Towards Validating the Effectiveness of Obstructive Sleep Apnea Classification from Electronic Health Records Using...

Area of Science:

• Sleep Medicine
• Biomedical Informatics
• Machine Learning

Background:

• Obstructive sleep apnea (OSA) is a prevalent sleep disorder characterized by airway obstruction during sleep.
• Polysomnography is the gold standard for OSA diagnosis but is costly and inaccessible.
• Effective screening tools are needed to identify individuals requiring further sleep evaluation.

Purpose of the Study:

• To develop and evaluate machine learning algorithms for classifying obstructive sleep apnea (OSA) using routinely collected clinical data.
• To identify key clinical predictors for OSA classification.
• To assess the feasibility of using electronic health records for OSA screening.

Main Methods:

• A machine learning model was trained on 1479 records from the Wisconsin Sleep Cohort dataset.
• Features included demographics, lab results, physical measurements, sleep history, and comorbidities.
• Support vector machines (SVM) were optimized using Bayesian Optimization and Genetic Algorithms with five-fold cross-validation.

Main Results:

• Key predictors for OSA included waist-to-height ratio, neck circumference, BMI, daytime sleepiness, and snoring.
• The SVM model achieved an accuracy of 68.06%, sensitivity of 88.76%, and specificity of 40.74%.
• The model demonstrated high negative predictive value (73.33%), suggesting utility in ruling out OSA.

Conclusions:

• Routinely acquired clinical data can be effectively utilized for OSA classification.
• Machine learning models show promise for screening and prioritizing patients for polysomnography.
• This approach can improve the accessibility and efficiency of OSA diagnosis.