The site, pattern and degree of upper airway collapse is associated with the outcome of
different OSA therapies. In current clinical practice, this information is assessed
during DISE, in which the upper airway is evaluated during a light sedation, mimicking
natural sleep. Information on the site of collapse is not currently available from
routine clinical sleep studies.
Recent research has shown that the site of upper airway collapse seen during endoscopy
can be recognized from the distinct airflow shape patterns. For example, greater
"negative effort dependence" (NED) or inspiratory scooping is associated with
non-tongue-base sites of collapse. In preliminary analysis using >150 patients, the
investigators recently developed a logistic regression model predicting complete
concentric collapse at the level of the palate (CCCp). The model included 6 meaningful
airflow features (scoopiness NED, inspiratory skewness, peak flow in early inspiration,
inspiratory volume in the first 3rd of inspiration, inspiratory rise time, peak volume).
Each feature is calculated as the mean value during identified hypopneas. Other sites of
collapse were also assessed using the same six features in separate models (lateral
walls, tongue-base, epiglottis).
The study will proceed in two phases of development. In the first phase, the
investigators will prospectively validate the preliminary model for predictive CCCP in
300 patients (primary outcome test for phase I). Odds for true CCCP in the predicted CCCP
subgroup will be compared against the odds for CCCP in the predicted non-CCCP subgroup
(Fisher exact test). Secondary analyses will validate the other three model (lateral
walls, tongue-base, epiglottis).
Prior to the second phase, the investigators will develop a refined model to predict CCCP
using all available pooled data (N>450). The investigators will seek to include new
parameters or incorporate additional interactions between predictors, where appropriate,
Refined models for other sites will also be developed. Sensitivity analyses will examine
different patterns of collapse (e.g. anterior-posterior), different severities (partial
vs complete), how to optimally account for multi-site obstruction, and whether predicted
sites are correlated.
In the second phase, the investigators will prospectively apply the refined model to
N=700 patients, following the same analytic approach as for phase 1.
Interim analysis will be performed but the study will not be terminated for early
success).
Patients will undergo therapies per clinical indication and will be documented; In the
entire dataset, the investigators will explore whether the presence of CCCP (per model
prediction, and per DISE results) is associated with reduced efficacy of oral appliances,
hypoglossal nerve stimulation, positional therapy, greater efficacy of upper airway
surgery, greater CPAP pressure requirement, reduced CPAP efficacy, and reduced CPAP
adherence. Associations wlil also be performed using other predicted and actual site of
collapse information.