BASH-GN: a new machine learning-derived questionnaire for screening obstructive sleep apnea.

Area of Science:

• Sleep Medicine
• Artificial Intelligence in Healthcare
• Biostatistics

Background:

• Obstructive sleep apnea (OSA) is a prevalent condition requiring accurate screening.
• Current screening questionnaires have limitations in predicting OSA risk effectively.
• Identifying distinct OSA risk factor subtypes can potentially enhance diagnostic accuracy.

Purpose of the Study:

• To develop and validate a novel machine learning-based questionnaire, the BASH-GN (Brief Airway Sleepiness Heuristics - Genomics Network), for classifying obstructive sleep apnea (OSA) risk.
• To incorporate risk factor subtypes into the OSA risk prediction model.
• To compare the performance of the BASH-GN against established OSA screening tools.

Main Methods:

• Utilized data from the Sleep Heart Health Study (SHHS 1) and the Wisconsin Sleep Cohort (WSC) for model development and independent testing.
• Employed mutual information to rank potential risk factors, selecting the top 50% for analysis.
• Developed the BASH-GN using two logistic regression classifiers to predict OSA risk based on identified subtypes.

Main Results:

• The BASH-GN demonstrated superior performance compared to the Four-Variable, Epworth Sleepiness Scale, Berlin, and STOP-BANG questionnaires on both SHHS 1 and WSC test sets.
• Achieved high area under the receiver operating characteristic curve (AUROC) values (SHHS 1: 0.78, WSC: 0.76) and area under the precision-recall curve (AUPRC) values (SHHS 1: 0.72, WSC: 0.74).
• The BASH-GN questionnaire is publicly accessible at https://c2ship.org/bash-gn.

Conclusions:

• Integrating OSA subtypes into risk assessment significantly improves screening accuracy.
• The BASH-GN represents a promising advancement in the development of more precise OSA screening tools.
• Machine learning approaches offer a powerful method for refining diagnostic questionnaires in sleep medicine.