Prediction of obstructive sleep apnea using visual photographic analysis

Study Objectives: Obstructive sleep apnea (OSA) has been historically underdiagnosed and may be associated with grave perioperative complications. The ASA and American Academy of Sleep Medicine recommend OSA screening prior to surgery; however, only a minority of patients are screened. The objective of this study was to determine the proficiency of anesthesiologists, otolaryngologists, and internists at predicting the presence of OSA by visual photographic analysis without the use of a computer program to assist, and determine if prediction accuracy varies by provider type. Design: Prospective case series. Setting: Tertiary care hospital based academic center. Patients: Fifty-six consecutive patients presenting to the sleep laboratory undergoing polysomnography had frontal and lateral photographs of the face and torso taken. Interventions: Not applicable. Measurements: Polysomnography outcomes and physician ratings. An obstructive apnea hypopnea index (oAHI) >= 15 was considered positive. Twenty anesthesiologists, 10 otolaryngologists, and 11 internists viewed patient photographs and scored them as OSA positive or negative before and after being informed of patient comorbidities. Main Results: Nineteen patients had an oAHI <15, 18 were >= 15 but <30, and 19 were >= 30. The mean oAHI was 28.7 +/- 26.7 events/h (range, 0-125.7), and the mean body mass index was 34.1 +/- 9.7 kg/m(2) (range, 17.4-63.7). Overall, providers predicted the correct answer with 61.8% accuracy without knowledge of comorbidities and 62.6% with knowledge (P <.0001). There was no difference betweenprovider groups (P =.307). Prediction accuracy was unrelated to patient age (P =.067), gender (P =.306), or race (P =.087), but was related to body mass index (P =.0002). Conclusion: The ability to predict OSA based on visual inspection of frontal and lateral photographs is marginally superior to chance and did not differ by provider type. Knowledge of comorbidities did not improve prediction accuracy. (c) 2016 Elsevier Inc. All rights reserved.