PDF
ePub
Citation
Print
INTRODUCTION
Obstructive sleep apnea (OSA) is considered to be a significant risk factor for cardiovascular morbidity12; the main pathophysiologic mechanisms include cardiovascular autonomic dysfunction, which is predominantly represented by sympathetic overactivation.34 There is growing evidence that OSA is associated with cerebral small-vessel diseases,56 including cerebral white-matter changes (WMC),789 which are associated with an increased risk of cerebrovascular events and vascular cognitive impairments.10 However, the mechanisms underlying cerebral WMC in patients with OSA are not completely understood.
Heart-rate variability (HRV) has been widely analyzed as a noninvasive method for evaluating cardiovascular autonomic conditions in various conditions.11 Among the frequency-domain HRV indices, the low-frequency (LF) component is considered to represent both sympathetic and parasympathetic activity, while the high-frequency (HF) component is accepted as a marker of parasympathetic activity. The LF/HF ratio is considered to reflect sympathovagal balance, with a higher LF/HF indicating a predominance of sympathetic activity and a lower LF/HF suggesting parasympathetic predominance.12 It is well document that sympathetic activity during rapid-eye-movement (REM) sleep is increased during physiologic sleep,13 and excessive sympathetic activity during both wakefulness and sleep has been reported in OSA patients based on HRV studies.14 It has recently been demonstrated that morbidity and mortality can be predicted more accurately using several nonlinear HRV indices than standard linear parameters in certain diseases, including cardiovascular disease.1516
We previously reported a significant association between moderate-to-severe OSA and indicators of WMC, including an increased burden of white-matter hyperintensities, asymptomatic lacunar infarctions, cerebral microbleeds, or perivascular spaces.6 Although WMC are suggested to be associated with increased sympathetic activity, and HRV analyses are useful for monitoring sympathetic activities during sleep stages, HRV analyses in association with WMC during sleep—including in OSA patients—are lacking.
In the current study we assessed linear and nonlinear HRV indices according to different sleep stages in order to identify the cardiovascular autonomic characteristics during sleep that are associated with cerebral WMC in OSA.
METHODS
Patients
In total, 135 subjects were analyzed in this study (Fig. 1). Among the patients who visited the Sleep Center of Ewha University Medical Center and underwent overnight polysomnography (PSG) between April 2005 and May 2014, we selected 185 patients in whom 1.5- or 3.0-T brain FLAIR and T2-weighted MRI images were obtained within 1 year before or after PSG. The indication for performing PSG was the presence of at least one OSA-related symptom (e.g., witnessed loud snoring between apnea episodes, episodes of gasping for air, or choking), sleep fragmentation/insomnia, nonrefreshing sleep, and being in the high-risk category for OSA according to the Berlin Questionnaire, as described previously.61718 Brain MRI was routinely recommended in our sleep center for evaluating underlying organic brain lesions.
Based on the obtained PSG results, 122 of the 185 patients were diagnosed with OSA, of which 64 had WMC on brain MRI. Thirty of the patients without OSA who had WMC were excluded since they were considered not appropriate for inclusion in a control group when investigating the impact of OSA on the development of WMC. Subjects were excluded from the HRV analysis if the artifact-free and arousal-free periods of their N1, N2, or REM sleep were less than 5 min on PSG. Finally, 60 patients were assigned to the OSA with WMC (OSA+WMC) group, 44 to the OSA without WMC (OSA−WMC) group, and 31 subjects to the control group. This study was approved by the Institutional Review Board of our hospital (IRB No. 11-21-016). The need to obtained informed consent from the patients was waived due to the retrospective design and observational nature of the study.
MRI protocol and assessment of white-matter changes
The brain MRI protocol applied in this study has been described in detail previously.619 Brain MRI image slices were acquired using the following parameters: for FLAIR, repetition time (TR)/echo time (TE)=12,000 ms/120 ms, pixel spacing=0.449 mm/0.449 mm, field of view (FOV)=183×230 mm2, and slice thickness=5 mm; for T2-weighted images, TR/TE=15,000 ms/90 ms, pixel spacing=0.240 mm/0.240 mm, FOV=176×220 mm2, and slice thickness=5 mm. The extent of WMC was determined on brain FLAIR and T2-weighted MRI images according to the modified Fazekas scoring system20: a Fazekas score of ≥1 was considered to indicate WMC, while score of 0 indicated the absence of WMC. The presence of WMC was independently investigated by two neurologists (T.J.S. and J.H.P.) while they were blinded to clinical information. Consensus was reached in cases of discrepancy regarding the presence of WMC.
Polysomnography
PSG was performed in accordance with a previously described protocol.618 Overnight PSG was performed with a comprehensive device (TWin® PSG Clinical Software, Grass Technologies, West Warwick, RI, USA) at the Sleep Center of Ewha University Medical Center. The electrocardiogram (ECG) was recorded at a sampling rate of 256 Hz. The PSG data were initially scored by an experienced sleep technician and then reviewed thoroughly by expert sleep physicians (H.W.L. and J.H.P.) following the 2014 American Academy of Sleep Medicine manual for sleep scoring and respiratory event criteria.21
Briefly, apnea episodes were defined when the airflow was reduced to ≥90% of the baseline values for at least 10 sec, and they were further classified as either the obstructive type if respiratory effort was noted on either the chest or abdominal belt channel, or the central type if no respiratory effort was noted. Hypopnea episodes were defined as a ≥30% reduction of airflow for at least 10 sec and accompanied by a decrease of ≥3% in oxygen saturation (SaO2) or arousal. The apnea-hypopnea index (AHI) was calculated by averaging the total number of obstructive apnea episodes and hypopnea episodes per hour of sleep, with OSA being diagnosed when AHI ≥5.
Assessment of heart-rate variability
For each patient we carefully selected artifact-free, apnea-free, awakening-free, 5-min ECG samples from each sleep stage: N1, N2, REM sleep, and wakefulness. ECG samples were selected from the earliest period within a particular sleep stage, and for the wakefulness period we analyzed a 5-min segment at the beginning of the PSG recording that was free of ectopic beats and artifacts. For the HRV analysis according to different sleep stages, one 5-min segment was analyzed for each sleep stage, to give a total of three 5-min segments for each patient. The HRV analysis during the entire sleep period utilized the entire ECG data set recorded between the onset and the end of sleep on PSG.
The ECG data were analyzed using Kubios HRV software.12 The RR variability was analyzed in both the time and frequency domains. Each RR interval was calculated using Welch's periodogram method. Mean RR intervals, the standard-deviation of normal to normal RR intervals (SDNN), and the root mean square of successive differences (RMSSD) were acquired as time-domain indices. Spectral components of HRV were classified as very low frequency (VLF, <0.04 Hz), LF (0.04–0.15 Hz), and HF (0.15–0.4 Hz). LF- and HF-variability components were calculated in both power and normalized units (LFnu and HFnu, respectively). LFnu and HFnu were calculated as LFnu=LF/(total power−VLF)×100 and HFnu=HF/(total power−VLF)×100.
The nonlinear properties of HRV were analyzed using Poincaré plots, detrended fluctuation analysis (DFA), approximate entropy (ApEn), and sample entropy (SampEn). A Poincaré plot is a graphical presentation of the correlation between consecutive RR intervals, and the standard-deviations of the points perpendicular to and along the line of identity (the line where RRj=RRj+1) are denoted by SD1 and SD2, respectively. SD1 represents the short-term beat-to-beat variability, and SD2 measures the long-term variability.12 SD12 represents the relationship between these components, which is the ratio of short-term variation to long-term variation (i.e., SD1/SD2).1522 DFA measures correlations within the data over different time scales.23 DFA1 (α1) and DFA2 (α2) are obtained from the plot by default within ranges of 4–16 beats and 16–64 beats, which represent short-term and long-term fluctuations, respectively. For the entropy-derived measures of ApEn and SampEn, lower values of entropy reflect data that are more predictable, while more-random and complex data are described by higher values of entropy.2425
Statistical analysis
Clinical variables were compared among the groups using the chi-square test for categorical variables and analysis of variance (ANOVA) for continuous variables. The Tukey honestly significant difference (HSD) test was used for post-hoc analysis of ANOVA. HRV indices were compared among the groups using ANOVA with the post-hoc Tukey HSD test for three-group comparisons, and the independent t-test for comparisons of control and combined OSA groups. Analysis of covariance (ANCOVA) was performed to control for the effect of age on HRV, and the Bonferroni test was applied for post-hoc comparisons. In the subgroup analysis of OSA patients selected by specific age ranges, the OSA+WMC and OSA−WMC groups did not conform to a normal distribution, and so the Mann-Whitney test was used to compare the mean values. The statistical analyses were conducted using SPSS software version 13.0 (SPSS Inc., Chicago, IL, USA), and p<0.05 was considered statistically significant.
RESULTS
Demographic characteristics and polysomnography findings
Age, sex, and body mass index (BMI) differed among the three groups (Table 1). Patients were significantly older in the OSA+WMC group (age 66.8±9.6 years, mean±standard deviation) than in the OSA−WMC group (age 49.1±10.2 years) and the control group (age 47.8±10.7 years). BMI was significantly lower in the control group than in either of the OSA groups. The incidence of hypertension and diabetes differed across the groups, being highest in the OSA+WMC group.
Sleep characteristics including daytime sleepiness, total sleep time, sleep latency, and percentages of time in the REM, N2, and N3 sleep periods did not differ significantly between the groups, with the exception of the percentage of N1 sleep being higher in the OSA+WMC group than in the control group (Table 1). The arousal index, AHI, REM AHI, non-REM (NREM) AHI, and respiratory disturbance index were significantly higher in both the OSA+WMC and OSA−WMC groups than in the control group, as expected. The minimum SaO2 differed significantly among the groups, with the minimum SaO2 being markedly lower in the OSA+WMC group than in the control group.
Heart-rate variability findings
Several indices in the time-domain and frequency-domain analyses of HRV differed significantly among the three groups (Table 2). Since most of the HRV parameters—including the frequency-domain indices and nonlinear indices—are significantly influenced by increasing age,2627 and patients in the OSA+WMC group were significantly older than those in the other groups, we performed ANCOVA to control for the effect of age on HRV parameters. Applying this age adjustment resulted in the differences in the time-domain and frequency-domain indices no longer being significant.
In terms of nonlinear HRV indices, SD12 was significantly higher and DFA1 was markedly lower in the OSA+WMC group than in the OSA−WMC and control groups during NREM sleep, REM sleep, and the entire sleep period (Table 2). ApEn was significantly lower in the OSA+WMC group during REM sleep.
When ANCOVA was performed with age controlled as a covariate, SD12 during the entire sleep period was the only HRV parameter that differed significantly among the three groups (p=0.045). However, DFA1 tended to be lowest in the OSA+WMC group during NREM sleep (p=0.051) (Table 2).
Subgroup analysis in subjects without hypertension or diabetes
In order to minimize the effect of confounding factors on HRV indices, we performed a subgroup analysis in subjects who did not have hypertension or diabetes, comprising 24, 33, and 18 subjects in the control, OSA−WMC, and OSA+WMC groups, respectively. ANCOVA was performed with age controlled as a covariate. SD12 during the entire sleep period differed significantly among the groups, being 0.41±0.13, 0.50±0.16, and 0.60±0.21 in the control, OSA−WMC, and OSA+WMC groups, respectively (p=0.028). DFA1 during NREM sleep was also significantly different among the groups, being 1.15±0.37, 0.97±0.28, and 0.86±0.36 in the control, OSA−WMC, and OSA+WMC groups, respectively (p=0.025) (Table 2).
Subgroup analysis in selected age groups
Since more than two-thirds of the general population older than 65 years have WMC,28 we selected middle-aged patients (age 45–64 years) for inclusion in an additional subgroup analysis, involving 18 patients in each of the OSA−WMC and OSA+WMC groups. The mean AHI was higher than 20 in both groups, which indicates that most of these patients had moderate-to-severe OSA. The demographic characteristics and sleep study parameters did not differ significantly between these two groups (Table 3).
In the HRV analysis, none of the frequency-domain indices differed between the OSA−WMC and OSA+WMC groups (data not shown). Regarding the nonlinear indices, ApEn during REM sleep was significantly lower in the OSA+WMC group (0.79±0.25) than in the OSA−WMC group (0.94±0.23) (p=0.031) (Fig. 2A and Table 3). Receiver operating characteristics analysis was performed to assess the effectiveness of ApEn for differentiating OSA+WMC from OSA−WMC. The value of 0.71 for the area under the curve indicated a diagnosis of OSA+WMC. When ApEn was lower than 0.85, the sensitivity, specificity, positive predictive value, and negative predictive value were 55.6, 77.8, 71.4, and 63.6%, respectively (Fig. 2B). None of the other nonlinear indices differed significantly between the two groups (data not shown).
DISCUSSION
With the aim of identifying the cardiovascular autonomic characteristics associated with cerebral WMC in OSA patients, we performed HRV analysis based on sleep stages obtained from PSG in a large number of OSA patients and control subjects. The frequency-domain indices did not exhibit any meaningful differences among the study groups. The nonlinear index of SD12 during the entire sleep period was increased in the OSA+WMC group, which was the only significant difference observed after age adjustment. Additionally, DFA1 during NREM sleep tended to be lower in the OSA+WMC group. These indices were altered regardless of the presence of hypertension or diabetes. In subgroup analysis of middle-aged patients (age 45–64 years) with moderate-to-severe OSA, ApEn during REM sleep was significantly lower in the OSA+WMC group than in the OSA−WMC group. Overall our findings suggest that changes in nonlinear HRV indices reflect sympathetic overactivity in the OSA+WMC group, which might be related to the pathophysiology of WMC in OSA patients.
Changes in HRV indices according to different sleep stages have been reported both in healthy subjects and in patients with OSA.13 Parasympathetic activity normally increases as a subject approaches a deeper sleep stage during NREM sleep, resulting in increases in the HF component and decreases in the LF component as well as in the LF/HF ratio. The opposite alterations occur during REM sleep and wakefulness, characterized by the predominance of sympathetic activity, resulting in increases in the LF component and LF/HF and a decreased HF component.1113 Previous HRV analyses in OSA patients revealed an overall predominance of sympathetic activity during either wakefulness or sleep, characterized by lower total variability and increases in the LF component and LF/HF, and a decrease in the HF component.4112429
Our data differed somewhat from previous findings for HRV changes in frequency-domain indices in OSA patients during sleep.3031 The frequency-domain HRV indices did not differ among the groups in our study. Analyzing frequency- domain HRV indices in OSA patients is especially challenging due to the presence of repetitive apnea episodes, arousals, and leg movements, which induce biologic noise and can modify the findings of HRV analyses.11 We consider these factors to be the main reasons for the inconsistencies in findings for frequency-domain HRV indices in OSA patients between previous studies,3132 and the current study.
Nonlinear HRV indices may be superior for assessing autonomic cardiovascular regulation in OSA patients. Many of the heart-rate fluctuations occur over a broad frequency range, showing complex, and irregular variability. This suggests that the mechanisms involved in cardiovascular regulation occur in a nonlinear fashion over a long time period, and therefore nonlinear methods will be superior to conventional (linear) measurement techniques for detecting alterations in heart-rate dynamics.33 Several studies have shown that nonlinear indices are more effective than conventional indices for predicting the condition of the heart and the prognosis in cardiac patients.15162534 Moreover, nonlinear indices are known to be less affected by artifacts,23 which makes them more appropriate for use in OSA patients who intrinsically display many artifacts on the ECG due to recurrent arousals.
SD12 during the entire sleep period was significantly increased and DFA1 during NREM sleep had a tendency to decrease in the OSA+WMC group after age adjustment, which together suggest the predominance of sympathetic activity in the OSA+WMC group. These alterations were consistently observed in patients without hypertension or diabetes, suggesting that our findings are not attributable solely to the presence of hypertension or diabetes. SD12 may reflect the complex nature of HRV changes since it combines two nonlinear characteristics in a single parameter: increased SD12 can reflect increased short-term beat-to-beat variability (SD1) and/or decreased long-term variability (SD2). It is meaningful that SD12 measured during the entire sleep period was significantly higher in the OSA+WMC group, because assessing the ECG data during the entire sleep period can avoid selection bias when choosing ECG samples. Increased SD12 and decreased DFA1 have been reported to be associated with mortality after myocardial infarction,15 and SD12 has been negatively correlated with DFA1 both in healthy subjects and in patients with coronary artery disease.22 Increased sympathetic activity is thought to be reflected in a lower DFA134; increasing the infusion doses of norepinephrine resulted in increased SD12 and decreased DFA1.3536
ApEn during REM sleep was the only parameter that differed significantly between the OSA+WMC and OSA−WMC groups in a subgroup analysis of middle-aged patients with moderate-to-severe OSA. ApEn reflects the randomness of HRV, with a lower ApEn indicating decreased randomness,12 and decreased HRV randomness is associated with simplification of cardiovascular regulation, which is observed in aging and pathologic conditions.11 Patients in the OSA+WMC group demonstrated lower ApEn, which suggests a reduction of parasympathetic tone.25 A lower ApEn has been reported to be associated with postoperative ventricular dysfunction25 and poor outcome in arrhythmia patients.16 We suggest that decreased ApEn, especially during REM sleep, is a useful marker for predicting WMC in OSA patients.
The identified nonlinear indices that were dysregulated in the OSA+WMC group were concordantly related to sympathetic overactivity and parasympathetic dysfunction, which is a well-known autonomic property of OSA. It is not yet known if increased SD12, decreased DFA1, and decreased ApEn are directly associated with the pathophysiology of cerebral WMC. However, we assume that sympathetic overactivity would play a crucial role in the development of WMC. OSA patients experience recurrent intermittent hypoxemia with consequent sympathetic activation and marked surges in blood pressure, with each of these mechanisms possibly impairing endothelial function37 and eventually leading to cerebral WMC.38 It has been reported that sympathetic nerve terminals exist in the tunica media of human cerebral arteries39 and that the intracranial capillary diameter is regulated by pericyte contraction in response to noradrenergic neurotransmitters.40 While continuous-positive-airway-pressure treatment can reportedly improve linear HRV parameters,4142 whether such treatment can normalize nonlinear indices remains unclear. However, we propose that these nonlinear indices can be used as indicators of WMC in OSA patients, and consequently may be useful when selecting OSA patients on whom to perform brain MRI. For example, the presence of WMC in middle-aged OSA patients could be predicted by ApEn during REM sleep with a moderate positive predictive value (71.4%) and specificity (77.8%).
In addition, some of the time-domain indices differed considerably in OSA+WMC patients. It is noteworthy that SDNN and RMSSD were significantly increased during NREM sleep in the OSA+WMC group (the oldest group), because both parameters are known to decrease with increasing age.27 Increased nighttime RMSSD has been reported to be independently associated with the progression of small-vessel disease.43 However, the differences in SDNN and RMSSD in the OSA+WMC group were inconsistent after age adjustment, and so further studies are required to clarify the associations between these parameters and cerebral WMC.
This study was subject to a few limitations. First, the baseline demographic characteristics and the prevalence of chronic conditions such as hypertension and diabetes differed among the included groups. Patients in the OSA+WMC group were significantly older (by a mean of almost 20 years) and had a higher prevalence of hypertension or diabetes than those in the other groups. HRV indices are significantly affected by age, and hypertension and diabetes may impact HRV measures.2627 We therefore performed ANCOVA with age controlled as a covariate, and performed additional subgroup analyses in order to minimize the effect of these potential confounders. Second, most of the subjects in the control group had multiple sleep-related complaints, including insomnia, sleep fragmentation, snoring, nonrefreshing sleep, and so they might not have constituted a reliable control group. Third, apnea-free, artifact-free, 5-min ECG samples were selected from each sleep stage for HRV analysis, which might not represent the HRV for each sleep stage. It is known that HRV changes during REM sleep, and can begin during NREM sleep several minutes earlier than the transition to REM sleep13; also, the degree of sympathetic activation during REM sleep is affected by the previous sleep stage.32 We consider these limitations to be the main reasons for the inconsistencies between the findings for the frequency-domain HRV indices between the current study and previous studies.
In summary, certain nonlinear HRV indices were significantly altered in OSA+WMC patients, with SD12 being increased and DFA1 and ApEn being decreased during sleep. In particular, an increased SD12 during the entire sleep period and a decreased ApEn during REM sleep could be useful as markers for predicting cerebral WMC in OSA patients. Future work should investigate the validity of these nonlinear HRV parameters and their causal relationships with the pathophysiologic mechanisms underlying WMC.
XML Download