The Prediction of Obstructive Sleep Apnea Using Data Mining Approaches.

Area of Science:

• Medical Informatics
• Sleep Medicine
• Data Mining

Background:

• Obstructive sleep apnea (OSA) is a prevalent sleep-disordered breathing condition with significant health and economic burdens.
• Current diagnosis relies on polysomnography (PSG), a resource-intensive standard method.
• Developing alternative diagnostic models is crucial for efficient patient identification.

Purpose of the Study:

• To develop and compare data mining models, specifically C5.0 decision tree and logistic regression model (LRM), for predicting OSA.
• To identify the superior model for OSA prediction, potentially reducing reliance on PSG.
• To facilitate early identification of patients with OSA for timely intervention.

Main Methods:

• A cross-sectional study utilizing data from 333 patients referred to a sleep disorders research center (2012-2016).
• All participants underwent standard overnight polysomnography (PSG).
• Performance of a stepwise LRM was compared against a C5.0 decision tree using accuracy, sensitivity, and specificity metrics.

Main Results:

• The C5.0 decision tree achieved an accuracy of 0.757 (sensitivity: 0.66, specificity: 0.809).
• The logistic regression model (LRM) yielded an accuracy of 0.737 (sensitivity: 0.693, specificity: 0.78).
• C5.0 decision tree demonstrated superior performance in diagnosing OSA compared to LRM.

Conclusions:

• The C5.0 decision tree model exhibits better diagnostic performance for OSA than LRM.
• This data mining approach presents a viable alternative to traditional PSG for OSA diagnosis.
• The findings support the use of C5.0 decision trees for more accessible OSA screening.