Sleep-disordered breathing is associated with brain vascular reactivity in spinal cord injury

Abstract

Objective

To determine the population-level odds of individuals with spinal cord injury (SCI) experiencing fatigue and sleep apnea, to elucidate relationships with level and severity of injury, and to examine associations with abnormal cerebrovascular responsiveness.

Methods

We used population-level data, meta-analyses, and primary physiologic assessments to provide a large-scale integrated assessment of sleep-related complications after SCI. Population-level and meta-analyses included more than 60,000 able-bodied individuals and more than 1,800 individuals with SCI. Physiologic assessments were completed on a homogenous sample of individuals with cervical SCI and matched controls. We examined the prevalence of (1) self-reported chronic fatigue, (2) clinically identified sleep apnea, and 3) cerebrovascular responsiveness to changing CO₂.

Results

Logistic regression revealed a 7-fold elevated odds of chronic fatigue after SCI (odds ratio [OR] 7.9, 95% confidence interval [CI] 3.5–16.2), and that fatigue and trouble sleeping are correlated with the level and severity of injury. We further show that those with SCI experience elevated risk of clinically defined sleep-disordered breathing in more than 600 individuals with SCI (pooled OR 3.1, 95% CI 1.3–7.5). We confirmed that individuals with SCI experience a high rate of clinically defined sleep apnea using primary polysomnography assessments. We then provide evidence using syndromic analysis that sleep-disordered breathing is a factor strongly associated with impaired cerebrovascular responsiveness to CO₂ in patients with SCI.

Conclusions

Individuals with SCI have an increased prevalence of sleep-disordered breathing, which may partially underpin their increased risk of stroke. There is thus a need to integrate sleep-related breathing examinations into routine care for individuals with SCI.

Individuals with spinal cord injury (SCI) have a diverse array of adverse secondary health consequences.^1,2 In particular, single cohort and small scale studies provide preliminary evidence that sleep-disordered breathing (SDB) of obstructive, central, and mixed origin are highly prevalent after SCI.^3–14 Evidence from able-bodied (AB) individuals suggests that SDB underlies widespread fatigue^15,16 and cerebrovascular disease.^17–19 As those with SCI also have elevated rates of all these health consequences, sleep-related breathing may be an extremely powerful modifiable risk factor to reduce all-cause morbidity and mortality and improve quality of life in this population.

Our objective was to use population-level data (>60,000 respondents) to examine whether individuals with SCI have elevated risk of chronic fatigue, a condition often characterized by SDB.^20,21 Next, we leveraged a recent detailed survey of 1,500 individuals with SCI to determine if sleep problems and fatigue are associated with cervical level and complete injuries. We then combined this analysis with prior studies that employed gold standard polysomnography assessments of sleep quality from more than 600 individuals with SCI to further establish the risk of clinically defined sleep-related breathing problems. Next, we aimed to confirm these findings in our own experimental dataset, where we collected more than 90 metrics of sleep quality using laboratory-based polysomnography in a homogenous sample of cervical SCI and matched controls. Finally, we implemented syndromic analyses techniques to examine whether sleep-related breathing problems are associated with impaired cerebrovascular responsiveness to CO₂ after SCI (figure 1).

Figure 1. Experimental overview.

We used population-level data (>60,000 respondents) from the Canadian Community Health Survey combined with a recent survey of 1,500 individuals with spinal cord injury (SCI) and a meta-analysis of 682 individuals with SCI to examine global trends in sleep-disordered breathing. We combined this with our own primary dataset, where we collected laboratory-based polysomnography and cerebrovascular responsiveness to changing CO₂ levels in a homogenous sample of patients with cervical SCI and matched controls to ultimately provide an integrated understanding of sleep-disordered breathing after SCI.

Methods

Population-level statistics

To assess whether individuals with SCI experience greater rates of chronic fatigue at a population level, we utilized data from the Canadian Community Health Survey (CCHS) 2010 Annual Component. The CCHS is a comprehensive national cross-sectional survey conducted by Statistics Canada, containing data on more than 60,000 individuals in a public health care setting. To conduct the survey, Statistics Canada uses a multistage, stratified cluster sampling design, which we have accounted for in our analysis using probability weighting. The primary outcome in this analysis was self-reported chronic fatigue syndrome (CCC_251). Individuals were instructed to answer “yes” only to conditions diagnosed by a health care professional (table 1). The explanatory variable in this analysis was self-reported SCI status (NEUDSIR). Individuals without valid responses (or nonrespondents) for both the primary and explanatory variables were excluded from the analysis. We treated age (DHHGAGE) and sex (DHH_SEX) as categorical risk factors for traumatic and nontraumatic SCI.^22–26 For the statistical analysis, logistic regression models were obtained for the outcome measure chronic fatigue syndrome. In the bivariable model, only SCI status was used as the explanatory variable. In the multivariable model, age, sex, and body mass index (BMI) (HWTGBMI) were input as additional explanatory variables. BMI was categorized into the following groups: normal (18.5–25), overweight (25–30), obese Class I (30–35), obese Class II (35–40), obese Class III (>40). Our sensitivity analysis included confounders such as hypertension (CCC_072), migraines (CCC_081), mood disorders (CCC_280), and anxiety disorders (CCC_290); lifestyle factors such as smoking status (SMKDSTY), physical activity levels (PACDFR, PACDPAI), alcohol consumption (ALC_2), fruit and vegetable intake (FVCDTOT), and self-perceived mental health and life stress (GEN_02B); and socioeconomic factors including household income (INCGHH) and education level (EDUDH04). We examined potential interaction effects with each variable as well as its effect on our adjusted model. We then present a fully adjusted model (AOR2 including all these potential explanatory variables). Using the logistic models, unadjusted and adjusted odds ratios (ORs) with 95% confidence intervals (CIs) are presented. Goodness of fit for the full model was assessed using a receiver operating characteristic curve. Data are presented in accordance with the STROBE and STROND guidelines of reporting.

Table 1.

Characteristics of the population-based survey by chronic fatigue (CF) status (weighted responses)

Clinically annotated SCI dataset analysis

Data from the SCI Community Survey (SCICS) were used to assess the relationship between level and severity of injury and feelings of fatigue and tiredness in individuals with SCI. Data were collected through a National SCI Community Survey using measures developed for the Rick Hansen Spinal Cord Injury Registry Community Follow-up Version 2.0.²⁷ We included 1,549 individuals with SCI who completed the SCICS (table e-1, doi.org/10.5061/dryad.302jq43). We assessed the contribution of level and severity of injury on the self-reported measures fatigue and trouble sleeping. Each outcome was scored from 0 to 5 (table e-2, doi.org/10.5061/dryad.302jq43). We transformed these scores into a binary outcome, whereby individuals reporting 0 or 1 were grouped together and those reporting 2–5 were grouped together. We examined the contribution of injury severity as a binary variable (motor complete [American Spinal Injury Association Impairment Scale [AIS] A/B] vs motor incomplete [AIS C/D]) and injury level as a binary variable (tetraplegia vs paraplegia). AIS scores were evaluated indirectly from participants' answers to questions about their sensorimotor abilities, as described previously.²⁷ For the statistical analysis, logistic regression models were obtained as above.

Meta-analysis

A total of 236 articles were retrieved, using a PubMed search (May 2017) of the keywords “sleep” and “spinal cord injury,” and 11 studies were included in the final analysis (table e-3 and figure e-1, doi.org/10.5061/dryad.302jq43). Our inclusion criteria included (1) gold standard polysomnography tests with valid apnea–hypopnea index (AHI) scores and (2) individuals with SCI. To provide a comparison to normative data, we derived median population values by taking the average of the second quartile scores from published AB data. We note that one study⁴ used a cutoff of 10 for severe AHI; however, for simplicity, we have included this study in our severe (AHI = 15) analysis. We compared the prevalence of clinically defined sleep apnea (based on AHI score) to AB normative data derived from the Sleep Health Heart Study, which included AHI information on 5,615 individuals.²⁸ The mean value from the second quartile was extracted for AHI >5, AHI >15, age, BMI, and neck circumference. Only reference AB data from men were used as our meta-analysis sample was made up of 85% men, and therefore the use of a pooled reference dataset may overinflate the estimated OR. ORs and 95% CIs were extracted using a random effects model (R Package rmeta). In all cases, we first tested for heterogeneity of variance using the Woolf test and examined funnel plots for publication bias. Summary statistics for both AHI >5 and AHI >15 were extracted by pooling the data from all experiments investigating individuals with SCI, accounting for heterogeneity within the data (n = 682).

Standard protocol approvals, registrations, and patient consents

Ethical approval for the CCHS was obtained through the publicly available data clause from the University of Calgary. For the SCICS, written informed consent was obtained from each participant in the survey, and from local research ethics boards to recruit from SCI centers across Canada. For our physiologic data collection, 25 individuals participated in this study, which was approved by the ethics committee of the University of Split School of Medicine, and conformed to the standards set by the Declaration of Helsinki (see table 2 for participant characteristics). Participants received written and verbal descriptions of experimental procedures and provided written informed consent.

Table 2.

Characteristics of the polysomnography participants

Polysomnography

Full-night attended polysomnography (Alice 5LE or Alice 6LDxN, Philips Respironics, Eindhoven, the Netherlands) was performed. Recordings included EEG, electrooculography, mental and tibial EMG, ECG, nasal airflow, pulse oximetry, thoracic and abdominal movements, and snoring intensity.^29–31 All data were manually scored and evaluated in accordance with the published American Academy of Sleep Medicine guidelines by the same certified sleep physician who was blinded to the participant's involvement in the study. Participants were nonsmokers free of medication and did not have a known history of cardiovascular disease. In total, we analyzed 91 metrics of sleep quality (table e-4, doi.org/10.5061/dryad.302jq43).

The dimensionality of our generated matrix M_i,j, where i = polysomnography outcome measures and j = independent participants, was reduced using principal component analysis, implemented using the principal component analysis function in the R package FactoMineR. Principal component analysis was chosen to compute our high dimensional data into a low dimensional representation by linearly projecting outcome vectors along basis vectors to maximize the captured variation within each vector. The principal component score (i.e., eigenvalue) for each column (individual) can then be obtained by projecting its outcome vector (i.e., eigenvector) along the principal directions. Individual outcome measure loadings (i.e., the correlation to the principal component) were then examined for each principal component. For syndromic analyses, a similar procedure was completed after combining all polysomnography and cerebrovascular responsiveness outcome measures into one matrix prior to principal component analysis.³²

Cerebrovascular responsiveness

Participants arrived at the laboratory having been instructed to abstain from exercise and alcohol intake for 24 hours and items containing caffeine or a large meal for 4 hours. Hypercapnia was induced by switching inspired gas from room air to 4% CO₂ (in 21% O₂ with the balance N₂) for 4 minutes. The PETco₂ was recorded throughout. Following hypercapnia, participants were instructed to increase their rate and depth of breathing to generate a level of hypocapnia to match, in an equal and opposite direction, the rise in PETco₂ incurred during hypercapnia, and maintain this for 3 minutes. The changes in PETco₂ were not different between groups for hypercapnia (SCI 15.11 ± 4.06 mm Hg, AB 15.08 ± 1.86 mm Hg; p = 0.5872; independent samples t test) or hypocapnia (SCI −10.59 ± 4.82 mm Hg, AB −11.58 ± 5.84 mm Hg; p = 0.6499; independent samples t test) The velocity of blood in the right middle cerebral artery (MCA) and left posterior cerebral artery (PCA) were simultaneously measured using a 2 MHz pulsed Doppler ultrasound system (Spencer, Redmond, WA).^33,34 Beat-by-beat arterial blood pressure and heart rate were monitored using finger photoplethysmography (Finometer PRO, Finapres Medical Systems BV, Enschede, the Netherlands) and ECG, respectively. End-tidal CO₂ was sampled from a leak-free mouth piece and measured by a gas analyzer (ML206; ADInstruments, Colorado Springs, CO). All data were acquired continuously at 1,000 Hz using an analog-to-digital converter (Powerlab/16SP ML795; ADInstruments) interfaced with a laptop computer that collected data using LabChart (version 8, ADInstruments).

To characterize the cerebrovascular responsiveness, we wrote a custom R script to iterate through a range of measures (table e-5, doi.org/10.5061/dryad.302jq43). Each outcome was calculated as a function of MCA velocity, PCA velocity, or respiratory rate. Additional outcomes calculated were pulsatility ([MCA/PCA max velocity − MCA/PCA min velocity]/MCA/PCA mean velocity) and pulsatility index, calculated as pulsatility/mean arterial pressure. For each metric, we calculated outcome measures from the absolute and percentage change, and for both hypercapnia and hypocapnia. For each measure, we calculated the peak and time to peak. We computed all outcome measures with the inclusion of specific time windows (time windows included 30, 60, 90, 120, 150, and 180 seconds from the onset of the stimulus).

Statistics

All statistical procedures were conducted using R (R Core Team, 2017). Data were assessed for normality prior to any statistical procedures, after which the relevant parametric or nonparametric tests were completed. Alpha was set a priori at 0.05 for all null hypothesis testing.

Data availability

Supplementary data are provided online (doi.org/10.5061/dryad.302jq43).

Results

Population-level data reveal that individuals with SCI have increased odds for chronic fatigue

To understand whether symptoms of fatigue in individuals with self-reported SCI were pervasive at the population level, we examined the CCHS. The 2010 CCHS consists of 62,909 unique individuals. After excluding individuals with invalid responses for the primary explanatory and outcome variables, our final study sample included 57,436 individuals (table 1). Of these 57,436 individuals, 328 unique individuals with SCI provided a valid response, and were included in the analysis (table 1). Median age categories, as well as sex stratification across our samples, can also be found in table 1.

In individuals without SCI, the prevalence of chronic fatigue was 1.3% (table 1). In contrast, the prevalence of chronic fatigue in individuals with SCI was 10.3% (table 1). The risk of chronic fatigue was 9.0 times greater in individuals with SCI compared to individuals without SCI (95% CI 5.9–13.2; table 3). In the age-, sex-, and BMI-adjusted model, the OR for chronic fatigue was still elevated, but reduced to 8.0 (95% CI 5.2–11.9; table 3). To ensure the robustness of our results, we included a sensitivity analysis, and added other potential confounding variables. Specifically, we included additional examination of disease comorbidity confounders such as hypertension, migraines, mood disorders, and anxiety disorders; lifestyle factors such as smoking status, physical activity levels, alcohol consumption, fruit and vegetable intake, self-perceived mental health, and life stress; and socioeconomic factors including household income and education level. After inclusion of any one of these variables, our results remained significant, with adjusted ORs ranging from 4.7 to 11.0. In addition, even after inclusion of all these variables, our results remained significant (adjusted OR 7.9, 95% CI 3.5–16.2; receiver operating characteristic 0.84).

Table 3.

Odds ratios (ORs) (95% confidence intervals [CIs]) for chronic fatigue (CF) (probability-weighted)

Level and severity of injury are correlated to the severity of sleep-related disorders in a clinically annotated dataset of individuals with SCI

Next, we aimed to assess the potential effects of severity and level of injury on sleep and fatigue-related disorders using a targeted SCI survey that includes level and severity of SCI. We included 1,549 individuals with SCI who completed the SCICS. Of these individuals, 569 were motor and sensory complete (AIS A), 116 were motor complete, sensory incomplete (AIS B), 302 were AIS C, and 319 were AIS D, while the remainder had poorly defined AIS scores. The mean age of the sample was 50 years with an SD of 13.9 years. A total of 412 had nontraumatic injuries and 1,137 had traumatic SCI. A total of 1,041 individuals were men and 508 were women.

Logistic regression revealed that individuals with tetraplegia (see table e-1, doi.org/10.5061/dryad.302jq43, for participant and descriptive characteristics) reported trouble sleeping more than individuals with paraplegia (OR 1.5, 95% CI 1.2–2.0; table e-6 and figure e-2, doi.org/10.5061/dryad.302jq43). This effect persisted after controlling for age and sex (OR 1.5, 95% CI 1.2–2.0; table e-6, doi.org/10.5061/dryad.302jq43), and increased in a stratified sample of individuals with only motor-complete injuries (OR 1.8, 95% CI 1.2–2.6; table e-6, doi.org/10.5061/dryad.302jq43). In a stratified sample of individuals with only tetraplegia, we found that those with complete injuries were more likely to experience trouble sleeping, even after controlling for age and sex (OR 1.8, 95% CI 1.2–2.9; table e-6, doi.org/10.5061/dryad.302jq43). We found that individuals with SCI experienced fatigue at a significant rate (72% for individuals with paraplegia, 75% of those with tetraplegia) (table e-1, doi.org/10.5061/dryad.302jq43). We did not find that fatigue was dependent on the severity and level of injury (table e-7, doi.org/10.5061/dryad.302jq43). However, there was significant agreement between trouble sleeping and feelings of fatigue (κ = 0.35; 95% CI 0.31–0.42; Z = 10.9; p < 0.001).

Individuals with SCI exhibit an increased prevalence of sleep apnea

We further supported our population-level data with previous studies examining sleep-related breathing disorders in SCI. We found 11 studies meeting our criteria (see table e-3, doi.org/10.5061/dryad.302jq43, for a list of included studies). We used the literature-curated normative data to extract ORs for elevated AHI at a mild cutoff (AHI >5; n = 509 individuals with SCI; n = 2,648 AB individuals) and a more severe cutoff (AHI >15; n = 300 individuals with SCI; n = 2,648 AB individuals). The prevalence of mild sleep apnea in a meta-analysis of individuals with SCI was 72.5% (369/509) vs 33.0% (873/2,648) in AB individuals. This elevated rate of sleep apnea persisted when using a more conservative cutoff (AHI > 15), where the prevalence of sleep apnea in individuals with SCI was 63% (189/300) vs 25.0% (662/2,648). In this sample, we found that individuals with SCI were at elevated risk of experiencing AHI-defined sleep apnea (AHI 5: OR 3.6, 95% CI 1.6–7.7; AHI 15: OR 3.1, 95% CI 1.3–7.5; figure 2A), despite similar neck circumference, age, and BMI (figure e-3, doi.org/10.5061/dryad.302jq43).

Figure 2. Epidemiology of sleep-disordered breathing in individuals with spinal cord injury (SCI).

(A) Random effects meta-analysis reveals significantly elevated odds of both mild (left) and severe (right) sleep apnea in individuals with SCI, compared to median able-bodied (AB) values derived from the literature. (B) Our primary data examining the apnea–hypopnea index (AHI) through gold standard polysomnography in AB individuals and individuals with SCI reveals a high prevalence of both mild and severe sleep apnea. (C) Dot plot demonstrates the number of obstructive apneic events (OA), mixed apneic events (MA), and central apneic events (CA) in AB individuals and those with SCI. Each row represents 1 individual, with the size of the dot representing the number of events. Colors indicate the type of apneic episode (OA, MA, CA). (D) Bar plots (mean with standard error) demonstrate a significant difference (p = 0.010) in the number of obstructive apneic events in individuals with SCI (Wilcoxon rank sum test with a 5% false discovery rate). A similar trend was found for MA and CA. CI = confidence interval.

Next, we conducted gold standard polysomnography examinations on 25 individuals (n = 13 tetraplegia, n = 12 AB). Bivariate linear regression revealed a significant relationship between injury status and AHI (β 17.5, 95% CI 1.3–33.7; p = 0.035). This relationship persisted in a multivariable model controlling for known risk factors including age, sex, BMI, and neck circumference (β 18.5, 95% CI 1.3–35.8; p = 0.036). We found that individuals with SCI had a high prevalence of clinically defined mild sleep apnea (AHI >5; 46%; figure 2B), moderate sleep apnea (AHI >10; 46%; figure 2B), and moderate to severe sleep apnea (AHI >15; 31%; figure 2B), compared to our AB sample. Further, these individuals experienced significantly high rates of having at least one episode of obstructive (76.9%), mixed (61.5%), or central (53.8%) apnea during laboratory testing (figure 2C). We observed a significant difference for the number of obstructive apnea events in individuals with SCI (Figure 2D), and a similar trend for mixed and central apneic events (figure 2D).

Dimensionality reduction reveals key variables associated with SDB in individuals with SCI

While the AHI provides clinically relevant information relating to the diagnosis of sleep apnea, it does not capture or reveal which specific aspects of sleep are primarily disrupted. Given that gold standard polysomnography provides a host of variables to examine, we sought to gain a deeper understanding of what specific aspects of sleep are disrupted in individuals with SCI. We took 91 variables (table e-4, doi.org/10.5061/dryad.302jq43) derived from our gold standard sleep studies on all individuals who participated in our sleep studies (table 2). We performed principal component analysis to compute our high dimensional data into a low dimensional representation. After reducing the dimensionality of polysomnography-derived variables, we found that individuals with SCI significantly deviated along the first principal component (figure 3, A and C). The 2 individuals who most significantly deviated along this linear projection were clinically recognized as presenting with significant SDB, and immediate clinical intervention occurred (figure 3A). Key variables that most highly correlated to the first principal component were the number and length of apneic episodes and the number of obstructive apneic episodes (figure 3, B and C). In contrast, the second principal component explained leg movements, while principal component 3 tended to examine broad sleep metrics such as sleep efficiency and total sleeping time.

Figure 3. Principal component (PC) analysis of polysomnography in individuals with spinal cord injury (SCI).

Individuals with SCI deviate along the first PC axis. Ninety-one polysomnographic variables were input into a principal component analysis to examine differential trends following SCI. (A) Those with the most severe sleep apnea were found to have the lowest PC1 score, and more individuals with SCI were found to deviate along this axis compared to able-bodied (AB) individuals. (B) Variable loading scores revealed that PC1 primarily explained variables related to apnea–hypopnea (AH) quantification indices, while PC2 explained leg movement indices and PC3 explained broad sleep indices. (C) Individuals with SCI were significantly different from controls for their PC1 scores and variables highly correlated to PC1, whereas this trend was not found for PC2 (D) and PC3 (E). ARL = arousing leg movements; I = variable index (n/total sleep time); OA = obstructive apnea; PLMS-Ar = passive leg movements leading to arousal; SE = sleep efficiency; TST = total sleep time. * p < 0.05 according to the Wilcoxon rank sum test following false discovery rate correction.

We found variables highly associated with principal component 1, but not principal components 2 and 3 (figure 3, D and E), when tested independently, were significantly different in individuals with SCI. Individuals with SCI experienced significantly longer total apnea–hypopnea duration during sleep, number of obstructive apneic episodes, as well as the total number of apneic episodes compared to AB individuals (figure 3C). Alarmingly, individuals with SCI also exhibited significantly reduced oxygen saturation during sleep, which was highly correlated to principal component 1 (figure 4D).

Figure 4. Syndromic analysis.

Cerebrovascular responsiveness and polysomnography variables were input into a unified matrix (n = 483 parameters total). (B) This matrix was then subject to principal component analysis (PCA). (C) The first principal component explained 16% of the variance in the data and was highly correlated to measures of apnea–hypopnea, indicating a clear association with sleep-disordered breathing. Key measures of both anterior and posterior responsiveness to CO₂ were highly negatively correlated with the first principal component, indicating a potentially causal relationship between cerebrovascular responsiveness and sleep-disordered breathing. Arrows and annotated numbers represent the loading value of that variable (e.g., max apnea duration, number of apneic events) to the first principal component of the unified matrix. (D) Oxygen desaturation data (highly correlated to the first principal component in our sleep PCA). Our data reveal that individuals with spinal cord injury (SCI) have significantly lower minimum O₂ saturation during sleep (A), as well as lower average O₂ saturation across all sleep (B), REM sleep (C), and non-REM sleep (D). (E) Example pulsatility metrics that are highly correlated to PC1 in a bivariate correlation analysis, and significantly different between groups. Our data reveal that individuals with SCI have significantly lower peak posterior pulsatility during hypocapnia (A), as well as lower peak posterior pulsatility ratio during hypocapnia (B), and peak anterior pulsatility during hypocapnia (C). ** p < 0.01, *** p < 0.001. Statistical tests completed using a one-tailed Wilcoxon rank sum test with a false discovery rate of 5%. AB = able-bodied; AHI = apnea–hypopnea index.

Syndromic analysis reveals association between CO₂ reactivity and SDB

Repetitive hypoxic insults may impair cerebrovascular health and reduce the capacity of the cerebrovasculature to respond to changes in CO₂ tension. We therefore examined cerebrovascular responsiveness to changing CO₂ levels in these same individuals (figures e-4 and e-5, doi.org/10.5061/dryad.302jq43) and determined whether cerebrovascular responses to CO₂ were correlated with measures of SDB using an analysis strategy known as syndromics (figure 4, A and B).^35,36 This revealed that the first principal component (which explained 13% of the variance within the data) was highly correlated to AHI (loading value of 0.65), indicating that this principal component was representative of SDB (figure 4C). We also found that metrics of pulsatility responsiveness to CO₂ from both the anterior and posterior cerebral circulation were correlated to this principal component (figure 4C), suggesting that small-vessel cerebrovascular responsiveness to CO₂ is highly related to SDB. These data are supported by significant between-group differences for these metrics (figure 4E; see table e-5, doi.org/10.5061/dryad.302jq43, for a complete list of pulsatility metrics). In line with our findings of increased number of desaturations after SCI (figure 4D), the number of desaturations was also strongly correlated to the first principal component (r = 0.67).

Discussion

In this study, we used a combination of population-level data, meta-analyses, and primary physiologic assessments to provide a large-scale integrated analysis of sleep-related complications after SCI. Using data from 60,000 individuals, we found (1) a 7-fold elevated risk of chronic fatigue after SCI and (2) that fatigue and trouble sleeping are correlated with the level and severity of injury. Meta-analysis data from 682 individuals confirmed that the high rate of fatigue was mirrored by elevated odds of clinically defined SDB. We confirmed in our own primary data that individuals with cervical SCI experience a high rate of clinically defined sleep apnea. Finally, we provide clinical evidence using syndromic analysis that impaired sleep-related breathing problems in individuals with SCI are strongly associated with cerebrovascular dysfunction.

SDB is an established risk factor associated with stroke and other cerebrovascular disorders in non-SCI populations.^18,19 Our analysis revealed that people with SCI have SDB at similar rates to those with heart failure, which is a condition widely appreciated to have rampant SDB.³⁷ We also found that individuals with tetraplegia were more likely to experience trouble sleeping. This observation may indicate that individuals with tetraplegia should be more closely screened for sleep-related complications. Providing insight into the association between SDB and cerebrovascular reactivity will allow us to understand if sleep-related breathing problems are affecting cerebrovascular reactivity after SCI,^{3–14,17,38–42} where there is a risk-factor-adjusted 3- to 4-fold increased risk of stroke.^24,43

Our recent work has demonstrated impaired cerebrovascular function after SCI as well as elevated odds of stroke.^{24,33,34,44,45} Our current syndromics approach revealed that both apnea episodes and sleep time oxygen saturations are highly related to cerebrovascular responsiveness to CO₂, which may indicate that hypoxic insults during sleep are impairing cerebrovascular reactivity after SCI.^46,47 We also know that declining cerebrovascular function is associated with the development of sleep-related disordered breathing.^48–50 Specifically, impaired cerebrovascular responsiveness to CO₂ can act in part to destabilize breathing by increasing the gain of the central ventilatory reflex, and is related to inappropriate breathing while asleep.⁴⁹ The current work does not attempt to delineate this complex bidirectional relationship between sleep-related breathing and cerebrovascular dysfunction. However, given that the exact relationship between cerebrovascular dysfunction and SDB is population-specific, our results provide important evidence that both SDB and cerebrovascular dysfunction are prevalent and associated in individuals with SCI. These data therefore may inform decisions regarding which data to collect prospectively in registries (e.g., the specific type of sleep-related breathing problems, cerebrovascular responses to CO₂ levels), whether or not to screen individuals with SCI for sleep-related breathing disorders, and whether sleep-related breathing problems are a factor in elevated rates of cerebrovascular disease.

There are possible limitations to the current study. First, the data are primarily derived from cross-sectional study designs. Therefore, it is not possible to assess if SCI preceded the sleep-related breathing problems. Furthermore, the CCHS relies on self-report to define the presence of SCI, which may lead to an overestimation. Another limitation is that the variable “chronic fatigue syndrome” within the CCHS is nonspecific, and may be a result of unidentified factors. Within our clinically annotated dataset, we were limited to responses to the questions “Do you experience fatigue?” and “Do you experience trouble sleeping?” These questions may not be as specific as the clinical determination of sleep disorders, using tools such as polysomnography testing.

These limitations notwithstanding, the increased odds of sleep-related complications within people with SCI highlights the need for future investigations. Indeed, the association between sleep-related breathing and cerebrovascular health is not consistent, and appears to be population-dependent. For example, those who have already sustained a stroke have greater rates of sleep-related breathing problems,⁵¹ and in men (not women) sleep-related breathing problems appear to be associated with increased risk of stroke.⁵² Small and medium scale studies have not found a relationship between sleep-related breathing problems and white matter disease in generally healthy adults.^53,54 Other work showed that sleep-related breathing problems after stroke had a very weak but significant association with white matter disease.⁵⁵ One very small study observed that cerebrovascular reactivity was impaired compared to controls; however, no correlation between these factors was reported.⁵⁶ On the other hand, in a follow-up study of a similar size, this group difference did not occur.⁵⁷ Together, these findings would suggest the relationship is not clear, and that examination of these relationships within specific clinical populations is warranted. For individuals with SCI, detailed respiratory and cardiovascular physiologic assessments in patients will be needed to explore the high plausibility for a causal relationship between SCI and SDB, and cerebrovascular capacity to respond to changing CO₂ levels. Overall, there may be a need to integrate sleep-related breathing examinations into routine care and perhaps even to consider prophylactic treatment for individuals with SCI and thereby potentially reduce the cerebrovascular burden on this population.

Glossary

AB

able-bodied

AHI

apnea–hypopnea index

AIS

American Spinal Injury Association Impairment Scale

BMI

body mass index

CCHS

Canadian Community Health Survey

CI

confidence interval

MCA

middle cerebral artery

OR

odds ratio

PCA

posterior cerebral artery

SCI

spinal cord injury

SCICS

SCI Community Survey

SDB

sleep-disordered breathing

Appendix. Authors

Study funding

The Phillips Lab is supported by the Wings for Life Foundation (Project Grant), Compute Canada (Resources for Research Groups), Natural Sciences and Engineering Research Council (Canada; Discovery Grant), the Canadian Institutes for Health Research (Project Grant), Alberta Innovates Health Solutions, Campus Alberta Neuroscience, the Libin Cardiovascular Institute of Alberta, the Hotchkiss Brain Institute, and the Rick Hansen Institute. This work was also supported by an ICORD Seed Grant funded through the Blusson Integrated Cures Partnership. Dr. Barak is supported by a grant (IO-175037) that is funded by theMinistry of Education, Science and Technological Development of the Republic of Serbia. The study was also supported by the Rick Hansen Institute (grant 2010-03) and the Ontario Neurotrauma Foundation (grant 2010-RHI-SURVEY-812). J.W. Squair is supported by a Banting Postdoctoral Fellowship from the Canadian Institutes of Health Research, a Killam Postdoctoral Fellowship, and an Alberta Innovates Postdoctoral Fellowship.

Disclosure

The authors report no disclosures relevant to the manuscript. Go to Neurology.org/N for full disclosures.