A Baseline Prediction Model for First-Year CPAP Treatment Disengagement in Adults with Obstructive Sleep Apnea: A Secondary Analysis of a Danish Cohort

Abstract

Purpose To estimate and internally evaluate a fixed baseline model for recorded first-year continuous positive airway pressure (CPAP) treatment disengagement before device-use data are available and compare it with more complex alternatives. Patients and Methods We conducted a secondary analysis of a Danish single-center cohort that initiated CPAP in 2012–2013. The outcome was the recorded 52-week treatment-status event, which did not distinguish active discontinuation from loss to follow-up. Of 695 participants, 662 had determinate status and 94 events. The fixed logistic model included current smoking, Epworth Sleepiness Scale (ESS) score, and apnea–hypopnea index (AHI), with coefficients estimated in 632 complete cases (85 events). Performance was evaluated by repeated 10-fold cross-validation with bootstrap confidence intervals (CIs). Results Current smoking was associated with higher odds of the recorded outcome (odds ratio [OR] 2.45, 95% CI 1.47–4.07). Higher ESS (OR per 5-point increase, 0.68; 95% CI, 0.53–0.87) and AHI (OR per 10 events/hour increase, 0.56; 95% CI, 0.45–0.68) were associated with lower odds. The cross-validated area under the curve was 0.762 (95% CI, 0.709–0.812), the Brier score was 0.103 (0.087–0.119), the calibration intercept was −0.001 (−0.259 to 0.226), and the calibration slope was 0.927 (0.624–1.298). In 662 participants, extended and full models did not consistently improve prediction error. The 94 events could not be separated into active discontinuation and loss to follow-up. Conclusion Three routinely available baseline measures provided moderate predictive information, while broader and penalized models offered no consistent improvement. These findings provide a fixed-horizon benchmark for external validation in contemporary CPAP cohorts that directly assess device use, treatment preferences, and treatment burden.