ACF-SAP: A machine learning framework for predicting obstructive sleep apnea severity using anthropometric and...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Sleep Medicine Research

Background:

• Obstructive sleep apnea (OSA) is a prevalent condition requiring effective screening.
• Current diagnostic methods like polysomnography (PSG) can be resource-intensive.
• There is a need for accessible, non-invasive tools for OSA severity prediction.

Purpose of the Study:

• To develop and validate ACF-SAP, a machine learning (ML) framework for predicting OSA severity.
• To utilize routinely collected, non-invasive clinical features for OSA assessment.
• To create a scalable and cost-effective screening solution.

Main Methods:

• Leveraged anthropometric and clinical data (sex, BMI, height, weight, neck circumference, nocturia).
• Employed ML-based feature selection to identify key predictors.
• Utilized unsupervised clustering for data-driven severity labels, followed by ensemble classifier training.

Main Results:

• The ACF-SAP framework achieved a classification accuracy of 0.84.
• Demonstrated strong F1-scores and balanced sensitivity across different OSA severity levels.
• The model effectively integrated feature selection and clustering for robust prediction.

Conclusions:

• ACF-SAP facilitates early identification of patients at high risk for OSA.
• The framework can serve as a first-line screening tool, prioritizing PSG referrals.
• This scalable, low-cost solution improves triage efficiency and timely diagnosis, especially in resource-limited settings.