Do Sleep Questionnaires Add Predictive Value Beyond Demographics for Obstructive Sleep Apnea Severity? Evidence from 67,397 Patients

In a retrospective cohort of 67,397 adults referred for in-laboratory polysomnography, demographics-only AUROC for OSA severity (AHI thresholds) was within 0.016–0.038 of all-features models.

• Why it matters: Sleep questionnaires are commonly used to screen for obstructive sleep apnea, but their incremental predictive value beyond basic demographics is not well quantified.
• What they did: XGBoost models were trained using 169 variables, questionnaire items only, or demographics only, and evaluated on a held-out 10% test set for AHI and RDI at 5, 15, and 30 events/h.
• The result: Questionnaire-only models performed worse (AUROC 0.674–0.699 for AHI), while age, BMI, and gender dominated SHAP contributions; questionnaire information added at most 0.038 AUROC.

AI summary of the abstract · may contain errors · not medical advice · read the paper.