This study confirmed that the predictive value of the OSA-18 for SDB severity is weak. Some questions perform better than others statistically, and the seven categories of response introduce significant statistical noise, raising the possibility that modification of the OSA-18 may improve its performance in the prediction of OSA severity.