polysomnography (PSG). To minimize this bias, a built-in algorithm integrated into the RP software was used to classify wake, NREM and REM sleep stages (RP-Algorithm). The aim was to determine the accuracy of the algorithm in detecting the three-sleep stages and to assess the agreement between OAHI obtained by RP-Algorithm or RP and OAHI by PSG. PSG recordings from 88 children (median age [1 st ; 3 rd quartile]: 10.5 [7.5; 13]years) were included. Several weeks after the PSG analysis, the scoring was deleted, and recordings blindly reanalyzed considering only respiratory signals and the built-in algorithm (Nox BodySleep™ 1.0, ResMed; RP-Algorithm). Few weeks later, a final analysis was performed after deleting the previous, and recordings were blindly reanalyzed as RP. The RP-Algorithm showed an accuracy of 90% and a Cohen’s kappa of 0.68 for the three-stage sleep classification. OAHI was underestimated by RP-Algorithm (3.5 [2.2; 6.2/h) or RP (3.0 [1.9; 5.5]/h) compared to PSG (4.8 [3.3; 6.6]/h; p<0.0001 for both), confirmed by Bland-Altman analysis (mean difference -1.39/h (95%CI: -3.62–0.84) for RP-Algorithm and -1.92 events/h (95%CI: -4.41–0.57) for RP). Severity category changed in 13 children (15%) when diagnosed by RP-Algorithm and in 23 children (26%) when diagnosed by RP. The built-in algorithm incorporated into RP interpretation demonstrated high accuracy for three-stage sleep classification and reduced the underestimation of the OAHI compared with standard RP, when PSG was used as gold standard, with only a few children changing the severity category.