265 Connecting Electronic Medical Record Sub-phenotypes of Obstructive Sleep Apnea to Adverse Outcome Risks

Victor Borza, Raghu Upender, Wei‐Qi Wei, Bradley MalinJournal of Clinical and Translational Science 2023 · DOI

Abstract

Obstructive Sleep Apnea (OSA) is a heterogeneous disorder with varying clinical presentations and long-term risks. While previous research has identified OSA subtypes, these studies often lack the longitudinal depth provided by large-scale clinical data. This study leverages Vanderbilt’s Electronic Medical Record (EMR) database to identify clinically meaningful OSA sub-phenotypes and evaluate their associations with future adverse outcomes. Using a cohort of over 61,000 patients, including 12,516 with detailed polysomnography data, we applied clustering algorithms to define distinct patient profiles. Preliminary results identify specific clusters, such as a high-BMI/severe OSA group and a high nocturnal limb movement group. Our findings suggest that these sub-phenotypes are linked to varying risks of comorbidities like hypertension, diabetes, and cerebrovascular disease, paving the way for personalized risk assessment and treatment.

Introduction

Obstructive Sleep Apnea (OSA) is characterized by repetitive pauses in breathing during sleep, but its clinical manifestation varies significantly across the population. Some patients present with classic obesity-related airway obstruction, while others exhibit significant sleep fragmentation or periodic limb movements. Despite this diversity, clinical management often relies on a "one-size-fits-all" approach based primarily on the Apnea-Hypopnea Index (AHI).

Recent research has shifted toward identifying "sub-phenotypes" to better capture this heterogeneity. However, many existing studies are limited by short follow-up periods or small sample sizes. By utilizing the longitudinal richness of Electronic Medical Records (EMRs), we can track patient health over years, providing a more comprehensive view of how specific OSA patterns correlate with long-term health risks. This study aims to bridge the gap between sleep study data and real-world clinical outcomes.

Methods

Study Population

The study utilized Vanderbilt University Medical Center’s EMR database, identifying a cohort of 61,000 adult patients with a literature-validated definition of OSA. The median follow-up period after diagnosis was four years. A subset of 12,516 patients had comprehensive polysomnography (PSG) data integrated with EMR variables, including age at the time of the study and most recent Body Mass Index (BMI).

Analytical Approach

To identify natural groupings within the data, the research team applied several clustering methods. A key feature of this analysis was the use of fuzzy membership techniques, which allow a single patient to belong to multiple clusters to varying degrees. This approach acknowledges the biological reality that phenotypes often exist on a spectrum rather than as discrete, mutually exclusive categories.

Statistical Modeling

Once the clusters were established, the researchers planned to analyze the associations between these sub-phenotypes and adverse outcomes. Statistical tools included:

  • Logistic Regression: To assess the odds of specific comorbidities.
  • Cox Proportional Hazards Regression: To model the time-to-event for long-term outcomes.

Models were adjusted for potential confounders, including age, gender, and medication use, to isolate the impact of the OSA sub-phenotype itself.

Results

Identification of Sub-phenotypes

Preliminary clustering using PSG and EMR data revealed distinct patient profiles that align with existing literature. Two notable clusters emerged:

  1. Severe Metabolic Cluster: Characterized by high BMI, severe OSA (high AHI), and a high percentage of non-REM stage 1 sleep.
  2. Limb Movement Cluster: Defined primarily by high nocturnal limb movement indices.

Incidence of Adverse Outcomes

The study tracked the incidence of several OSA-associated conditions following diagnosis. Notable outcomes with incidence rates greater than 3% included:

  • Essential Hypertension: 43.4%
  • Hyperlipidemia: 28.8%
  • Type 2 Diabetes: 21.9%
  • Anxiety Disorder: 19.2%
  • Coronary Atherosclerosis: 14.9%
  • Cerebrovascular Disease: 7.7%
  • Pulmonary Heart Disease: 5.9%

These results demonstrate the high burden of cardiovascular and metabolic disease within the OSA population and provide a baseline for assessing how risk varies by cluster.

Discussion and Significance

The ability to map EMR-derived sub-phenotypes to specific health risks represents a significant step toward personalized medicine in sleep disorders. By identifying which patients are at the highest risk for specific outcomes—such as stroke or heart failure—clinicians can move beyond treating the AHI alone and instead focus on the patient's specific risk profile.

Limitations and Future Directions

While the preliminary results are promising, the study is ongoing. Future work will focus on:

  • Deeper EMR Integration: Incorporating laboratory results and medication histories to further refine the sub-phenotypes.
  • Causal Inference: Determining whether these sub-phenotypes are direct drivers of disease or markers of broader systemic health issues.
  • Clinical Implementation: Developing predictive tools that can be integrated into clinical workflows to assist providers in tailoring treatments.

Ultimately, this research highlights the power of "real-world" data in uncovering the complex relationships between sleep architecture and long-term health, offering a path toward more precise and effective OSA management.

Summary updated

Abstract

OBJECTIVES/GOALS: Prior work has established subtypes of OSA and linked them to risks of future adverse events but rarely with the longitudinality and richness of data available in the EMR. Our goal is to leverage EMR data identify clinically meaningful sub-phenotypes of OSA and better study how they affect risks of adverse outcomes. METHODS/STUDY POPULATION: Vanderbilt’s EMR database has over 61,000 adult patients with a literature-validated EMR definition of OSA with a median EMR follow-up period of 4 years after OSA diagnosis. Of these patients, 12,516 have fully recorded sleep study data in addition to EMR variables such as age at study and most recent BMI. We applied several clustering methods including to identify natural sub-phenotypes of OSA and assessed cluster quality. We also applied techniques which allow a single patient to belong to multiple clusters in various degrees. After selecting final clusters, we plan to analyze the associations between OSA sub-phenotypes and risks using statistical tools like logistic regression and Cox proportional hazards regression, with and without adjusting for factors such as age, gender, and certain medications. RESULTS/ANTICIPATED RESULTS: Preliminary clustering with primarily sleep study data has shown overlap with literature-described patient clusters, including a severe, high non-REM stage 1 sleep, high BMI cluster and a high nocturnal limb movement cluster. As we incorporate more EMR variables, we will select a final set of OSA sub-types. We anticipate patients in different clusters to have different risks of various adverse OSA-associated outcomes that are tracked in our EMR data. Notable outcomes with sufficient incidence rates (>3%) after OSA diagnosis include essential hypertension (43.4%), hyperlipidemia (28.8%), type 2 diabetes (21.9%), anxiety disorder (19.2%), coronary atherosclerosis (14.9%), cerebrovascular disease (7.7%), and pulmonary heart disease (5.9%). DISCUSSION/SIGNIFICANCE: If our results match anticipations, we will show how EMR data can be used to define OSA sub-phenotypes and predict patient risks of various OSA-associated outcomes. This analysis enables work in personalized risk and treatment predictions for OSA patients. By better understanding these risks, providers can better tailor treatments to patients.
BibTeX
@article{borza2023connecting,
  author = {Victor Borza and Raghu Upender and Wei‐Qi Wei and Bradley Malin},
  title = {{265 Connecting Electronic Medical Record Sub-phenotypes of Obstructive Sleep Apnea to Adverse Outcome Risks}},
  journal = {Journal of Clinical and Translational Science},
  year = {2023},
  doi = {10.1017/cts.2023.324},
}
Figure