Performance of Automatic Algorithms for Oxygen Desaturation Detection

Study Objectives: To evaluate the accuracy of Wesper Lab’s automated oxygen desaturation detection algorithm by comparing automatically derived oxygen desaturation index (ODI) values with blinded manual scoring performed by an independent registered polysomnographic technologist.

Methods: Adult patients scheduled for in-laboratory polysomnography (PSG) across four U.S. sleep clinics were recruited to wear two Wesper Lab patches and a fingertip pulse oximeter concurrently with standard PSG. Blood oxygen saturation signals were independently scored for 3% desaturations by a blinded registered polysomnographic technologist in randomized order. Wesper Lab’s automated algorithms independently analyzed the same signals for 3% desaturations. ODI was calculated for both methods as the number of desaturation events divided by total recording time in hours (ODI-Manual and ODI-Wesper). Agreement between methods was assessed using Pearson correlation and linear regression.

Results: Fifty-nine studies from 59 participants were analyzed (34 male; age 21–76 years; BMI 23.0–56.1 kg/m²). Skin tone data were available for 51 participants, of whom 43% had darker skin tones (Fitzpatrick IV–VI or self-reported). Automated ODI showed strong agreement with manual scoring (r = 0.976), with Bland–Altman analysis demonstrating minimal bias (−3.76 to +6.17 events/h) and near-unity linear equivalence across the observed ODI range (Figure 1).

 

Figure 1: Pearson’s correlation and Bland-Altman analysis between Wesper-ODI and PSG-ODI

 

Conclusions: Wesper Lab’s automated oxygen desaturation detection algorithm demonstrates excellent agreement with expert manual scoring across a diverse adult population. These findings support the accuracy and reliability of the algorithm for automated ODI estimation and its suitability for clinical interpretation of sleep-disordered breathing.