A prediction model based on an artificial intelligence system for moderate to severe obstructive sleep apnea

Area of Science:

• Medical diagnostics
• Artificial intelligence in healthcare
• Sleep medicine

Background:

• Obstructive sleep apnea (OSA) is a prevalent condition, yet accurate diagnosis remains challenging.
• Current screening tools like the Berlin, Rome, and BASH'IM questionnaires lack optimal sensitivity and specificity.
• There is a need for improved methods to identify patients with moderate to severe OSA.

Purpose of the Study:

• To develop and evaluate an artificial intelligence (AI) method for screening moderate to severe obstructive sleep apnea (OSA).
• To compare the performance of AI-derived models against traditional logistic regression for OSA screening.
• To identify patients with an apnea-hypopnea index (AHI) of 15 or higher.

Main Methods:

• A newly developed questionnaire was administered to 120 patients prior to polysomnography (PSG).
• Data from 110 validated questionnaires were used to build five predictive models using a genetic algorithm (GA).
• Logistic regression (LR) was employed for comparative analysis.

Main Results:

• Genetic algorithm (GA) models demonstrated high sensitivity (81.8%–88.0%) and specificity (95%–97%).
• Logistic regression (LR) models exhibited significantly lower performance, with sensitivity of 55.6% and specificity of 57.9%.
• The GA approach proved superior in identifying patients with moderate to severe OSA.

Conclusions:

• Genetic algorithms offer a robust solution for developing effective screening models for moderate to severe OSA.
• The developed questionnaire is self-administered and does not require biochemical data.
• The GA models show promising diagnostic accuracy, with potential for further improvement through larger patient cohorts.