A Prediction Nomogram of Severe Obstructive Sleep Apnea in Patients with Obesity Based on the Liver Stiffness and Abdominal Visceral Adipose Tissue Quantification

The diagnosis of severe OSA still relies on polysomnography, which causes a strong sense of restraint in patients with obesity. However, better prediction tools for severe OSA applicable to patients with obesity have not been developed.

Relevant clinical data of 1008 patients with OSA who underwent bariatric surgery in our hospital were collected retrospectively. Patients were divided into training and test cohorts by machine learning. Univariate and multivariate logistic regression analysis was used to screen associations, including liver stiff measurement (LSM) and abdominal visceral tissue (aVAT), and to construct a severe OSA risk prediction nomogram. Then, we evaluated the effectiveness of our model and compared our model with the traditional Epworth Sleepiness Scale (ESS) model. Finally, our associations were used to explore the correlation with other indicators of OSA severity.

Our study revealed that age, biological sex, BMI, LSM, aVAT, and LDL were independent risk factors for severe OSA in patients with obesity. A severe OSA risk prediction nomogram constructed by six indicators possessed high AUC (0.845), accuracy (77.6%), and relatively balanced specificity and sensitivity (72.4%, 82.8%). The Hosmer-Lemeshow test (P=0.296, 0.785), calibration curves, and DCA of the training and test cohorts suggested better calibration and more net clinical benefit. Compared with the traditional ESS model, our model had higher AUC (0.829 vs 0.545), sensitivity (78.9% vs 12.2%), PPV (77.9% vs 53.3%), and accuracy (75.4% vs 55.2%). In addition, the associations in our model were independently correlated with other indicators reflecting OSA severity.

We provided a simple, cheap, and non-invasive nomogram of severe OSA risk prediction for patients with obesity, which would be helpful for preventing further complications associated with severe OSA.