Using assessments of craniofacial phenotypic information, obtained using quantitative facial photography, we demonstrated differences in craniofacial phenotype between supine-isolated and non-positional OSA. In particular, the supine-isolated cohort had smaller craniofacial dimensions. Given that these observed craniofacial difference do not remain after adjustment for obesity, it suggests that they are primarily explained by different obesity levels between these OSA clinical phenotypes. Overall, it appears that any craniofacial phenotypic differences between these groups seem to be due to differences in body size. It may not be possible to detect craniofacial skeletal differences between these OSA clinical phenotypes using this method due the inherent clinical differences between these groups which also influence facial measures. Therefore, conclusions about craniofacial skeletal differences and an anatomical basis of supine-dependent OSA cannot be made from this study. However, facial analysis is picking up differences in these clinical characteristics related to the supine-isolated OSA phenotype.

There were no differences in demographics (age, gender, and ethnicity) in our cohorts. Our findings of lower BMI with smaller body circumferences in supine-isolated OSA are well reported. It is yet unclear whether these anthropometric differences could be explained by supine-isolated OSA being on the pathway from snoring to OSA in all positions which may be progressed by increasing fat deposition around the chest and upper airway, or whether supine-isolated OSA is a separate OSA phenotype altogether. The supine-isolated group also had much less severe OSA based on total AHI (on average only mild OSA) compared to the non-positional group. BMI and OSA severity are also related to facial characteristics [2] and hence are confounders in comparing craniofacial characteristics between OSA clinical phenotypes in this sample. To more completely understand if positional OSA is identifiable in facial phenotype beyond these factors, a matched sample on body size/fat distribution and OSA severity would be desirable. Craniofacial measurements also differ by ancestry/race and OSA risk also varies with ancestral background [10]. Although there was no statistical difference in the ethnicity profile between the OSA groups in our cohort, the sample is majority White ethnicity (~ 60%). Although other ethnicities were recruited, in this sample, we did not have sufficient numbers to stratify analysis by ethnicity. Therefore, this is a limitation to the current work. It could be that facial measurements relate more directly to positional OSA phenotypes in specific ethnic groups, such as Asian populations where craniofacial factors have been previously shown to be more associated with OSA risk than soft-tissue factors [5]. Therefore, there could be differences in these craniofacial variables that associate with positional OSA by ethnicity which would need to be explored in larger samples; however, matching on ethnicity and body size was not possible in the sample available and this is a limitation of this analysis.

Facial analysis is a non-invasive and increasingly feasible tool to apply to clinical settings and large-scale research given advances in technologies for image capture and automated analysis. Facial phenotyping has been shown to predict OSA risk and relate to OSA treatment outcomes [2,3,4,5]. In recent years, there has been growing interest in understanding OSA phenotypes for potential to improve outcomes via individualised management [1]. Craniofacial structure and position dependency are phenotypes of OSA that have implications for diagnosis and treatment. In this study, we explored whether facial analysis could have application in identifying OSA phenotypes, in this case position dependency, as well as potential to explain differences in how these phenotypes generate, i.e., underlying craniofacial structure. Although this work is preliminary, it highlights further potential for OSA phenotyping and the richness of information that could be acquired towards individualised management using facial analysis technology.