Undiagnosed Obstructive Sleep Apnea in Primary Care Clinics

Obstructive sleep apnea (OSA) is a major public health problem that is commonly seen in primary care clinics where a majority of outpatient office visits occur. Prior studies suggest that OSA is severely under-diagnosed in primary care practices, but these studies relied on patient reported symptoms rather than objective diagnosis.

Consequently, the prevalence of OSA and rate of under-diagnosis of OSA among primary care practices are not currently known. Given the recent studies implicating the excessively sleepy OSA symptom subtype to the development of cardiovascular events, it is also important to ascertain the prevalence and underdiagnosis rate

of this OSA subtype among patients seen at primary care clinics. In addition to estimating prevalence, it is equally important to evaluate the factors contributing to the likelihood of underdiagnosis across different primary care practices. Barriers to OSA diagnosis include patient factors related to lack of awareness and

understanding of the disease, physician perceptions of OSA risk and knowledge about the heterogeneous presentation of OSA, and system and practice-specific factors such as time constraints. The growth of electronic health records (EHRs) in recent years also provides the unique ability to identify these undiagnosed

and untreated cases via the application of automated algorithms. Our preliminary data suggest that combining readily available data in EHRs including comorbidities using machine learning techniques offers an opportunity to develop a reliable OSA phenotypic risk score that would facilitate OSA case identification in primary care.

New paradigms for identification of OSA and its subtypes centered on primary care and using data from the EHR, including comorbidities and patient symptoms, are needed to meet the high prevalence of OSA. In Aim 1, using a two-stage sampling strategy in primary care practices, we will determine the prevalence of OSA and its

subtypes, and robustly estimate the rate of OSA under-diagnosis in primary care clinics. In Aim 2, we intend to develop a root cause analysis of the underdiagnosis of OSA problem in primary care practice setting using a systems engineering framework. We will elicit the barriers and facilitators of the OSA diagnostic process within

the work system of the primary care practice setting using stakeholder engagement methods with multi-level stakeholders. Using these results, we will survey primary care providers and staff to estimate the frequency and priority rank-order of OSA diagnostic process barriers and facilitators within the primary care work system.

In Aim 3, using a novel machine learning pipeline, we will develop an efficient and accurate tool for determining OSA risk in primary care participants recruited in Aim 1 and an efficient tool to determine OSA subtype that could be employed within the context of primary care practice. This proposal will ultimately lead to increased

identification of OSA and its subtypes in primary care clinics and prioritize treatment of patients that will most benefit from therapy which may reduce the burden of significant health problems that are associated with this very common condition.