Application of machine learning to predict obstructive sleep apnea syndrome severity

Area of Science:

• Medical Informatics
• Pulmonology
• Artificial Intelligence

Background:

• Obstructive sleep apnea syndrome (OSAS) is a significant public health issue.
• Polysomnography (PSG) is the gold standard for OSAS diagnosis but is costly and time-consuming.
• There is a need for efficient methods to assess OSAS severity.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) models in predicting OSAS severity.
• To utilize demographic and questionnaire data for OSAS severity prediction.
• To assess the clinical applicability of ML for prioritizing patients for PSG.

Main Methods:

• Collected data from 313 patients with diagnosed OSAS, including demographics, spirometry, gas exchange, and symptoms.
• Applied principal component analysis to select 19 key variables.
• Trained and validated seven classification and five regression ML models using stratified 10-fold cross-validation.

Main Results:

• Support vector machine and random forest models showed superiority in classification.
• Support vector machine and linear regression were best for predicting the apnea-hypopnea index (AHI).
• High predictive accuracy was achieved with a limited number of features, though overall classification accuracy was 44.7%.

Conclusions:

• Predicting AHI or OSAS severity class solely from pre-PSG data is challenging.
• ML methods show promise for prioritizing patients for PSG but cannot automate diagnosis.
• Further research may refine ML models for improved OSAS severity assessment.