PS 3-55 WEIGHTING OF AHI SCORES AND SYMPTOMS IN THE DIAGNOSIS OF SLEEP APNEA BY SLEEP SPECIALISTS

Tuesday, October 25, 2016
Bayshore Ballroom ABC, Lobby Level (Westin Bayshore Vancouver)
Poster Board # PS 3-55

Nicholas Mitsakakis, MSc PhD1, Lusine Abrahamyan, MD, MPH, PhD1, Valeria E. Rac, MD, PhD1, Suzanne Chung, BEd, CCRP2, Petros Pechlivanoglou, MSc, PhD3, Michael Fitzpatrick, MB, BCh, DCH, FRCPI, MD, FRCPC, FAASM4 and Murray Krahn, MD, MSc, FRCPC1, (1)Toronto Health Economics and Technology Assessment (THETA) Collaborative, Toronto, ON, Canada, (2)Toronto Health Economics and Technology Assessment (THETA) Collaborative, University of Toronto, Toronto, ON, Canada, (3)Child Health Evaluative Sciences, The Hospital for Sick Children, Toronto, ON, Canada, (4)Sleep Disorder Clinic, Kingston General Hospital, Queen’s University, Kingston, ON, Canada
Purpose: Obstructive Sleep Apnea (OSA) is characterized by periodic stoppage of breathing during sleep and it is usually diagnosed with the use of a sleep study, often in a lab (polysomnography), which estimates the Apnea-Hypopnea Index (AHI). Clinical guidelines recommend the clinical diagnosis of OSA to be made based of a combination of the AHI and symptoms (choking during sleep, unrefreshing sleep, fatigue, sleepiness). We hypothesized that sleep specialists rely heavily on the AHI scores alone, regardless of symptoms, when making a diagnosis. We examined this hypothesis using data from a large clinical trial.

Method: Data of AHI scores and clinical diagnoses from 269 patients participating in the SIESTA trial were used. A composite binary variable was created summarizing symptoms of daytime sleepiness and fatigue. ROC curve analysis and logistic regression models were used for assessing how well AHI scores alone can discriminate patients and predict a diagnosis of OSA, and how much the discriminatory power increases when additional information on symptoms is considered, using  methods including the Area Under the Curve (AUC), optimal cutoff and Wald χ2 tests for the model covariates. In addition, classification of patients using the recommended guidelines was compared against the actual clinical diagnosis.

Result: The AHI alone was able to discriminate the diagnosed patients with high precision (AUC=0.9702), with an optimal cutoff of 5.2. When added to the logistic regression model as covariate, the composite symptom variable created a marginal improvement (AUC=0.9721), but without being statistically significant (p=0.267). The classification based on the guidelines gave a very high positive predictive value (0.99) but an unsatisfactory negative predictive value (0.73), due to a significant number (n=17) of false negatives. Interestingly, all of these false negatives had AHI < 5 and only 4 of them had signs of daytime sleepiness. This indicates that sleep specialists assign a positive diagnosis almost always to patients with AHI ≥ 5, but occasionally also to patients with AHI < 5, which is against the recommended guidelines.

Conclusion: Our analysis has shown that an AHI score ≥5 alone predicts very accurately the clinical diagnosis of OSA by sleep specialists. However, it also demonstrates that some patients are diagnosed as having OSA using criteria other than the AHI and symptoms of daytime sleepiness. Additional investigation is needed.