Current knowledge/Study rationale: Phenotypes of sleep apnea in relation to cardiac comorbidities is not a well-studied topic. The rational of this study is to define phenotypes of sleep apnea and see the applicability of that information for future care of patients.
Study impact: We were able to demonstrate three major phenotypes. There needs to be more phenotypic studies in sleep apnea patients.
Obstructive Sleep Apnea (OSA) is a complex disease process with a known significant association with cardiovascular diseases and the metabolic syndrome. This study aimed to define phenotypes of OSA based on sleep studies and cardiovascular comorbidities and to further investigate whether there would be any meaningful association between these disease processes. Defining phenotypes could assist in individual targeted treatments for patients with OSA.
We conducted a retrospective chart review on sleep studies between 12/6/2015 and 5/18/17 and identified 1056 adult patients. We documented all aspects of the sleep studies and then did a chart review on the identified patients in our Electronic Medical Record (EMR) to study cardiovascular disease processes of hypertension, atrial fibrillation, coronary artery bypass surgery, severity of diabetes and the presence of prior stroke.
After comparing all our data, we found that lowest saturation, baseline saturation, N1, BMI, and N3 had strong correlations with AHI. Presence of diabetes and ESS number had no correlation. Hypertension and age had moderate while Rapid Eye Movement (REM) cycle, ECG abnormalities and sleep efficiency had small correlation with Apnea-Hypopnea Index (AHI).
Only hypertension had a significant effect on the clustering. The rest of the majority of the clusters were formed by differences in sleep stages, Respiratory Disturbance Index (RDI), lowest saturation points and sleep efficiencies. Based on these results, our study did not show a significant association between the cardiac comorbidities and sleep study outcomes as clustering.
Sleep apnea, Cardiovascular disease, Cluster analysis, Sleep apnea phenotype
OSA: Obstructive Sleep Apnea; AHI: Apnea-Hypopnea Index; IRB: Institutional Review Board; ESS: Epworth Sleepiness Scale; ECG: Electrocardiogram; EMR: Electronic Medical Record; BPV: Blood Pressure Variability; RDI: Respiratory Disturbance Index
Obstructive Sleep Apnea (OSA) is a common but complex sleep disorder characterized by sleep fragmentation and hypoxia, which if left untreated may cause harmful sequelae.
It is estimated that in the United States, the prevalence of moderate to severe OSA (Apnea-Hypopnea Index ≥ 15 per hour) may be as high as 10-17% in men and 3-9% in women [1]. This prevalence will likely continue to increase as the rate of obesity rises worldwide [1,2].
Patients with OSA have a higher risk of having cardiovascular comorbidities such as hypertension, diabetes and stroke which increase patients' morbidity and mortality. One of the most impactful sequelae of OSA is its effect on the cardiovascular system and the metabolic syndrome [3,4]. A proposed mechanism of the impact of OSA in the cardiovascular system is the repetitive cycles of hypoxia and arousals that cause swings in negative intrathoracic pressure, decrease in myocardial contractility, and decrease response from the parasympathetic nervous system. This leads to increase in blood pressure and heart rate, activation of oxidative stress and systemic inflammation, and impairment of endothelial function [3].
There have been several studies that have described the association between OSA and hypertension [5,6]. Patients with both hypertension and OSA have been found with increases in urine catecholamines, suggesting increased sympathetic activity in these patients [7].
Studies have also found a link between diabetes with OSA [8]. Patients with OSA have a higher risk of having diabetes and increasing OSA severity is associated with increased likelihood of having diabetes with worse diabetic control [8,9]. In patients with the metabolic syndrome, OSA is independently associated with increased glucose and triglyceride levels, as well as markers of inflammation, arterial stiffness, and atherosclerosis [2].
An observational study demonstrated that untreated patient with severe OSA had a higher incidence of fatal cardiovascular events which included myocardial infarction, acute coronary syndrome, and stroke, than untreated patients with mild-moderate OSA [10]. It is clear that OSA is a complex disease which affects other systems, predominantly the cardiovascular, yet, OSA continues to be classified simply with the Apnea-Hypopnea Index (AHI). Few studies have been done studying OSA and other comorbid conditions [11-13]. Given the importance of proper classification of OSA, we conducted a study aimed to better classify OSA taking into account several conditions associated with cardiovascular disease that abound in these patients.
We conducted a retrospective chart review on sleep studies performed at our hospital affiliated sleep center between 12/6/2015 and 5/18/17 and identified 1056adult patients. Our study was approved by the institutional review board (IRB). As seen in (Table 1), from the sleep studies we documented AHI, lowest saturation levels, Epworth Sleepiness Scale (ESS), Electrocardiogram (ECG) abnormalities, and percent saturation less than 90%. We then did a chart review on the identified patients in our Electronic Medical Record (EMR) and documented cardiovascular disease processes which were grouped into subgroups of hypertension, atrial fibrillation, and coronary artery bypass surgery.
Table 1: Retrospective chart review on sleep studies. View Table 1
We also documented the presence and severity of diabetes and the presence of prior stroke. Hypertension was classified based on the number of medications the patient was taking for control: 1 medication = mild, 2 medications = moderate, and 3 or more medications = severe (Table 2).
Table 2: The presence and severity of diabetes and the presence of prior stroke. View Table 2
Other data collected included age, date of birth, gender, ethnicity, height, weight, and Body Mass Index (BMI). In preparing the data for analysis, variables with less than 10 cell counts were excluded from the analysis leaving a total of 15 candidate variables. We then used correlation analysis and Chi square tests to examine which variables were related to AHI as outcome.
Given the categorical nature of the data, we adopted latent class analysis as a methodology to group patients into distinct groups based on the characteristics in (Table 3). The three cluster solution from the analysis is distributed as follows by AHI Category.
Table 3: How Candidate variables are correlated with AHI. View Table 3
After comparing all our data, we found that lowest saturation, RDI (Respiratory Disturbance Index), Baseline saturation, N1, BMI, and N3 had strong correlations with AHI. Presence of diabetes and ESS score had no correlation. Hypertension and age had moderate while REM cycle, ECG abnormalities and sleep efficiency had small correlation with AHI (Table 3).
In (Table 4), we have a chi square analysis on all the studies.
Table 4: Chi-square analysis on candidate variables. View Table 4
Based on the results from correlations and chi-square tests, the final variables selected for clustering are listed below along with their categories. All variables are ordinal in nature (Table 5).
Table 5: Final variables selected for cluster analysis. View Table 5
We performed a cluster analysis using latent class analysis (Table 6).
Table 6: 3 Cluster solution distributions by AHI Category. View Table 6
The (Table 7) blow provides the probabilities of belonging to a given category for each cluster.
Table 7: Estimated Probabilities and Standard Errors. View Table 7
Based on this, we have 3 clusters with descriptions below (Table 8).
Table 8: Cluster descriptions. View Table 8
Cluster Seperation (Figure 1).
Figure 1: Cluster Separation. View Figure 1
Our study demonstrates the difficult and intricate relationship between cardiovascular comorbidities and OSA. To our knowledge this is the most comprehensive study to date that includes many cardiac comorbidities. Multiple studies [13-17] have demonstrated a relationship between obstructive sleep apnea and cardiac comorbidities but have not been able to investigate the same number of studies combined with the number of variables in our study. There are very limited number of prospective studies in the literature and one large prospective observational study with a mean follow up of 10.1 years found that when comparing healthy controls matched for age, sex and weight those with severe untreated OSA had more fatal and non-fatal cardiovascular events whereas those treated by CPAP did not different significantly from the controls [10].
Cluster analysis has also been used to help differentiate comorbidities found in obese patients; Reategui, et al. looked at 14 obesity comorbidities from 1237 discharge summaries from the i2b2 2008 Obesity dataset and found relationships between obesity, hypertension and OSA [18]. A small prospective study of 45 patients analyzed OSA's association with cardiovascular disease finding associations with severe OSA and moderate to severe coronary artery [19]. With a similar goal to our study Sweed, et al. retrospectively studied 244 patients diagnosed with OSA to assess the prevalence of associated comorbidities and the most common comorbidities they identified were obesity, hypertension, and diabetes mellitus [20]. Hypertension is a well-studied disease process that has been shown to be better controlled with CPAP [21] and our study also demonstrated moderate correlation of the severity of hypertension and AHI. Some studies have attempted to treat hypertension in OSA patients by using beta blockers instead of diuretics given the OSA triggered excessive sympathetic nervous system response [22]. Interestingly, OSA was found to increase 24 h diastolic blood pressure variability (BPV) in hypertensive patients and night time systolic in multiple studies [23-26]. High AHI, was associated with elevated diastolic blood pressure in the morning and was postulated as one symptom that could be diagnostic for OSA [27].
Finding well defined clusters in a population as big and diverse as OSA is difficult but if noted could prove fruitful. The best scenario would be to have groups of the population that could be treated in an individual manner for focused care creating a targeted algorithm that may in the long turn achieve better control and higher compliance with improved outcomes.
The strength of our study is the number of sleep studies that were looked at in detail and the multiple EMRs used within our health system. The limitation of the cluster study is that we were not able to reach a unified clustering based on differences in the levels of comorbidities. We aimed for clusters to form based on comorbidity severity and in correlation to the AHI severity, however there were comorbidities that had no association with the AHI and therefore not included in the cluster. We only used Age, baseline saturation, BMI, ECG abnormalities, hypertension, lowest sat, N1 sleep stage, N3 sleep stage, RDI, REM and sleep efficiency. Based on these variables, our study resulted in three major clusters. When all clusters are compared, we were able to demonstrate different characteristics of sleep studies but only hypertension had a significant effect on the clustering. The rest of the majority of the clusters were formed by differences in sleep stages, RDI, lowest saturation points and sleep efficiencies. Based on these results, our study did not show a significant association between the cardiac comorbidities and sleep study outcomes as clustering.
This study demonstrates how severity of sleep apnea has a direct relationship with hypertension and perhaps sleep efficiency. Nocturnal hypoxia has already been an established cause of hypertension [28] and prospective studies have been also performed to establish this phenomenon [6-10]. It is thought that the cause of systemic hypertension in patients are due to persistently elevated sympathetic nerve activity due to hypoxemia, with associated increased blood pressure and heart rate. This happens even during wakefulness and muscle sympathetic nerve activity is attenuated during apnea under hypoxic conditions [29,30] pointing out to a direct link with hypoxemia.
Baroreflex sensitivity is diminished in OSA and this leads to intermittent hypoxemia [31,32] that has also been shown to be reduced with CPAP use.
We are not fully confident that a larger study would be able to demonstrate better clustering with these disease processes. We believe that one of the explanations why we were not able to show associations between the cardiovascular disease processes and severity of sleep study is the heterogeneity of each cardiovascular disease process and we were simply looking at a cross section of the disease in whatever time frame was available in our EMR. The uncontrolled or undiagnosed diseases prior to the recording in the EMR or not recording might account for the clustering issues we faced. We believe that a prospective study with these diseases aiming at a controlled way of collecting the data might yield more uniform results. It might also yield better clustering that may yield individualized treatment algorithms for patients suffering from OSA and its associated comorbidities.