Estimation of the probability of disturbed breathing during sleep before a sleep study.

We have investigated the ability of a statistical model developed from clinical data and questionnaire responses to predict disturbance of breathing during sleep. Data from 100 consecutive patients referred for sleep study for suspected sleep apnea were used to develop the model using logistic regression analysis. For each subject, the model predicted the probability of having an apnea-hypopnea index (AHI) greater than 15; this probability was compared with the AHI measured from sleep study. A probability cutoff point (= 0.15) was decided on that minimized the number of subjects with false-negative predictions. Four terms--apneas observed by bed partner, hypertension, body mass index, and age--were found to contribute significantly to the model with observed apneas being by far the most predictive term of the four (adjusted odds ratio 19.7). When the model was tested to estimate the probability of an AHI greater than 15 for 105 patients from a second group of consecutive patients referred for sleep study, the model correctly classified 33 of 36 patients with a measured AHI greater than 15 (sensitivity = 92%) and 35 of 69 patients with a measured AHI less than or equal to 15(specificity = 51%). This study shows that analysis of clinical features of patients presenting with suspected sleep apnea may reduce the need for sleep studies by about one-third yet still lead to the identification of the great majority of patients with abnormal breathing during sleep.