Optimizing screening of severe obstructive sleep apnea in patients undergoing bariatric surgery

Abstract Background Obstructive sleep apnea is common in patients waiting for bariatric surgery (BS). International consensuses have recommended assessment of obstructive sleep apnea in the preoperative evaluation to avoid perioperative complications. Polysomnography is the standard diagnostic method but is expensive and time-consuming. The aim of our study was to detect those patients who merit treatment before BS using a simple predictor model. The study was conducted at 3 university hospitals (Hospital de Bellvitge, Hospital de la Santa Creu i Sant Pau, Hospital Clinic de Barcelona). Methods A prospective cross-sectional study was conducted of 136 consecutive bariatric subjects. The outcome variable was severe obstructive sleep apnea, defined as an apnea-hypoapnea index of ≥30 events/hr by polysomnography. The predictors evaluated were anthropometric and clinical in the first model, with an oxygen desaturation index of ≥3% added to the second model. Predictive models were constructed using multivariate logistic regression analysis. The best model was selected according to the area under the receiver operating characteristic curve. Results The first model identified 4 independent factors: age, waist circumference, systolic blood pressure, and witnessed apnea episodes, with a sensitivity of 78%, specificity of 68%, and area under the receiver operating characteristic curve of .83 (95% confidence interval .76–.90, P < .001). The second model identified 2 independent factors (witness apnea episodes, oxygen desaturation index of ≥3%), with a sensitivity of 91%, specificity of 85%, and area under the receiver operating characteristic curve of .94 (95% confidence interval .89–.98, P < .001). The 2-step model predictive values were sensitivity of 90%, specificity of 91%, and accuracy of 90% (95% confidence interval 84–94%). After applying the first model and then the second, 45% of subjects would have been ruled out (15% and 30%, respectively) and 55% would require additional sleep management before BS. Conclusion The proposed model could be useful for improving the management of complex patients before BS and optimizing limited polysomnography resources.