A LightGBM prediction model for sleep apnea syndrome (SAS) treated with CPAP was developed and internally validated using Japanese insurance claims, annual health checkup data, and personal health records (PHRs from the Pep Up platform) across 1,858,566 individuals with 18,692,873 observations from January 2022 to July 2024.
The model achieved an AUROC of 0.898 (95% CI 0.895–0.901). Positive predictive values in the top 1% and 10% of predicted risk were 28.3% and 10.3%, respectively. Male sex, age, BMI, and waist circumference were the top predictors. Adding PHR data improved IDI by 1.07% in light users and 4.57% in heavy users.
- Outcome was CPAP-treated SAS only (prevalence 1.6%), excluding undiagnosed or untreated SAS; model performance may be overestimated due to healthcare-access bias. - Population was limited to Japanese company employees and Pep Up users, limiting generalizability. - Internal validation only; external validation in other populations has not been performed.
This model can flag high-risk individuals for PSG or home sleep testing using routinely collected data, without added testing burden. Clinicians should consider sex-specific thresholds, given marked SAS underdiagnosis in women (prevalence 0.2% vs. 2.4% in men).
Explore related topics
