Article Text

Download PDFPDF

P019 Surface craniofacial phenotype and prediction models for obstructive sleep apnoea: a 3-D evaluation
  1. Bahn Agha1,2,
  2. Lifong Zou1 and
  3. Ama Johal1
  1. 1Queen Mary University of London, The London School of Medicine and Dentistry, London, UK
  2. 2Mustansiriyah University, College of Dentistry, Baghdad, IRAQ


Aims Obstructive sleep apnoea (OSA) is considered as a major healthcare problem, remaining relatively underdiagnosed and associated with significant comorbidity (Rejón-Parrilla et al., 2014). The present study aims to explore: the existence of a craniofacial phenotype in adults with OSA, the ability to predict the condition from clinical and surface craniofacial structures and the presence of a surface facial marker for OSA.

Methods A case-control study was conducted with 118 middle-aged Caucasian males (56 controls; 62 OSA). Each undergoing a clinical examination including body mass index, Mallampati airway classification, sleep apnoea clinical score, Epworth sleepiness scale and 3-D stereophotography for surface craniofacial analysis (figure1).

Abstract P019 Figure 1

3-D Sterophotogrammetry shows surface surface landmarks of the face, neck an upper torso

Results Surface craniofacial risk factors (phenotype) were identified for OSA Caucasian males, with the predominant characteristics being: an enlarged neck circumference (p<0.001), short neck (p<0.001), large mandibular width (p<0.001), forward head posture (p<0.001) and increased lower anterior facial height (P<0.002). Multiple regression analysis for the surface predictors, both with and without clinical variables, identified a range of prediction models with moderate to high sensitivity and specificity, with an area under the receiver operating characteristics curve (AUC) between 0.73–0.82. A higher positive and lower negative likelihood-ratios were identified for the combination model of the surface and clinical variables (LR+6.02, LR–0.40 respectively) when compared to the surface model alone. A high positive post-test probability and odds ratio (87%, OR=6.8, respectively) were also identified in the combination model. The surface model, with and without clinical variables, not only successfully identified OSA subjects from controls (AUC=0.77, 0.82 respectively) but also presented as a marker (figure 2).

Abstract P019 Figure 2

a) Receiver operating characteristic (ROC) curve for backward logistic regression model for surface and clinical variables with area under the curve (AUC) of 0.82. b) Box plot showing logistic regression model score for OSA subjects and controls. A cut-off value of 0.67 (red line, b) produced 65% sensitivity and 89% specificity for identification of OSA patients with an AHI≥5 events/hour. c) Prediction profilers for logistics multiple regression – surface and clinical predictors (backward model). Prediction formula: Prob (OSA) = 1/(1+Exp (-((-7.88) + 0.23 *BMI + 1.88 * MAC)))

Conclusion This case-control study demonstrated the existence of a surface craniofacial phenotypic pattern, identified a predictive model and marker for OSA in Caucasian males, using 3D-surface imaging analysis and clinical tools.

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.