Obstructive Sleep Apnea Patients With Atrial Arrhythmias Suffer From Prolonged Recovery From Desaturations

Objective: We aimed to investigate how acute and long-term effects of atrial arrhythmias affect the desaturation severity and characteristics determined from the oxygen saturation signal in obstructive sleep apnea (OSA) patients. Methods: 520 suspected OSA patients were included in retrospective analyses. Eight desaturation area and slope parameters were calculated from blood oxygen saturation signals recorded during polysomnographic recordings. Patients were grouped based on whether they had previously diagnosed atrial arrhythmia (i.e., atrial fibrillation (AFib) or atrial flutter) or not. Furthermore, patients with a previous atrial arrhythmia diagnosis were sub-grouped based on whether they had continuous AFib or sinus rhythm during the polysomnographic recordings. Empirical cumulative distribution functions and linear mixed models were utilized to investigate the connection between diagnosed atrial arrhythmia and the desaturation characteristics. Results: Patients with previous atrial arrhythmia diagnosis had greater desaturation recovery area when the 100% oxygen saturation baseline reference was considered (β = 0.150--0.127, p ≤ 0.039) and more gradual recovery slopes (β = −0.181 to −0.199, p < 0.004) than patients without a previous atrial arrhythmia diagnosis. Furthermore, patients with AFib had more gradual oxygen saturation fall and recovery slopes than patients with sinus rhythm. Conclusion: Desaturation recovery characteristics in the oxygen saturation signal contains essential information about the cardiovascular response to hypoxemic periods. Significance: More comprehensive consideration of the desaturation recovery section could provide more detailed information about OSA severity, for example when developing new diagnostic parameters.

included in retrospective analyses. Eight desaturation area and slope parameters were calculated from blood oxygen saturation signals recorded during polysomnographic recordings. Patients were grouped based on whether they had previously diagnosed atrial arrhythmia (i.e., atrial fibrillation (AFib) or atrial flutter) or not. Furthermore, patients with a previous atrial arrhythmia diagnosis were sub-grouped based on whether they had continuous AFib or sinus rhythm during the polysomnographic recordings. Empirical cumulative distribution functions and linear mixed models were utilized to investigate the connection between diagnosed atrial arrhythmia and the desaturation characteristics. Results: Patients with previous atrial arrhythmia diagnosis had greater desaturation recovery area when the 100% oxygen saturation baseline reference was considered (β = 0.150-0.127, p ≤ 0.039) and more gradual recovery slopes (β = −0.181 to −0.199, p < 0.004) than patients without a previous atrial arrhythmia diagnosis. Furthermore, patients with AFib had more gradual oxygen saturation fall and recovery slopes than patients with sinus rhythm. Conclusion: Desaturation recovery characteristics in the oxygen saturation signal contains essential information about the cardiovascular response to hypoxemic periods. Significance: More comprehensive consideration of the desaturation recovery section could provide more detailed information about OSA severity, for example when developing new diagnostic parameters.

I. INTRODUCTION
C ARDIORESPIRATORY and cardiovascular systems are linked both functionally and through several anatomical structures and pathophysiological interactions [1], [2]. These connections are highly sensitive for the autonomic nervous system (ANS) activations as well as cardiovascular and respiratory disturbances [1], [2]. For example, hypoxemia, caused by respiratory disruptions, activate several ANS mechanisms leading to cardiac responses [3], [4]. In turn, cardiac rhythm disturbances impact respiration activity and function [5]. Consequently, cardiovascular diseases (CVDs) and respiratory disorders often appear simultaneously and have a significant association; however, the association is not completely understood at the pathophysiological level [2], [6]. Obstructive sleep apnea (OSA) is a sleep-related breathing disorder that affects globally nearly a billion people and disrupts cardiorespiratory activations [3], [7]. OSA-related complete (apnea) and partial (hypopnea) repetitive upper airway collapses can lead to intermittent hypoxemias (i.e., repetitive desaturations), reoxygenations, substantial heart rate, and blood pressure variation as well as intrathoracic pressure fluctuations. These consequences can further trigger atrial arrhythmias [8], [9]. Thus, OSA patients have over twice the risk for atrial arrhythmias compared to individuals without OSA, and the risk increases with increasing OSA severity [10], [11]. Several studies have investigated the pathophysiological pathways of how intermittent hypoxemias lead to cardiac arrhythmias in OSA patients [6], [9]. However, even though atrial arrhythmias are associated with flow abnormalities and endothelial dysfunction [12], it is not fully understood how atrial arrhythmias are related to the severity and morphology of hypoxemic episodes in OSA patients.
Currently, the apnea-hypopnea index (AHI) is used to assess the severity of OSA, estimating the average number of respiratory events per hour of sleep [13]. However, several studies show that AHI is not the best parameter in OSA severity estimation, as it does not consider the characteristics of the breathing disruptions [14], [15], [16], [17]. Furthermore, it has been shown that different oxygen desaturation thresholds for hypopnea scoring lead to different associations between OSA and CVDs [18], [19]. Instead, the desaturation event characteristics have been shown to describe the connection between OSA severity and OSA-related outcomes better compared to the AHI [15], [20], [21]. Moreover, desaturation areas, characterizing the depth and the duration of desaturations, are associated with severe OSA-related health consequences, such as increase in mortality rate and cardiovascular morbidity [22], [23]. Currently, desaturations are determined as a ≥ 3% drop in the oxygen saturation (SpO 2 ) signal [13]. However, no standard scoring rules exist for the desaturation endpoint placement, even though differences in the desaturation endpoints substantially affect the value of related desaturation area parameters [24].
Based on the prominent connection between cardiac arrhythmias and intermittent hypoxemias, we hypothesized that suspected OSA patients with a history of atrial arrhythmias (atrial fibrillation or flutter) have more severe desaturation events. Secondly, we hypothesized that OSA patients with previous atrial arrhythmia diagnosis have slower reoxygenation processes leading to larger desaturation recovery areas. Thus, our first aim was to investigate how previously diagnosed atrial arrhythmia in OSA patients is related to desaturation severity and morphology estimated from SpO 2 signal. Additionally, we investigated the acute effects of atrial arrhythmias on the severity and morphology of desaturation events in OSA patients with a previous atrial arrhythmia diagnosis.

A. Dataset
The initial study population consisted of 933 consecutive patients who underwent complete overnight polysomnography (PSG) between January 2015 and December 2017 at the Sleep Disorders Centre, Princess Alexandra Hospital, (Brisbane, Australia) due to clinical suspicion of OSA. PSGs were conducted and scored retrospectively using the Compumedics Grael acquisition system and the Compumedics ProFusion PSG 4 software (Compumedics, Abbotsford, Australia). Recordings were scored manually in conformity with the American Academy of Sleep Medicine 2012 scoring rules [13]. The specific scoring protocol of apneas, hypopneas, and desaturations are delineated in earlier publications [20], [25]. The Institutional Human Research Ethics Committee of the Princess Alexandra Hospital approved the use of this retrospective data (HREC/16/QPAH/021 and LNR/2019/QMS/54313).
Out of the 933 patients, 385 were excluded based on incomplete PSG recordings, demographic information, or desaturation event information, having AHI < 5, chronic obstructive pulmonary disease (COPD), or having suffered a cardiac failure. The exclusion criteria are detailed in Fig. 1, (Exclusion, Step 1). The patients with COPD or cardiac failure were excluded as overlapping COPD and OSA, or overlapping cardiac failure and OSA, have a deteriorative effect on cardiorespiratory function and oxygen saturation [5], [26]. For the remaining 548 patients, nocturnal atrial arrhythmias were evaluated based on nocturnal electrocardiogram (ECG) signals from the overnight PSG  (28 patients in total) were excluded from further analysis ( Fig. 1, Exclusion, Step 2). The final population of 520 patients (with a total of 58244 desaturation events) was divided into two groups based on whether the patient was previously diagnosed with atrial arrhythmia (i.e., AFib or atrial flutter) or not. Patients with a previous diagnosis of atrial arrhythmia were further divided into two subgroups, based on whether the patient had SR or continuous AFib during the overnight PSG. Comorbidities were reported based on the medical history of the patient and an interview in the sleep clinic. Patient demographics and polysomnographic information are presented in Table I. A flowchart describing the subgrouping of the patients is presented in Fig. 1.

B. Signal Processing and Calculated Desaturation Area and Slope Parameters
Eight different desaturation parameters were calculated from the SpO 2 signal: desaturation area, fall area, recovery area, desaturation area with a 100% baseline (i.e., desaturation area with a baseline reference at 100%), fall area with a 100% baseline, recovery area with a 100% baseline, fall slope, and recovery slope (Fig. 2). The desaturation areas with and without a 100% baseline were integrated over the whole desaturation, the fall areas with and without a 100% baseline were integrated from the desaturation onset to the nadir, and the recovery area with and without a 100% baseline from the nadir to the endpoint of the desaturation (Fig. 2). All desaturation area parameters were integrated numerically utilizing the trapezoidal method. The fall slope was calculated by dividing the saturation difference between the onset and the nadir with the time difference between these points and the recovery slope was calculated similarly between the nadir and the endpoint of desaturation (Fig. 2). In addition, median values of desaturation area and slope parameters were calculated for each patient.

C. Statistical Analysis
Differences in the desaturation area and slope parameters between the patients with and without diagnosed atrial arrhythmia were investigated using empirical cumulative distribution functions (CDFs) and linear mixed model analysis. The CDFs of the desaturation area and slope parameters were calculated separately in patients with and without diagnosed atrial arrhythmia using the medians of each patient. In addition, the CDFs of desaturation area and slope parameters were calculated separately in patients with continuous AFib and normal SR using all desaturation events. Statistical significance of the differences between the CDFs were calculated using a two-tailed Kolmogorov-Smirnov test.
In linear mixed model analyses, three different models were constructed. Model 1 was unadjusted, Model 2 was adjusted with age, gender, body mass index (BMI), and smoking, and   Model 3 was adjusted with the AHI, type 2 diabetes mellitus (T2DM), and hypertension. Desaturation area and slope parameters were inputted separately as response variables in each model. The parameter distributions were logarithmically transformed to ensure normal distribution before linear mixed model analyses. The previous diagnosis of atrial arrhythmia and the adjusted variables were used as fixed effects and the patient research numbers were used as a random effect. Furthermore, the normality of residuals of the linear mixed model analyses was verified visually. The limit of statistical significance was set at p < 0.05. All calculations and statistical analyses were performed using MATLAB software (version R2019b; MathWorks; Natick, MA, USA).

A. Cumulative Distribution Functions
Significantly (p ≤ 0.032) larger desaturation and recovery areas were observed in patients with diagnosed atrial arrhythmia compared to patients without a previous atrial arrhythmia diagnosis when the 100% baseline was considered (Fig. 3). Furthermore, the recovery slope was significantly (p = 0.045) steeper in patients without a previous atrial arrhythmia diagnosis compared to patients with a previous atrial arrhythmia diagnosis (Fig. 3). In the CDFs of the subgroups of arrhythmia patients, all other area parameters were larger, but the recovery area with the 100% baseline was smaller in AFib patients compared to patients with SR (p < 0.001) (Fig. 4). In addition, substantially more gradual fall and recovery slopes were observed in patients with continuous AFib (p < 0.001) (Fig. 4).

B. Descriptive Statistical Results of the Desaturation Area and Slope Parameters
The greatest differences between OSA patients with and without previous atrial arrhythmia diagnosis were observed in the desaturation area parameters when the 100% baseline was considered. That is, patients with a previous atrial arrhythmia diagnosis had significantly larger desaturation areas (Table II). Furthermore, patients with previously diagnosed atrial arrhythmia had more gradual fall slope and recovery slopes than patients without a previous atrial arrhythmia diagnosis (Table II). When the comparison was done between patients with SR and continuous AFib during the PSG recordings, AFib patients were Authorized licensed use limited to the terms of the applicable license agreement with IEEE. Restrictions apply.
observed to have more gradual desaturation slopes and larger desaturation areas (Table II). As the only exception, the recovery area with the 100% baseline was significantly smaller (p < 0.05) in patients with continuous AFib (Table II).

C. Linear Mixed Model Analyses
With the unadjusted model, a previous atrial arrhythmia diagnosis in OSA patients was associated with an increase in desaturation area parameters when the 100% baseline reference was considered (β ≥ 0.133, p ≤ 0.031, Table III). However, this association diminished in the adjusted models, as recovery area with the 100% baseline was the only statistically significant predictor for the prevalence of an atrial arrhythmia diagnosis in Model 3 (β = 0.127, p = 0.039) (Table III). In contrast, the predictive power of the fall and recovery slopes was higher in the adjusted models compared to the unadjusted model. That is, prevalence of atrial arrhythmia was associated with a decrease  (Table III).

IV. DISCUSSION
In this study, we examined how atrial arrhythmias are related to the severity and morphology of oxygen desaturations in OSA patients. We found that patients with a previous atrial arrhythmia diagnosis had significantly greater recovery area when a 100% SpO 2 baseline reference was considered and more gradual desaturation recovery slope. Moreover, patients who had continuous AFib during the PSG recordings had substantially more gradual desaturation slopes than patients with SR. These results show that the desaturation recovery section and desaturation rates determined from SpO 2 signal contain essential information about the pathophysiological impacts of breathing cessations in OSA patients. Thus, this study suggests that the desaturation events should be scored to include also the desaturation recovery section, not only until the nadir of desaturation. This would lead to utilization of more detailed information on the desaturations in OSA severity estimation and diagnostics. Furthermore, results illustrate that the 100% SpO 2 baseline reference should be considered in the desaturation area calculations as it contains the information on the baseline of the patient's nocturnal oxygen saturation level.
In linear mixed model analysis, the prevalence of atrial arrhythmia diagnosis was associated with an increase in the recovery area with the 100% baseline parameter (Table III). Furthermore, the prevalence of having previously diagnosed atrial arrhythmia was associated with a decrease in the recovery slope parameter. These results support our hypothesis that the reoxygenation from desaturations is prolonged in OSA patients with a previous atrial arrhythmia diagnosis compared to patients without a previous atrial arrhythmia diagnosis. The possible causes behind these findings are manifold. First, the irregular rhythm can cause an acute impairment of the cardiorespiratory regulation and function. Second, OSA-related recurrent desaturations may promote the development of AFib by inducing electrical and structural cardiac remodelling as well as autonomous system dysregulation [6], [28]. Therefore, the homeostasis of the cardiovascular system may be impaired and lack the adequate response to the changes of peripheral oxygenation in OSA patients with a previous atrial arrhythmia diagnosis. These findings provide interesting new insight in understanding the pathophysiological cascade seen in patients with OSA and atrial arrhythmias. Therefore, a more detailed assessment of the clinical history of the patients, the overnight ECG, and the pathophysiology of desaturations, would benefit the clinical decision making and OSA severity assessment.
Previous studies have shown that the desaturation severity describes the severity of OSA and its connection with the related long-and short-term outcomes better than the AHI [14], [15], [25], [29]. For example, we have previously shown that desaturation severity parameters are significant predictors for cardiovascular comorbidity and other health consequences of OSA such as severe excessive daytime sleepiness (EDS) and impaired psychomotor vigilance task (PVT) performance [20], [25], [29]. Therefore, the present results together with the previous findings support the conclusion that the more detailed examination of desaturation characteristics and recovery section in OSA diagnostics could be beneficial to estimate the true severity of OSA. Currently, no uniform guidelines exist for scoring desaturations endpoints [13]. The results of the current study suggest that both the fall, and the recovery parts should be included in the scored desaturations. A uniform way to score desaturations could contribute to achieve more coherent information related to desaturations and assist in finding the best parameters to diagnose OSA.
In accordance with our hypothesis, the present results showed that OSA patients with previously diagnosed atrial arrhythmia have larger desaturation areas. Interestingly, this phenomenon appeared only when the 100% baseline reference in oxygen saturation was considered, and against our hypothesis, the connection between the desaturation severity and a previous diagnosis of atrial arrhythmia was not observed when the areas were restricted to the desaturation onset level (Fig. 3, Tables II and III). In addition, patients with previously diagnosed atrial arrhythmia were observed to have deeper desaturations when either the onset level or the 100% SpO 2 baseline was considered compared to patients without previously diagnosed atrial arrhythmia (Table I). It has been shown that lower levels in the nocturnal oxygen saturation and prevalence of intermittent hypoxemias have major effects on cardiac and vascular dysfunction and are one of the most important pathophysiological consequences of OSA [6], [30]. Furthermore, desaturation area-based parameter which includes 100% SpO 2 baseline has been associated with predicting CVD outcomes of OSA [31]. Together with previous knowledge and results of this study, we support this finding in the case of patients with atrial arrhythmia when developing parameters assessing OSA severity. That is, the baseline reference could be set at 100% SpO 2 value in order to factor differences in baseline blood oxygenation as well as the relative decrease in blood oxygenation.
To study the acute influence of AFib on the desaturations in OSA patients with a history of atrial arrhythmias we further compared patients having continuous AFib to patients having SR during the PSG recordings. Results showed that the patients with AFib had greater desaturation areas. As an exception, the recovery area with 100% SpO 2 baseline was smaller in the patients with continuous AFib (Table II, Fig. 4). However, the patients with continuous AFib had substantially more gradual desaturation slopes (Fig. 4) implying noticeably longer desaturations compared to the patients who had SR during the PSG (Table I). Previously, AFib has been associated with blood flow abnormalities and endothelial dysfunction [12]. These results are in line with the previous findings, showing that oxygenation changes in the peripheral vascular system are highly prolonged in patients suffering AFib during the night. In turn, it has been shown that OSA patients have impaired cardiac response causing acute hypoxia in the body and cerebral circulation [6], [30].
These findings indicate that this phenomenon is exceedingly severe in OSA patients suffering from acute AFib. It is acknowledged that OSA patients with cardiac comorbidity have a higher risk for nocturnal sudden death compared to patients without OSA [32]. Thus, these findings suggest that the identification of OSA patients having acute atrial arrhythmias in the early stage is highly important to reduce these more severe health consequences. For example, as the patients with SR had faster desaturation rates than patients with AFib, the results of this study reveal that rhythm control might be better than rate control in OSA patients suffering from acute AFib. Thus, also the ECG signal measured during PSG should always be analysed in detail for a more comprehensive view of the status of the patient. However, OSA is commonly diagnosed with home sleep apnea testing which can be conducted without ECG making the detection of atrial arrhythmias challenging. The present results suggest that it could be possible to differentiate between OSA patients with and without atrial arrhythmias based in more comprehensive analysis of the SpO 2 signal. Furthermore, it has been shown that detecting arrhythmias and OSA are feasible using ambulatory devices utilizing arterial tonometry measurement alongside oxygen saturation, indicating potential for detailed, at-home diagnosis [33], [34].
In the linear mixed models, the connection between desaturation areas and previously diagnosed atrial arrhythmias diminished when adjusted with demographic parameters. We acknowledge that the study population was relatively old (median = 53.3 years), particularly in the subgroup with a previous diagnosis of atrial arrhythmia (median = 65.0 years). It has been shown that the severity of desaturation increases with increasing age [35]. Furthermore, age is one of the major risk factors for the prevalence of atrial arrhythmias [36]. These, together with the biased sample size (n = 47 vs. n = 473, respectively) could explain these results (i.e., decreased β-coefficients and increased p-values) in the linear mixed model analysis when considering desaturation area parameters.
This study has some limitations. First, the patient subgrouping was done based on whether the patient had previously diagnosed atrial arrhythmia, and we had no information on whether the patient had paroxysmal or persistent AFib or atrial flutter during the primary atrial arrhythmia diagnosis. Furthermore, the information about the possible treatments and interventions for atrial arrhythmias were not available. Moreover, we lacked the information about the time between the atrial arrhythmia diagnosis and the PSG. It is acknowledged that different arrhythmia types might have different physiological consequences and responses to intermittent hypoxemias [6], [9]. However, in this dataset most of the arrhythmias observed in ECG signals revealed AFib. Therefore, the results can be generally associated with AFib and a more detailed investigation of other types of arrhythmias and their association with desaturation severity is warranted. Second, we acknowledge including patients with a history of multimorbidity as a limitation of the present study since several comorbidities can affect oxygen saturation levels and cardiac function. However, as OSA patients are usually multimorbid, excluding all possible comorbidities would have extensively reduced the number of patients. Therefore, only the most common and potential comorbidities, i.e., COPD and cardiac failure were considered in exclusions. Third, the desaturation values at start points and endpoints of desaturations were not evaluated. However, these values could give valuable information on how well patients recover from desaturations. For example, it is possible that the oxygen saturations do not always recover to the baseline level of the desaturation onset. As this aspect was not considered in this study, it warrants further investigation in the future. Finally, as there are no clinical guidelines for scoring desaturations (besides the minimum depth requirement), some desaturations were scored from the onset to the nadir in this data set. Those desaturations were excluded from analyses to enable the evaluation of the characteristics of fall and recovery areas. However, this study included around 58 000 desaturation events and the number of such events was only 693. Thus, the exclusion of these 693 events has negligible effects on the results.

V. CONCLUSION
This study revealed that acute and long-term effects of atrial arrhythmias are connected prolonged changes in peripheral SpO 2 levels in OSA patients. This was particularly highlighted in the recovery section of the desaturation. Therefore, both the decrease and the recovery sections of desaturations in SpO 2 signal should be considered in more detail in OSA severity assessment and diagnostics. Thus, uniform desaturation scoring protocols and rules are required to allow the determination of the most appropriate diagnostic parameters beyond the AHI.