0586 Advanced Tree Models to Predict Moderate-to-Severe Obstructive Sleep Apnea

Abstract

Abstract Introduction The aim of this study was to develop a predicting model for the moderate-to-severe obstructive sleep apnea (OSA) by using advanced tree models. Methods We retrospectively investigated the medical records of patients who undertaken overnight polysomnography (PSG) at our sleep disorders center. We divided the data to a training set (70%) and a test set (30%), randomly. We made a random forest and a XGBoost model to predict the moderate-to-severe OSA (apnea hyponea index [AHI] ≥ 15/h) by using the training set, and then applied each models to the test set. To compare the fitness of the models, we used an accuracy, and an area under curve (AUC). Results Finally, 1,426 patients (AHI < 5:AHI ≥ 15= 464:962) were enrolled. The random forest model showed an accuracy of 0.79, and AUC of 0.82. In the random forest model, the sleep apnea scale of the sleep disorders questionnaire (SA-SDQ), age, neck circumference, male sex, body mass index (BMI), hypertension, and hyperlipidemia appeared in order of a variance importance. The XGBoost model showed an accuracy of 0.75 and AUC of 0.79. Conclusion The random forest model to predict moderate-to-severe OSA showed better performance compared to the XGBoost model. The further study for validation is required. Support None