Likelihood ratios for a sleep apnea clinical prediction rule

Nocturnal polysomnography, the standard diagnostic test for sleep apnea, is an expensive and limited resource. In order to help identify the urgency of need for treatment, we determined which clinical features were most useful for establishing an accurate estimate of the probability that a patient had sleep apnea. Of 263 physician-referred patients, 200 were eligible for the study and 180 (90%) completed it. All patients had their histories recorded with a standard questionnaire, and underwent anthropomorphic measurements and nocturnal polysomnography. Sleep apnea was defined as more than 10 episodes of apnea or hypopnea per hour of sleep. Multiple linear and logistic regression models predictive of sleep apnea were compared with physicians' subjective impressions and previously reported models. Likelihood ratios were calculated for several levels of a sleep apnea clinical score produced by one of the linear models. Predictors of sleep apnea in the final model (R2 = 0.34) included neck circumference, hypertension, habitual snoring, and bed partner reports of nocturnal gasping/choking respirations. This model was superior to physician impression, slightly inferior to more detailed linear and logistic models, and comparable to previously reported models. A sleep apnea clinical score of less than 5 had a likelihood ratio of 0.25 (95% CI: 0.15 to 0.42) and a corresponding posttest probability of 17%, while a score of greater than 15 had a likelihood ratio of 5.17 (95% CI: 2.54 to 10.51) and posttest probability of 81%. These likelihood ratios can simply and accurately determined the probability of whether a patient has sleep apnea.