Introduction

People living with spinal cord injury (SCI) typically experience significant impairment from the complications arising from their injury. Quality of life following SCI is low and predicted by the presence of complications and community participation [1]. Sleep-disordered breathing (SDB) is a highly prevalent complication of tetraplegic SCI, leading to fragmented sleep and repetitive reductions in oxygen saturation. SDB is an umbrella term, incorporating obstructive sleep apnoea (OSA), central sleep apnoea (CSA) and sleep-related hypoventilation disorders. It has been associated with substantial neurocognitive impairment, daytime sleepiness and reduced quality of life in people with tetraplegia [2,3,4]. In the general population, strong associations have consistently been found between SDB and chronic diseases such as hypertension, diabetes, myocardial infarction and congestive heart failure [5].

Treatment with continuous positive airway pressure has been shown to alleviate daytime sleepiness associated with SDB in people with tetraplegia [6]. However, access to diagnosis and treatment in this population is frequently poor [7, 8]. Despite the high prevalence, surveys suggest that less than a quarter of people with SCI have been tested for SDB [9, 10].

A precise measure of SDB prevalence in tetraplegia is critical to service and research planning, however; reported prevalence estimates have ranged from 15 to 97% [11,12,13]. Heterogeneity in study design is likely responsible for the wide-ranging results. Types of sleep studies have varied from simple overnight oximetry to home based and in-laboratory polysomnography. The AASM have updated the rules for scoring respiratory events several times, which has impacted case detection rates significantly [14, 15]. Additionally, studies have employed different thresholds for case detection. Characteristics of study populations (by e.g. SCI severity, age and time since injury) have also varied substantially, and the frequently small sample sizes have undoubtedly affected the precision of their prevalence estimates [11, 12].

Reported risk factors for SDB in SCI have included: male sex, larger neck circumference, greater obesity (body mass index (BMI) and abdominal girth), older age, higher lesion level (tetra- vs. paraplegia), complete SCI, sleeping in the supine position, and cardiac and antispasmodic medications [2, 13, 16,17,18,19,20,21,22]. However, these risk factors are inconsistently reported and when reported, findings occasionally are conflicting between studies. Additionally, pathophysiological abnormalities associated with SDB in SCI include increased ventilatory loop gain and control, altered parasympathetic tone and CSA, especially during sleep onset and sleep state transitions [23, 24].

Several review papers have summarised the literature on SDB in SCI, however none have included a meta-analysis of SDB prevalence [11, 12, 25]. A recent empirical study investigating the relationship between SDB and cerebrovascular reactivity in SCI included a meta-analysis of published prevalence data and estimated the prevalence of SDB in SCI to be 73% [26]. However, the search strategy was not systematic and furthermore, several of the included studies shared the same dataset (i.e. non-independent samples).

The aim of this systematic review and meta-analysis is to estimate the prevalence of objectively measured SDB in tetraplegia and, where possible, to identify which factors are associated with higher prevalence.

Methods

Search and data extraction

The protocol for this study was published in PROSPERO (CRD42019132322), an international database of prospectively registered systematic reviews. We identified studies from systematic searches of bibliographic databases including MEDLINE, EMBASE and the Cochrane Central Register of Controlled Trials (CENTRAL) using SCI and SDB search terms (see Supplementary File for full search strategy). Grey literature sources (OpenGrey, Grey Matters, Trove) were similarly searched and screened for relevant articles.

Articles were restricted to publication dates between 1 January 2000 and 4 September 2020 (the search date). The decision to exclude articles published prior to 2000 was made because the American Academy of Sleep Medicine (AASM) published the first comprehensive set of rules for scoring of respiratory events during sleep in 1999, and the heterogeneity of SDB definitions prior to this date is considered to be substantial. Observational (cohort and cross-sectional) and intervention studies with objective assessment of SDB, in adults with traumatic tetraplegia were included. Studies with less than 100% non-traumatic cases were included. Studies with people with paraplegia and tetraplegia were included as long as the SDB prevalence results were reported separately. Studies needed to measure the presence of SDB with an overnight sleep study (Levels 1–4; see below), by reporting prevalence of SDB with an Apnoea Hypopnoea Index (AHI), Respiratory Disturbance Index (RDI) or Oxygen Desaturation Index (ODI). Studies were excluded if they diagnosed SDB with questionnaire alone or if they only included people with non-traumatic causes of SCI. The authors of any studies reporting SDB prevalence in SCI without sub-group reporting for participants with tetraplegia were contacted to ask whether prevalence in the tetraplegic group could be provided. If these data were not available, the study was excluded. If there was potential for duplicative datasets across reports, the authors were contacted to clarify and to suggest which article(s) to include. The intent was to only include English language publications because no member of the study group spoke a language other than English and there was no funding available for transcription. Regardless, the search strategy only yielded English articles.

Two researchers (MG and LM) independently performed a review of titles and abstracts against the defined criteria to exclude ineligible studies. Following this a review of full text articles was performed by the same two authors (MG and LM) to finalise inclusion. In both stages, any disagreements not resolved by discussion were referred to a third researcher (DJB) for resolution. The same authors then independently extracted the following data from included studies:

  • Sample characteristics: age, sex, SCI severity, SCI level, acute or chronic injury (time since injury). Injury severity was categorised using the American Spinal Injury Association (ASIA) impairment scale (AIS); AIS A, B, C and D.

  • Method: study design, sampling method, sample size, diagnostic method including type of sleep study, diagnostic definitions of SDB (including which version of the AASM rule set was used) and thresholds for diagnosis. Sleep study type was categorised as Level 1 (full in-laboratory attended polysomnography (PSG)), Level 2 (unattended PSG in homes or hospital beds), Level 3 (respiratory only signals) or Level 4 (oximetry and/or nasal flow only).

  • Results: mean (SD) or median (IQR) AHI, RDI, ODI. SDB prevalence rates at various thresholds of AHI, RDI, ODI (as reported). The results were extracted for overall sample and sub-groups (acute vs. chronic injury, age groups, sex, SCI level and severity, BMI) where possible. In a deviation from the PROSPERO protocol, data about respiratory event types (e.g. central vs. obstructive vs. mixed) were extracted where available.

Mild, moderate and severe SDB were defined as an AHI ≥5, ≥15 and ≥30 respectively. Where SDB prevalence at these thresholds was not provided in the paper, they were calculated from data provided in the paper or from individual participant data, if available. Any differences in data extraction were resolved by a third researcher (DJB).

Two researchers (MG and LM) independently assessed risk of bias, using the Joanna Briggs Institute critical appraisal checklist for studies reporting prevalence data, to assess the quality of each included study and determine suitability for inclusion in meta-analysis [27]. Any disagreements not resolved by discussion were referred to a third researcher (DJB) for resolution. Studies that scored 5 or less out of 10 were excluded from meta-analysis.

Statistical analyses

Data analysis was performed using MetaXL software (Version 5.3) [28]. All data were transformed using the double arcsine transformation method [28]. A random effects model was used to calculate weighted summary proportions with 95% confidence intervals from the extracted prevalence results at various diagnostic thresholds (AHI ≥5, ≥15, ≥30). A random effects model was selected over a fixed effects model because of the known issues with heterogeneity in study design. Heterogeneity was assessed statistically with the chi-squared test for heterogeneity and the I-squared statistic. A result of p < 0.1 (chi-squared test) and/or >50% (I-squared statistic) was considered to be evidence of high heterogeneity. A sensitivity analysis, with the results of the excluded studies, was also performed to determine the effect of excluding these studies on prevalence of at least moderate SDB (AHI ≥ 15).

Analysis of sub-groups or subsets

As per the method described above, weighted summary proportions of participants with AHI ≥ 15 were calculated for each sub-group (age, sex, AIS, SCI level, BMI, time since injury) using a random effects model in MetaXL. Chi-squared analyses were subsequently performed in Review Manager (RevMan Version 5.4; The Cochrane Collaboration) to assess differences between groups.

Results

The search yielded 836 articles (Fig. 1). None were in a language other than English. After duplicates were removed, 616 titles/abstracts were screened for inclusion. Sixty-two full text articles were reviewed. Of these 50 were excluded; nine because data had been reported in more than one study (duplicate data); 16 were abstracts not later published in full; nine were abstracts that had subsequently been published in full; and seven were published prior to 2000. Two articles reported SDB prevalence in SCI without sub-group reporting for tetraplegia [16, 29]. The relevant information, sub-group reports for tetraplegia, was supplied by the authors of one of these studies, which was subsequently included [29]; the other study was excluded. Three other studies in SCI populations provided sub-group reporting for tetraplegia and were therefore included [17, 30, 31]. Twelve articles were included in the systematic review.

Fig. 1
figure 1

PRISMA flow diagram.

Full size image

Risk of bias assessment

The results of the Joanna Briggs Institute critical appraisal checklist are presented in Table 1. The assessment identified three studies that were not suitable for inclusion in the meta-analysis [17, 22, 26]. Two of these studies utilised inferior testing methods (Level 3 sleep studies); one had poorly defined criteria for case detection and both chose non-conventional definitions for classifying SDB [17, 22]. The third study did not provide any information about how the participants were sampled [26]. The remaining nine studies used Level 1 or 2 polysomnography for testing and AASM scoring rules for counting events. Incorporation of symptoms (e.g. excessive daytime sleepiness, recurrent awakenings, impaired concentration) into the definition of SDB in the Leduc et al. study had minimal impact on the prevalence rate (53% vs. 56%) [18].

Table 1 Characteristics of included studies.
Full size table

Narrative synthesis

Table 1 summarises the design and subject characteristics of the 12 included studies. The combined sample size of the 12 articles was 752, which included 705 people with tetraplegia. Five studies (combined tetraplegic sample = 395) were conducted in the acute inpatient setting (<12 months since SCI), and six (combined sample = 297) were done in people with chronic SCI, defined in some as >6 and others as >12 months post injury. One study did not specify whether the participants had acute or chronic SCI [26]. Two studies conducted in the acute setting assessed the presence of SDB at multiple timepoints. One assessed participants immediately post injury, and at 2, 4, 13, 26 and 52 weeks post SCI [32], and the other assessed at 7 and 20 weeks [29]. For the meta-analysis, we included the data from the 13 and 7 week assessments respectively, because the sample size was largest at these timepoints. The average time since injury for studies reporting on chronic tetraplegia ranged from 9 to 15 years. The proportion of males in each study ranged from 62 to 100%, and the mean age of participants ranged from 34 to 50 years.

Regarding the type of sleep study used to identify SDB, four were Level 1 (full in-laboratory attended PSG), six were Level 2 (unattended PSG in homes or hospital beds) and two were Level 3 (respiratory only signals). Various rules were used to define hypopnoeas, with the AASM 1999 (ʻChicago’) criteria the most common (six studies).

Table 2 summarises the prevalence of SDB of the 12 studies at various AHI thresholds. Average AHI ranged from 13 to 36 events/h. Evidence of at least mild SDB, defined as AHI ≥ 5, was available in 11 of the 12 studies, and ranged from 46 to 97%. At least moderate SDB (AHI ≥ 15), also available in 11 studies, ranged from 22 to 81%, and severe SDB (AHI ≥ 30), available from six studies, ranged from 11 to 50%.

Table 2 Sleep-disordered breathing prevalence results of 12 included studies.
Full size table

SDB prevalence rates for pre-defined sub-groups (e.g. age group, sex, AIS) were rarely presented in the included studies, however many studies reported associations between these risk factors and SDB. Berlowitz et al. reported SDB prevalence (AHI ≥ 10) to be higher in complete (AIS A) than incomplete (AIS B, C, D) tetraplegia (91% vs. 56%, p < 0.05) [2]. They also found that the severity of SDB was associated with increasing age and waist circumference. Graco et al. also demonstrated greater risk of SDB as injury completeness (AIS A–D) increased [13]. Conversely Burns et al. reported SDB was present in only 21% of SCI participants with motor complete SCI (AIS A and B) and in 83% of participants with motor incomplete injuries (AIS C and D) [17]. Stockhammer et al. reported the prevalence of SDB to be higher in men (55%) than in women (20%, p < 0.05) [22]. This study also demonstrated associations between increasing severity of SBD and increasing age, BMI, neck circumference and time since injury. No relationship between SDB and cervical lesion level (C3–C5 vs. C6–C8) or AIS was demonstrated [22]. In the Sankari et al. study, males were more likely to have SDB than females, and older age increased the odds of SDB [31]. Leduc et al. found BMI ≥ 30 significantly predicted SDB in tetraplegia, but no association between cervical lesion level, AIS, sex, age and time since injury was observed [18].

Three studies (N = 462) provided data on the categories of respiratory event type. The proportion of central events ranged from 2 to 4%, with hypopneas followed by obstructive events comprising the vast majority of the AHI score [6, 13, 32]. When reported, the prevalence of predominant CSA, defined as a greater proportion of central than obstructive events, ranged from 0 to 4% [6, 13, 22, 29]. Burns et al. classified 10% of the total SCI sample (n = 20) with predominant CSA; however, hypopnoeas, the predominant event type in other studies, were not counted [17].

Meta-analysis

Meta-analyses of prevalence of at least mild, moderate and severe SDB in tetraplegia are presented in Fig. 2. The prevalence of at least mild SDB (AHI ≥ 5) was 83% (95% CI = 73–91), that of at least moderate SDB (AHI ≥ 15) was 59% (46–71) and that of severe (AHI ≥ 30) was 36% (26–46). The heterogeneity was substantial in all three meta-analyses, with I2 values ranging from 76 to 85% (Fig. 2). Sensitivity analysis using the inclusion of the three studies that were excluded from meta-analysis found the prevalence of at least moderate SDB (AHI ≥ 15) was 57% (46–67); thus there was no significant difference from the primary analysis (Supplementary Fig. 1).

Fig. 2: Prevalence of at least mild SDB (AHI ≥ 5), at least moderate SDB (AHI ≥ 15) and severe SDB (AHI ≥ 30).
figure 2

SDB sleep-disordered breathing, AHI apnoea hypopnoea index, CI confidence interval, Q Cochrane’s Q statistic, I2 I squared statistic.

Full size image

Sub-group analyses

There were insufficient data provided in the included studies to extract prevalence data for any of the a priori defined sub-groups, except for acute vs. chronic SCI. There was no statistically significant difference in SDB prevalence between studies focusing on acute (63%, 95% CI = 49–76%) and chronic (58%, 95% CI 34–81%) SCI populations (Table 3; Supplementary Fig. 2). Sub-group analyses to determine the effect of sleep study type (Level 1 vs. Level 2) and diagnostic criteria (AASM 1999 vs. AASM 2012) on the prevalence of SDB were also performed. Both found there were no statistically significant differences between these sub-groups (Supplementary Figs. 3, 4).

Table 3 Meta-analysis results of sleep-disordered breathing prevalence by sub-groups.
Full size table

Individual participant data were available for the three largest studies [2, 6, 13], and were published for another study [33]. This allowed summaries of prevalence by sub-groups to be calculated in a sub-set of the included studies. All four of these studies were from our laboratory. A sensitivity analysis demonstrated no difference in the prevalence of at least moderate SDB between the studies from our laboratory and the others (68 vs. 59%, p = 0.33, Supplementary Fig. 5).

Table 3 summarises the results of the meta-analysis of at least moderate SDB prevalence (AHI ≥ 15) by sub-groups. Prevalence of SDB by sex, age group, SCI severity (AIS A vs. BCD), cervical SCI level (high vs. low) and BMI group for the four studies with individual participant data can be found in Supplementary Table 1. Sub-group meta-analyses performed on these data found no difference in prevalence between those with high and low cervical SCI (Table 3). SDB prevalence increased significantly with increasing age (p < 0.001). Prevalence was non-significantly higher in males than females (71 vs. 59%, p = 0.06), as BMI increased (62 vs. 74 vs. 76%, p = 0.07) and complete versus incomplete injuries (76% vs. 61%, p = 0.06). When grouped by AIS classification, prevalence did not necessarily increase with severity, although the highest prevalence was found in AIS A (76%) and the lowest in AIS D (48%; See Table 3 and Supplementary Figs. 6–11).

Discussion

This study is the first systematic review of SDB prevalence in tetraplegia with a meta-analysis and it confirms that the prevalence is high. Our results suggest that over 80% of people with tetraplegia have at least mild SDB, almost 60% have at least moderate and over a third have severe SDB. To our knowledge SDB prevalence is higher in tetraplegia than in any other clinical population. Estimates of OSA prevalence in the general population range from 9 to 38% (AHI ≥ 5) and 6 to 17% (AHI ≥ 15), are higher in men and with increasing age [34]. Three meta-analyses of at least mild SDB prevalence in people with cerebrovascular disease/stroke reported pooled prevalence estimates ranging from 62 to 72% (AHI ≥ 5) [35,36,37]. A meta-analysis of three studies reporting SDB following traumatic brain injury estimated the prevalence to be 25% (95% CI = 20–30) [38].

Within an empirical study investigating cerebrovascular reactivity in tetraplegia, Squair et al. included a meta-analysis of SDB prevalence from 11 published studies (682 people with SCI) [26]. The prevalence (AHI ≥ 5) was estimated at 73%. However, of the 11 included studies, five contained data that were not independent from other included studies, resulting in potential duplication of over 200 participants in the meta-analysis [26].

Other reviews summarising the prevalence of SDB in SCI have concluded that the prevalence is high, however our results suggest it may be even higher than previously reported [11, 12, 25]. There are several potential explanations for this. Firstly, this review focussed on people with tetraplegia only. Lower lung volumes, respiratory muscle weakness, sleeping in supine and disruption to the autonomic nervous system have all been postulated as explanations for the higher prevalence in people with tetraplegia than in those with paraplegia [19, 24, 39]. Secondly, we only included studies published after 2000 and data on incident cases of SCI demonstrate that the average age at injury has increased substantially over the past two to three decades [40]. Age is a known risk factor for SDB, as demonstrated in this study, and thus the increasing age at time of SCI may be reflected in our high estimate of the prevalence of SDB [34]. Finally, our meta-analysis only included studies using Level 1 or 2 overnight sleep studies to detect SDB, which may detect more hypopneas than Level 3 or 4 studies [41]. However a meta-analysis of 23 studies in stroke did not find any difference in SDB prevalence by sleep study type [36].

Our narrative synthesis summarising the reported associations between the presence of SDB and participant characteristics revealed conflicting results between studies, however the results of these individual studies should be interpreted cautiously due to the potential sampling biases and small sample sizes. Our meta-analysis of sub-groups, albeit in a sub-set of included studies all from the same laboratory, found that increasing age was significantly associated with the higher prevalence of SDB. While not statistically significant, there was some indication that a higher prevalence of SDB may be present in males, and in those with a complete SCI and a higher BMI. Increasing age, male sex and a higher BMI are known risk factors for SDB in general populations and our results suggest that they may also be important factors in tetraplegia [34]. The potential relationship between injury completeness and SDB remains unclear. Associations between AIS and SDB reported in the literature have been inconsistent. Berlowitz et al. found that those with motor and sensory complete (AIS A) injuries were more likely to have SDB, and Graco et al. identified that the risk of SDB increased with increased severity of SCI (AIS A–D) [2, 13]. However, others have found no associations between AIS and SDB [18, 22, 29, 42]. As stated by Giannoccaro et al. interpretation of these conflicting results is compromised by the differences in study samples, with several studies limiting inclusion to more severe injuries [12]. The relatively small sample sizes likely would also contribute to the inconsistent findings. Additionally, the inter-rater reliability of the International Standards for Neurological Classification of SCI scoring is variable, particularly for incomplete injuries, and may have introduced further variability into the results [43, 44]. Our sub-group analysis did not find any difference in the prevalence of SDB between low and high cervical SCI.

We did not plan to distinguish SDB prevalence by obstructive and CSA. However, of the studies reporting both, OSA was clearly the predominant form of SDB. One study, not included in this systematic review, contradicts this conclusion. In a sample of 16 participants with SCI, Sankari et al. demonstrated a high prevalence of CSA, defined as an AHI ≥ 5 (in non-REM sleep) and a central apnoea index of ≥5, which was higher in tetraplegia (63%) than in paraplegia (13%) [45]. The results should be interpreted cautiously in light of the very small sample size. Another study assessing SDB in a group of 91 people with SCI, using a Level 3 sleep study, reported 83% prevalence of OSA and 24% prevalence of CSA [16]. OSA and CSA were defined as an obstructive apnoea/hypopnea index of ≥5 events/h and a central apnoea index of ≥5 events/h, respectively. This supports our conclusion that while central events may be commonly present in SCI, OSA is the predominant form of SDB. Hypopnoeas are the most common respiratory event type in people with SCI, yet to our knowledge none of the SDB prevalence studies have considered whether hypopnoeas are central or obstructive in nature. More research into the nature of the hypopnoeas in people with SCI is needed to understand the aetiology of SDB.

The results of this systematic review and meta-analysis will hopefully raise awareness of the prevalence of SDB, among people with tetraplegia and the clinicians who treat them. Our results indicate that almost two thirds of people with tetraplegia have at least moderate SDB. The detrimental effects of SDB on physical health, mental health and quality of life are well established in both non-disabled and SCI populations, yet SDB in SCI has been largely neglected by the health care system [2, 46]. Two audits of US veterans with SCI have revealed low rates of SDB testing and treatment [9, 10]. A recent qualitative study involving interviews with rehabilitation doctors from SCI units concluded that the management of SDB in SCI is highly varied; many SCI units do not routinely screen for signs and symptoms, and when SDB is suspected the predominant model for diagnosis and management involves referral to specialist sleep services [7]. However access to these services can be poor, which may further contribute to the low rates of testing. People with tetraplegia are reluctant to stay overnight in sleep laboratories, because it disrupts their daily routines and many of the facilities do not accommodate their disability [8]. To address this problem, our group has developed an ambulatory test involving questionnaire and overnight oximetry to detect moderate to severe SDB in people with tetraplegia, which may provide an accessible alternative to a full in-laboratory sleep study [13]. The use of simple ambulatory diagnostic models has the potential to substantially increase the detection of SDB and improve access to treatment.

Other barriers to diagnosis and treatment must also be addressed in this population. Complications from SCI are common, and addressing poor sleep is often a low priority for both the clinician and the person with tetraplegia [7, 8]. Further complicating matters, symptom recognition can be poor among people with tetraplegia. In a study investigating the experience of people with tetraplegia who had recently been diagnosed and treated for SDB, several participants reported being unaware of their symptoms, surprised by their diagnosis and the improvement to their wellbeing once treated [8]. Increasing awareness of the potential benefits of diagnosis and treatment, among clinicians and people with SCI, could improve rates of diagnosis and access to treatment.

Limitations

Our results should be considered in light of several limitations. While our inclusion criteria for studies attempted to limit heterogeneity in study design, the meta-analysis still demonstrated substantial statistical heterogeneity. Heterogeneity is common among studies of disease prevalence per se [47]. Further, identification of SDB is inconsistent for many reasons. The intra- and inter-rater reliability of scoring respiratory events is low; different studies/laboratories use different rules for scoring hypopneas, and the threshold for diagnosis is commonly inconsistent [48, 49].

Five of the nine studies (549 of the 630 participants) included in the meta-analysis were from our group, which is likely to have introduced measurement and selection biases. A sensitivity analysis showed that the prevalence was higher in our five studies than in the others (68 vs. 59%), but this difference did not reach statistical significance. In making this comparison, we did not control for demographic differences such as age, sex and injury severity. Ideally, we would have included all nine studies in the sub-group analyses of age, sex, injury severity and BMI but these data were not available. We calculated SDB prevalence for these sub-groups from the individual participant data of four studies, all from our laboratory, and thus this result is potentially subject to the biases detailed above.

Given the inconsistent reporting of results used within this systematic review, we recommend that all future studies reporting the prevalence of SDB in SCI provide prevalence rates at AHI thresholds of 5, 15 and 30, with the type of study and event counting rules clearly articulated. Where possible, sub-group analysis by sex, age group and SCI severity should also be included, as per guidelines provided by the International Spinal Cord Society [50].

Conclusion

SDB is extremely common in people with tetraplegia. Almost two thirds have a degree of SDB that is likely to affect their health and quality of life. As in the general population, the prevalence increases with age, and may be higher in men and in those who are overweight. The findings of our study highlight the importance of providing access to SDB screening and treatment for this population. Given the high prevalence, the effects on health and quality of life and the potential for improvement with treatment, we assert that all people with tetraplegia should be routinely investigated for evidence of the disorder and offered treatment where indicated.