C100-02 Prototyping OSA: A New Framework for Predicting Cardiovascular Outcomes and CPAP Benefit

Abstract Rationale The prevalence of obstructive sleep apnea (OSA) is one billion persons worldwide.1 While observational studies have demonstrated that OSA independently associates with major adverse cardiac events (MACE), randomized controlled trials have not shown a significant reduction in MACE among patients randomized to CPAP.2-8 This discrepancy may be explained by recent studies demonstrating that OSA is a heterogeneous disease with distinct subgroups.9-11 We hypothesized that unsupervised cluster analysis using multiple data domains could help us identify heterogeneity amongst OSA patients in terms of their baseline characteristics, intrinsic MACE risk, and response to CPAP therapy. Methods Secondary analysis was performed using SAVE trial data.6 All participants randomized to CPAP versus usual care were evaluated, and all baseline features were assessed. After data pre-processing, multiple clustering architectures were fitted using Gaussian Mixture Models, with the final number of OSA prototypes selected based on Bayesian Information Criterion and the C-Index.12 Prototype and treatment effects on MACE were estimated using Cox proportional hazards model. Using pairwise t-test or Wilcoxon’s rank-sum test, these prototypes were compared across quality of life (QoL), Epworth Sleepiness Scale (ESS), Framingham Risk Score (FRS), and individual treatment effects (ITE) of CPAP on MACE as determined by a previously developed causal survival forest model.13,14 Results Using 88 variables across all available data domains, we identified five distinct prototypes among the 2,717 SAVE participants. Prototype 1’s predictive characteristics included diabetes and metabolic disease, while Prototypes 2 and 4 patients had a greater prevalence of smoking. Prototype 5 patients had prior history of stroke. Prototypes 3 and 5 demonstrated the highest QoL (Figure 1A), while Prototype 1 had higher ESS scores (Figure 1B) (p 0.05), FRS (Figure 1C) (p 0.05), and rates of MACE (p 0.0005) (Figure 1D). Importantly, the ITE analysis (Figure 1E) demonstrated that only individuals within Prototypes 1 and 4 derive significant cardiovascular benefit from CPAP (p 0.05) while individuals within Prototypes 2, 3, and 5 showed potential harm (p 0.05). Conclusion We identified five distinct OSA prototypes that differed in baseline cardiovascular risk and CPAP treatment effects on MACE. Only two prototypes benefited from CPAP, while others showed potential harm, underscoring meaningful heterogeneity in treatment response. These findings offer an explanation for the neutral results of prior trials and support a precision medicine approach to OSA assessment and management. Validating these prototypes in future studies will be essential to advancing individualized treatment strategies that improve cardiovascular outcomes. This abstract is funded by: 5R01HL168897-02