Development and internal validation of a prediction model for sleep apnea syndrome treated with continuous positive airway pressure based on claims and health checkup data linked to personal health records

PurposeTo develop and validate a prediction model for sleep apnea syndrome (SAS) treated with continuous positive airway pressure (CPAP) in the general population. MethodsUsing claims and health checkup data held by JMDC Inc., linked to personal health records (Pep Up), we developed and internally validated a prediction model for SAS treated with CPAP, defined as a diagnosis of SAS and reimbursement records of CPAP. Every three months from January 1, 2022 to July 1, 2024 (i.e., 11 timepoints), we identified eligible individuals with available data both 1 year before and 1 year after that timepoint to define the presence/absence of SAS treated with CPAP, as well as 279 predictor variables. We developed a LightGBM model for the training and tuning datasets and evaluated its performance on the validation dataset. ResultsAmong 18,692,873 observations (mean age 44.8{+/-}11.3 years, women 37.5%) obtained from 1,858,566 people, 300,868 (1.6%) had SAS treated with CPAP. The area under the receiver operating characteristic curve was 0.898 (95% confidence interval 0.895-0.901). The positive predictive values among people with the top 1% and 10% prediction scores were 28.3% and 10.3%, respectively. According to the SHapley Additive exPlanations plot, male sex was the most important predictor, followed by age, body mass index, and waist circumference. We also demonstrated that personal health records significantly improved the predictive performance. ConclusionWe developed a prediction model to identify people at high risk of SAS and encourage them to undergo polysomnography or related tests.