Abstract
AIM: To determine the association between total sleep time (TST) spent in increased respiratory effort (RE) and the prevalence of type 2 diabetes in a large cohort of individuals with suspected obstructive sleep apnoea (OSA) referred for in-laboratory polysomnography (PSG). MATERIALS AND METHODS: We conducted a retrospective cross-sectional study using the clinical data of 1128 patients. Non-invasive measurements of RE were derived from the sleep mandibular jaw movements (MJM) bio-signal. An explainable machine-learning model was built to predict prevalent type 2 diabetes from clinical data, standard PSG indices, and MJM-derived parameters (including the proportion of TST spent with increased respiratory effort [REMOV [%TST]). RESULTS: Original data were randomly assigned to training (n = 853) and validation (n = 275) subsets. The classification model based on 18 input features including REMOV showed good performance for predicting prevalent type 2 diabetes (sensitivity = 0.81, specificity = 0.89). Post hoc interpretation using the Shapley additive explanation method found that a high value of REMOV was the most important risk factor associated with type 2 diabetes after traditional clinical variables (age, sex, body mass index), and ahead of standard PSG metrics including the apnoea-hypopnea and oxygen desaturation indices. CONCLUSIONS: These findings show for the first time that the proportion of sleep time spent in increased RE (assessed through MJM measurements) is an important predictor of the association with type 2 diabetes in individuals with OSA.