Development and Validation of an Explainable Machine Learning Model for Identifying Hypertension Status in Patients With Obstructive Sleep Apnea

Obstructive sleep apnea (OSA) is an independent risk factor for hypertension (HTN) worldwide; however, OSA-related HTN is frequently overlooked in clinical practice, partly because of the limitations of traditional clinical blood pressure measurements. Our study aimed to develop an interpretable machine learning model for identifying HTN status among patients with OSA. We analyzed data from 1771 diagnosed OSA patients and randomly split them into a training set (70%) and a testing set (30%). Using LASSO regression and the Boruta algorithm, we identified seven key predictors for prevalent HTN status, namely, age, body mass index, neck circumference, lowest oxygen saturation, percentage of time with oxygen saturation below 90%, the apnea-hypopnea index, and the triglyceride-glucose index. Fourteen machine learning models were evaluated using internal validation based on a single random test set. Among the 14 models, the adaptive boosting model demonstrated superior performance, achieving an area under the receiver operating characteristic curve of 0.830 on the test set. Shapley additive explanation (SHAP) analysis not only confirmed the model's logic but also revealed a dose‒response relationship between each feature and HTN status, highlighting the collective contribution of nocturnal hypoxia burden, metabolic factors, and anatomical factors to the model's prediction. We ultimately developed an online prediction tool to facilitate rapid clinical application. This interpretable machine learning model provides a powerful tool for achieving precise sleep medicine. Our model highlights the strong association of nocturnal hypoxia burden (over respiratory event frequency alone) with HTN status in patients with OSA.