The main findings of this study are that a machine learning–based model underpinned by OSA endotype concepts that includes routinely collected sleep study and clinical data inputs may help predict treatment outcomes to oral appliance therapy for people with OSA. In addition, unlike most machine-learning approaches in which the underlying decisions of the model are not transparent, the complementary decision tree learner approach used in the current study allows for visualization of the data and model decisions to provide unique physiological insight and the opportunity for clinical oversight. While further validation in larger clinical data sets is required, this novel approach has the potential to be a useful clinical tool to help identify patients for oral appliance therapy to increase treatment success rates.
The patient journey for those prescribed oral appliance therapy is often time-consuming and costly. For example, many have previously tried and failed CPAP therapy. Thus, by the time the patient receives an appropriately fitted and titrated oral appliance device they may have undergone multiple sleep studies and medical appointments (ie, sleep physician and dental visits). Accordingly, given the time and financial burden, understandably there is a strong desire and often an expectation for treatment success. Yet on average, roughly half of all patients prescribed oral appliance therapy currently have an incomplete therapeutic response. This leaves many patients frustrated and disgruntled and at risk of not pursuing further treatment options to alleviate their OSA symptoms.
Comments