Performance characteristics of obstructive sleep apnea screening instruments in adults with epilepsy

Obstructive Sleep Apnea (OSA) is a highly prevalent, but often overlooked comorbidity in adults with epilepsy (AWE). Screening instruments widely used in general populations include STOP (Snoring, Tiredness, Observed apneas, high blood Pressure) and STOP-BANG (+BMI, Age, Neck circumference (NC), and Gender). Novel screens such as STOP-BAG₂ may improve OSA risk stratification, but validity and performance characteristics of OSA instruments have not been studied in large AWE populations.

We analyzed 724 AWE with STOP-BANG data available within 1 year of polysomnography (PSG). The primary outcome was apnea-hypopnea index (AHI) ≥ 5. We compared STOP, STOP-BANG, STOP-BAG, STOP-BAG₂, and reweighted STOP-BANG₂. Sensitivity, specificity, positive predictive value, negative predictive value, and area under the curve (AUC) were estimated. Instrument performance was compared by differences in AUC, with optimal cutoffs identified by the Youden index and internally validated using 1,000 bootstrap resamples. Seizure severity and frequency were assessed with the Liverpool Seizure Severity Scale.

Higher STOP, STOP-BANG, STOP-BAG, and STOP-BAG₂ scores were associated with AHI ≥ 5. STOP-BANG, STOP-BAG, STOP-BAG₂, and STOP-BANG₂ were superior to STOP. STOP-BAG₂ and STOP-BANG₂ showed the strongest predictive power, each with an AUC of 0.70. At the Youden cutoff, STOP-BAG₂ yielded 67 % sensitivity and 65 % specificity, while STOP-BANG₂ yielded 72 % sensitivity and 60 % specificity. STOP-BAG₂ was superior to STOP-BANG and STOP-BAG (p = 0.003 and p = 0.001, respectively) and did not differ from STOP-BANG₂ (p = 0.84).

STOP-derived instruments showed limited-to-moderate discrimination for PSG-defined OSA. The continuous weighted models had the highest AUCs, but the gain over categorical scores was modest, and their selected cutoffs would miss a clinically relevant portion of OSA cases.