Multimodal phenotypic labelling using awake endoscopy, drug-induced sleep endoscopy and computational fluid dynamics for mandibular advancement device treatment outcome

Abstract

<b>Introduction:</b> Mandibular advancement device (MAD) treatment for obstructive sleep apnea (OSA) is variable and patient dependent. Hence, a global clinical applicable predictive model is needed. <b>Aims and objectives:</b> We aim to develop a prediction model of MAD outcome derived from characteristics obtained during awake nasendoscopy, drug-induced sleep endoscopy (DISE) and CT-scan based computational fluid dynamics (CFD). <b>Methods:</b> One hundred OSA patients were prospectively recruited and treated with a MAD at fixed 75% protrusion. 72 completed 3-month follow-up polysomnography with MAD and underwent awake nasendoscopy, DISE and CFD analysis at baseline. Treatment response was defined as reduction in apnea-hypopnea index (AHI) ≥50%, deterioration as increase in AHI ≥10% during MAD treatment. To deal with missing data, multiple imputation with predictive mean matching was used. Logistic regression, adjusting for BMI and baseline AHI, was used to combine all variables. Potential predictor variables considered were DISE (complete circular palatal collapse (CCCp) and tongue base collapse), CFD (total upper airway volume and velopharyngeal volume) and awake nasendoscopy (soft palatal position and oropharyngeal crowding). <b>Results:</b> Using multivariate logistic regression, only DISE observations remained significant. Response was significantly related to tongue base collapse (odds ratio: 3.60 [1.16-11.21], p=0.027). CCCp during DISE was boundary significantly related with deterioration (odds ratio: 19.27 [0.79-469.87], p=0.069). <b>Conclusions:</b> Our findings suggest DISE to be the most robust value associated with MAD treatment outcome.