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:
- Severe Metabolic Cluster: Characterized by high BMI, severe OSA (high AHI), and a high percentage of non-REM stage 1 sleep.
- 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