Surface cephalometric and anthropometric variables in OSA patients: statistical models for the OSA phenotype.

Study Objective: We used statistical modelling to probe the contributions of anthropometric and surface cephalometric variables to the OSA phenotype.
Design: The design is prospective cohort study.
Setting: The setting is community-based and sleep disorder laboratory.
Patients or Participants: Study #1-Model development study: 147 healthy asymptomatic volunteers (62.6 % Caucasian; age, 18-76 years; 81 females; median multivariable apnea prediction index=0.15) and 140 diagnosed OSA patients (84.3 % Caucasian; age, 18-83 years; 41 females; polysomnography [PSG] determined apnea-hypopnea index >10 events/h). Study #2-Model test study: 345 clinic patients (age, 18-86 years; 129 females) undergoing PSG for diagnosis of OSA.
Intervention: We measured 10 anthropometric and 34 surface cephalometric dimensions (calipers) and calculated mandibular enclosure volumes for study #1 and recorded age and neck circumference for study #2. Statistical modelling included principal component (PC), logistic regression, and receiver-operator curve analyses.
Measurements and Results: Model development study: A regression model incorporating three identified PC predicted OSA with 88 % sensitivity and specificity. However, a simplified model based on age and NC alone was equally effective (87 % sensitivity and specificity). Model test study: The simplified model predicted OSA with high sensitivity (93 %) but poor specificity (21 %).
Conclusion: We conclude that in our clinic-based cohort, craniofacial bony and soft tissue structures (excluding neck anatomy) do not play a substantial role in distinguishing patients with OSA from those without. This may be because craniofacial anatomy does not contribute greatly to the pathogenesis of OSA in this group or because referral bias has created a relatively homogeneous phenotypic population.