Positional sleep apnea phenotyping using machine learning and digital oximetry biomarkers

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Data Science

Background:

• Obstructive Sleep Apnea (OSA) is a common condition.
• Positional Obstructive Sleep Apnea (POSA) is a subtype influenced by body position.
• Accurate phenotyping of POSA is crucial for effective treatment.

Purpose of the Study:

• To assess the feasibility of using digital oximetry biomarkers (OBMs) and body position data.
• To identify distinct phenotypes of Positional Obstructive Sleep Apnea (POSA).

Main Methods:

• An extreme gradient boost (XGBoost) machine learning model was employed.
• Classified three POSA phenotypes: positional patients (PP) (supine-predominant OSA [spOSA] and supine-isolated OSA [siOSA]) and non-positional patients (NPP).
• Utilized 43 OBMs from supine and non-supine positions, along with demographic and clinical data (META) from 861 OSA patients in the MESA dataset.

Main Results:

• The multiclass classification achieved a median weighted F1 score of 0.79 (IQR: 0.06).
• Binary classification differentiating positional patients (PP) from non-positional patients (NPP) yielded a weighted F1 score of 0.87 (IQR: 0.04).

Conclusions:

• Digital oximetry biomarkers, when analyzed in conjunction with body position, can effectively distinguish between different POSA phenotypes.
• The developed data-driven algorithm shows potential for integration into portable home sleep testing devices.