View Abstract

A3973 - Development of an Enhanced Obstructive Sleep Apnea Screening Model in Individuals with Atrial Fibrillation
Author Block: A. M. May1, L. Wang2, R. Mehra3; 1Case Western Reserve University, Cleveland, OH, United States, 2Quantitative Health Services, Cleveland Clinic Foundation, Cleveland, OH, United States, 3Neurologic Institute, Cleveland Clinic Foundation, Cleveland, OH, United States.
Rationale: Although obstructive sleep apnea (OSA) is highly prevalent in atrial fibrillation (AF), standard OSA screening tools have suboptimal performance in AF. We leverage a dataset of well-phenotyped AF individuals to develop an enhanced OSA prediction model and hypothesize that incorporating variables important in OSA-related AF pathophysiology improves the performance of OSA screening tests in AF. Methods: The Sleep Apnea and Atrial Fibrillation Biomarkers and Electrophysiologic Atrial Triggers (SAFEBEAT-NCT02576587) is a case-control study including adults with paroxysmal AF and controls without AF. Each participant (n=150) completed OSA screening questionnaires and underwent attended 16-channel polysomnography. Models with STOP-BANG predictors (Snoring, Tired, Observed apneas, high blood Pressure, body mass index (BMI)>35kg/m2, Age>50, Neck circumference>40cm, and male Gender) with the addition of average heart rate from 3 resting measures and left atrial volume were constructed. Best subsets analysis were used to select subsets of predictors for evaluation. Models were tested using validation set approach with bootstrapping. Test performance for two thresholds was assessed: apnea-hypopnea index (AHI) ≥5 and ≥15. R statistical software was used for analyses. Results: The 150 participants with AF were 61.28±12.11 years with BMI of 31.20±6.62 and a mean AHI of 16.7±17.1; 65 (43.3%) had an AHI≥15. For AHI≥15, STOP-BANG (AUC 0.774±0.073) did not perform as well as NABS - neck circumference, age, and BMI as continuous variables and snoring - (AUC 0.813±0.076). The optimal model for AHI≥15 was NABS and heart rate (AUC 0.813±0.076, sensitivity = 0.848, specificity = 0.63), which had a significantly improved net reclassification index and integrated discrimination index. For AHI≥5, all models performed worse than for comparable models examined at the AHI≥15 cutoff, and NABS had the best performance characteristics (AUC 0.720±0.074, sensitivity = 0.35, specificity = 0.875); however, NABS was not significantly better relative to STOP-BANG based on net reclassification index or integrated discrimination index. Conclusions: The NABS with heart rate has improved characteristics over other models, including the STOP-BANG, for moderate to severe OSA. However, NABS alone had the best performance for a lower AHI threshold including mild OSA. These data suggest that aside from snoring, symptoms such as tiredness/sleepiness and observed apnea do not substantively contribute to OSA prediction in AF. Further studies are needed to replicate and validate this model in other AF populations.