Towards Precise Phenotype Discovery of Obstructive Sleep Apnea with a Data-Inclusive Multi-Study Analysis Using the National Sleep Research Resource (NSRR) - Project Summary/Abstract
Obstructive sleep apnea (OSA) is highly prevalent and associated with a spectrum of cardiovascular
(CV) diseases and adverse health outcomes. However, OSA treatment strategies tend to show inconsistent
treatment efficacy across individuals and little or no reduction in risk of CV diseases, events, or death.
Phenotype discovery is critical for precise risk stratification and targeted treatment of OSA. Substantial
heterogeneity among OSA patients is likely an important contributor to the suboptimal results of clinical trials.
Thus, it is critical to delineate the OSA heterogeneity and stratify patients into high-vs low-risk clusters (i.e.,
“phenotypes”) associated with markedly different outcomes for precise risk stratification and targeted treatment.
OSA data hold great promise to facilitate OSA phenotype discovery. Rigor of Prior Research: (1) We
and others identified new prognostic factors in OSA data that are associated with one or more adverse CV
outcomes. (2) Emerging OSA phenotypes were defined by machine learning and clustering algorithms from
multi-faceted OSA data. (3) Newly identified OSA phenotypes, predictive of patients’ benefit from OSA
treatments and risk for adverse CV outcomes, laid the foundation for OSA phenotypes’ clinical utility in targeted
treatment and precise prognosis. However, significant gaps exist in fully leveraging the OSA data for
phenotype discovery: There is a lack of “outcome-predictive”, “clinically-interpretable”, and “reproducible”
phenotypes, defined from multi-domain OSA data in a large diverse U.S. population.
To address these gaps, we propose a secondary multi-study analysis that seeks to develop new
classification criteria and identify phenotypes in OSA by integrating multi-domain OSA-related sleep common
data elements, including but not limited to patient socio-demographics, health habits, medical history,
anthropometrics, polysomnography measures, daytime sleepiness, quality of life, and cardiovascular
comorbidities and mortalities, combined across three of the largest epidemiological study cohorts deposited in
the NIH-funded National Sleep Research Resource (NSRR). This includes Sleep Heart Health Study, Hispanic
Community Health Study, and Multi-Ethnic Study of Atherosclerosis, with at least 5,336 OSA patients from a
diverse population of African American, Caucasian, Hispanic, and Asian American men and women.
Aim 1: Develop a novel sparse, outcome-predictive multi-domain Factor Mixture Model for OSA phenotype
identification from multi-domain mixed-typed patient pre-clinical features and clinical features. Aim 2: Apply the
developed model in Aim 1 to individual and pooled NSRR datasets to: (1) identify, characterize, and validate
OSA phenotypes; (2) evaluate consistency and reproducibility in findings supported by individual and pooled
analyses. Impact: We will identify, characterize, and validate OSA phenotypes that assist clinicians with
determining how aggressive to be with the treatment plans and assist researchers with selecting appropriate
patients to enroll in clinical trials of OSA treatment, eventually leading to precise prognosis and treatment of OSA.
