Introduction

Obstructive sleep apnea (OSA) is extremely common in middle aged men and women worldwide but particularly in North America. In one landmark study in 1993, 9 % of middle aged women and 24 % of men had an apnea–hypopnea index (AHI) > 5 h-1 [1]. As the obesity epidemic continues, estimates of prevalence obtained in the 1990s may underestimate the current problem as obesity is a major risk factor for the development of sleep apnea [2, 3]. The Wisconsin group has recently compared sleep apnea prevalence from 1988 to 1994 and from 2007 to 2010 using data collected for the Wisconsin sleep cohort study. Prevalence has increased substantially during this period due largely to increases in obesity but also perhaps due to other factors [4]. Patients can be defined as having the sleep apnea syndrome based on an abnormal AHI on polysomnography associated with sleepiness or sleep apnea can be identified solely on the basis of an abnormal sleep study. OSA is associated with many comorbidities particularly hypertension, cardiovascular disease, cerebral vascular disease and the metabolic syndrome [5, 6]. These associations are seen even if sleep apnea is identified solely on the basis of an abnormal AHI [7]. OSA is now recognized as a chronic disorder and guidelines are established for diagnosis and management of this condition to improve patient outcomes [8].

Recent studies have shown that sleep apnea is common in patients undergoing elective surgery [9–11]. The majority of patients with sleep apnea undergoing elective surgery have not been diagnosed prior to surgery. In one study, approximately 80 % of patients felt to have OSA pre-operatively had not been previously diagnosed [12].

Anesthetic agents and analgesics used for pain in the post-operative period may be particularly problematic in patients with OSA. These agents by relaxing upper airway muscles could exacerbate the patient's sleep apnea [13, 14]. There are reports of apnea/hypopnea episodes and respiratory arrest associated with opioid analgesics [15, 16]. There is no definitive data to attribute increased perioperative complications in OSA patients to the use of anesthetic agents and analgesics. However, given that these agents can cause airway collapsibility, immediate post-operative complications in the OSA patients may plausibly be attributed to these agents. The current guidelines, therefore, recommend avoidance of opioids, where possible, and cautious selection of the appropriate surgical setting [17].

There have been quite a few studies comparing the risk of post-operative complication rates in patients with and without sleep apnea [9, 12, 18–34]. The majority of studies have shown a higher post-operative complication rate in patients with sleep apnea. However, considerable heterogeneity has been noted between studies. A recent meta-analysis found that OSA was associated with a significantly higher risk of desaturation, intensive care unit (ICU) transfer, cardiovascular complications and acute respiratory failure [35]. However, three studies that met the inclusion criteria were not included in the meta-analysis [24, 30, 31]. These studies reported data on outcomes, such as hypoxemia, respiratory complications, cardiac complications, neurologic complications, and ICU transfer, and thus could substantially affect the effect estimates. Moreover, four studies included in the meta-analysis also reported on neurologic complications but this outcome was not analyzed in the prior meta-analysis.[18, 23, 27, 29]. There have been several studies published since then and thus substantial additional data is now available that should improve the precision of these estimates. As well, more recent data is now available to examine the incidence of neurological complications in patients with and without sleep apnea [36, 37]. For these reasons, we thought that a meta-analysis at this time would provide useful information to further refine the effects of sleep apnea on post-operative complications.

Polysomnography is the gold standard for the diagnosis of OSA. However, this test is expensive and resource-intensive. Thus, there has been considerable interest in developing questionnaires that identify patients at high risk for sleep apnea. These questionnaires might be particularly useful in identifying patients who are at high or low risk of sleep apnea before elective surgery as it is not practical to perform polysomnography on all patients undergoing elective surgery. A recent systematic review of screening questionnaires analyzed the characteristics of these questionnaires and provided a comprehensive review of validated screening questionnaires [38]. Accordingly, we included in our meta-analysis, patients who were diagnosed with or without sleep apnea by polysomnography as well as patients identified as high risk or low risk for sleep apnea based on a validated screening questionnaire. We performed a subgroup analysis looking only at the patients who were classified with or without sleep apnea by polysomnography.

Materials and methods

Eligibility criteria

We included observational studies (prospective and retrospective cohort studies, and case-control studies) that reported post-operative complications from non-upper airway surgery in adult patients. Studies conducted on pediatric or adolescent population, case reports, case series, and abstracts were excluded. Studies reporting complications from upper airway surgery and endoscopic procedures were also excluded. Studies that did not use a validated prediction model or diagnostic tool for OSA and those that established OSA diagnosis based on International Code Diseases (ICD)-9 codes were also excluded from the meta-analysis (Table 2). Post-operative complications resulting from surgery itself like bleeding and infection were excluded from analysis.

Search strategy

In August 2012, we electronically searched Medline (1950 onwards; access via Ovid), Embase (all years; access via Ovid), and Evidence-Based Medicine Reviews using a detailed search strategy with search terms outlined in Appendix 1. The search strategy was outlined by the authors after initial detailed discussion of the aim of the study and the research question. There was no restriction on the language in which studies were reported. Preliminary review of related studies and keywords for some of the best studies done in the field were used as well. The search strategy was reviewed and approved by two librarians who are experts in literature search field. As we proceeded further in the literature search, using the selected studies, we searched for “related articles” in PubMed. We also manually searched the references of the included studies to confirm the inclusion of all related articles. In April 2013, we repeated the search for related articles in Pubmed, and to our knowledge, there are no other new publications that we did not include on our review.

Selection process

Using a standardized screening guide, two reviewers independently reviewed the titles and abstracts of the citations resulting from the search. Full text for the articles which were thought to be eligible by at least one of the reviewers was obtained. Each reviewer independently reviewed these full texts to judge eligibility to be included in our review. Any disparities within the two reviewers about including studies were discussed and resolved by a third reviewer.

Data abstraction

Two reviewers independently extracted data from the included studies using a standardized and pilot-tested data abstraction form. After completion, the reviewers tried to resolve differences in extracted data by discussion and, if not resolved, by discussion with the third reviewer. The pilot-tested form included study design, population, method used to diagnose or predict OSA, methodological characteristics of the study and results reported. We assessed for the validation of the tests used to establish a diagnosis of OSA and also the validity of any questionnaire used to predict OSA in the study population. Results from each study were assessed to exclude perioperative complications like bleeding and infection which are a result from the surgery as such. We recorded the effect measures derived from the regression models that adjusted for the maximum number of covariates. Results from the data abstraction from the included studies are shown in Table 1. We rated the overall quality of evidence for each outcome using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach [39]. Overall quality of evidence from the included studies for most outcomes is estimated to be moderate to low.

Table 1 Methodological characteristics of studies reporting (included in meta-analysis) the risk of post-operative outcomes in OSA patients
Full size table

Data analysis

We calculated the kappa statistic to evaluate the agreement between the two reviewers who assessed the studies full texts for inclusion eligibility. Meta-analysis was conducted for all the outcomes that were reported by at least three studies in which the population were either diagnosed with OSA by a polysomnography (PSG) or were predicted to have OSA using a validated questionnaire. Sub-group analysis was done on the patient population who were diagnosed with OSA using polysomnography. In studies that did not report actual number of incidents, percentages or other available data was used to calculate the number of events for the outcomes evaluated. For each outcome, we pooled the odds ratios (OR) of eligible studies using the generic inverse variance and the random effects model in Review Manager Version 5.0.20. The random effects model was used as the included studies evaluated different patient populations. We measured homogeneity across study results using the I 2 statistic. We checked for possible publication bias using inverted funnel plots.

Results

Our extensive electronic database search resulted in 5,127 articles. Appendix 2 shows an outline of the search process. One study [9] that used a nonvalidated questionnaire for OSA risk assessment and two other studies [19, 20] that used ICD codes for OSA diagnosis were reviewed in detail, but excluded from meta-analysis. A study conducted by Finkel et al. [12] used the ARES questionnaire to categorize patients into high-risk and low-risk OSA groups. However, the outcomes were monitored only in a sub-set of the high-risk group who were diagnosed with OSA after a PSG. Therefore, the study was excluded since the outcomes were not compared to a non-OSA group. Studies by Neligan et al. [33] and Rennotte et al. [32] were randomized controlled trials which did not compare the outcomes with a non-OSA group. The study by Hallowell et al. [26] was a quality improvement study. Changes in practice over time in terms of obtaining sleep studies and managing patients peri-operatively made the global estimates of post-operative complications inaccurate, and thus, the study could not be included in our analysis. Another study [40] was excluded since only an abstract form was available which provided insufficient information to calculate the number of events for evaluated outcomes. Iyer et al. [34] compared outcomes in severe OSA group and nonsevere (including non-OSA, mild OSA, and moderate OSA groups) OSA groups. The study was excluded due to the inability to extract data for the group without OSA as the reference population.

Our literature search resulted in 18 observational studies which reported the incidence of hypoxemia, re-intubation, respiratory complications, cardiac complications, neurological complications, unplanned ICU transfer, prolonged oxygen therapy, and length of hospital stay [9, 18–25, 27–31, 36, 37, 41, 42]. The patient population in these studies was categorized into those with a diagnosis or at high-risk for OSA vs. those without a diagnosis or at low-risk for OSA. Table 1 contains the methodological qualities and results reported by the 15 studies included in our meta-analysis [18, 21–25, 27–31, 36, 37, 41, 42]. Polysomnography was used to confirm the diagnosis in most studies [18, 23, 24, 27–31, 36, 37], while others used nocturnal oximetry [22] or validated questionnaires like Berlin questionnaire [25], Flemon's criteria [21, 41] or STOP-BANG questionnaire [42] to assess the risk for OSA. Table 2 contains the methodological qualities and results reported by the three studies included in our systematic review but excluded from the meta-analysis as outlined above [9, 19, 20]. Various post-operative outcomes were reported as the number of patients with the outcome in both comparison groups. Agreement between the reviewers for study eligibility was excellent (kappa = 0.946). Using GRADE approach, the quality of evidence from the studies included in our review was found to be moderate to low.

Table 2 Methodological characteristics of studies (excluded from meta-analysis) reporting the risk of post-operative outcomes in OSA patients
Full size table

Hypoxemia

A total of 12 studies that compared post-operative hypoxemia or oxygen desaturations in patients with and without OSA were included in analysis [18, 21–24, 27–29, 31, 37, 41, 42]. All these studies reported the outcome as number of patients experiencing post-operative hypoxemia in both the groups. Study conducted by Ahmad et al. [25] reported the number of hypoxemic events in each group rather than number of patients experiencing the event. Therefore, results from this study could not be included in the analysis. Auckley [31] reported the number of patients requiring supplemental oxygen, and these patients were assumed to have hypoxia, and data used for analysis. The overall OR for the outcome is 3.06 (95 % CI 2.35–3.97; I 2 = 0 %), indicating that hypoxemia is significantly higher in OSA patients after non-upper airway surgery. Subgroup analysis including only the studies that used polysomnography [18, 23, 24, 27–29, 37] showed an OR of 2.75 (95 % CI 1.89–4.00; I 2 = 0 %) with the significance still persistent (Fig. 1).

Fig. 1
figure 1

Association between OSA and post-operative hypoxemia

Full size image

Re-intubation

Data from six observational studies was included in the analysis of incidence of re-intubation [18, 25, 27–29, 37]. All the included studies used PSG to diagnose OSA. Meta-analysis with the reported data resulted in OR of 1.37 (95 % CI 0.65–2.91; I 2 = 0 %). Results indicate that the incidence of re-intubation in patients with OSA was not significantly higher (Fig. 2).

Fig. 2
figure 2

Association between OSA and reintubation

Full size image

Respiratory complications

Results from ten studies have been included in the analysis to compare the incidence of any respiratory complication in OSA group vs. non-OSA group [21–25, 27, 28, 31, 37, 42]. Included studies reported incidence of one or more respiratory complications varying from hypoxemia to re-intubation (detailed description in Table 1). Data from studies that reported only a single respiratory complication was included in the analysis of this outcome. Two other studies [29, 41] reported the number of patients experiencing each respiratory complication. However, these studies did not report the total number of patients experiencing at least one of the respiratory complications, and therefore, the results could not be included in the meta-analysis. Analysis resulted in an overall OR of 2.77 (95 % CI 1.73–4.43; I 2 = 23 %) showing a statistically significant increased risk in the OSA group. Subgroup analysis including only studies that used polysomnography [23–25, 27, 28, 37] did show a significantly increased incidence in OSA group compared to control group; OR of 2.03 (95 % CI 1.13–3.64; I 2 = 22 %; Fig. 3).

Fig. 3
figure 3

Association between OSA and post-operative respiratory complications

Full size image

Cardiac complications

Twelve observational studies reported various post-operative cardiac complications including dysrhythmias, abnormal heart rate, myocardial infarction and ischemia, hypotension, and congestive heart failure [18, 21–25, 27–30, 37, 42]. Included studies reported the incidence of one or more cardiac complications as reported in the data extraction tables. Data from studies that reported a single cardiac complication was also included in the analysis of this group of outcomes. Study by Kaw et al. [18] reported incidence of individual cardiac outcomes, such as myocardial ischemia and infarction, but did not report the total number of patients experiencing any type of cardiac complication. We attempted to obtain the data from the author and did not get any response. However, a recent meta-analysis by the same author [35] included the data which was not reported in the original paper. Numbers from this study have been included in our meta-analysis. Data analysis showed a significantly increased incidence of the outcome in the OSA group with an OR of 1.76 (95 % CI 1.16–2.67; I 2 = 0 %). Subgroup analysis showed a persistent significantly increased incidence in patients diagnosed with OSA using only PSG [18, 23–25, 27–30, 37]; OR of 1.83 (95 % CI 1.16–2.90 I 2 = 0 %; Fig. 4).

Fig. 4
figure 4

Association between OSA and post-operative cardiac complications

Full size image

Neurological complications

Seven studies compared the incidence of post-operative neurological complications like delirium, agitation, confusion, and excessive drowsiness [18, 23, 24, 27, 29, 36, 37]. All the included studies used PSG to diagnose OSA. Statistical analysis revealed a significantly increased incidence of the outcome in patients with OSA (OR = 2.65; 95 % CI 1.43–4.92; I 2 = 0 %; Fig. 5).

Fig. 5
figure 5

Association between OSA and post-operative neurologic complications

Full size image

Unplanned ICU transfer

Number of patients requiring post-operative unplanned ICU transfer was compared and reported by ten observational studies [18, 21, 23, 24, 27, 29, 31, 37, 41, 42]. Most of the studies reported number of patients requiring unplanned ICU transfer, whereas others [18, 21, 26, 27, 42] reported number of ICU transfers. Review of the descriptions provided in these studies does suggest that these ICU transfers were unplanned. For study by Kaw et al. [27], our attempt to get additional data from the authors failed. Therefore, we included the numbers reported by the same authors in their recent meta-analysis [35]. Chung et al. [24] reported no unplanned ICU transfers in both the groups. Meta-analysis of the data yielded an OR of 2.97 (95 % CI 1.90–4.64; I 2 = 1 %). Subgroup analysis after exclusion of the studies that used a questionnaire to predict occurrence of OSA [21, 31, 41, 42] still showed a significantly increased incidence in the sleep apnea group; OR of 2.36 (95 % CI 1.26–4.45; I 2 = 21 %; Fig. 6).

Fig. 6
figure 6

Association between OSA and post-operative unplanned ICU transfer

Full size image

Other outcomes

Length of hospital stay

Eight studies reported results comparing the length of post-operative hospital stay in OSA vs. non-OSA patients. Statistically significant increased duration of stay was reported by three studies [29, 30, 43], whereas the results from five other studies [22, 25, 27, 37, 44] showed no significant effect of OSA on the outcome, as reported in Table 1. Due to inconsistency in the way the data were reported, analysis for a combined effect could not be done.

Prolonged oxygen therapy

Three studies reported the effect of diagnosis or risk of OSA on the need for oxygen therapy after transfer out of post-operative recovery unit (PACU) [23, 24, 29]. All the studies found that patients with OSA or at high risk of OSA required prolonged oxygen therapy post-operatively, compared to patients without OSA or at low risk of OSA.

Discussion

The principal findings of our study is that patients with OSA or at high risk of OSA are more prone to adverse perioperative outcomes including respiratory, cardiac, neurological complications, and unplanned ICU transfer . Subgroup analysis including only studies in which sleep apnea was diagnosed by polysomnography yielded similar results.

Respiratory complications

Not surprisingly, patients with sleep apnea had a greater risk of having hypoxemia post-operatively than those without sleep apnea. Since many of the studies were retrospective in nature, episodes of hypoxemia may have been missed that would have been detected with a more systematic prospective approach. However, there is no reason why hypoxemia would be preferentially missed in the non OSA group so that the differences between groups are significant, and the overall magnitude of the problem may have even been underestimated. Hypoxemia by itself is more of a sign than a complication per se. Nevertheless, hypoxemia is clearly undesirable and can potentially precipitate more serious complications including respiratory failure, re- intubation, cardiac ischemia or cardiac arrhythmias. If persistent, it must also be treated. Thus, its presence complicates the post-operative period and may be a harbinger of more serious complications. We also found that patients with sleep apnea had a greater risk of being on oxygen for a prolonged period of time.

We found no increase in the rate of re-intubation in patients with OSA. Similar results have been reported previously [35]. Our analysis included two additional studies including over 400 patients while excluding one study where re-intubation rates were incompletely reported.

Total respiratory complications were increased in patients with sleep apnea. The type of respiratory complications measured varied substantially between studies which decreased the utility of this measure. When only studies that included polysomnography were included, this variable remained statistically significant. Further studies preferably with a prospective study design are needed before definitive statements can be made about the effect of sleep apnea on total respiratory complications. However, it does appear that hypoxemia (sustained requiring treatment) is more common in patients with sleep apnea. In a huge administrative database study, patients with sleep apnea had a greater risk of aspiration pneumonia, adult respiratory distress syndrome, and mechanical ventilation after orthopedic and general surgeries [19]. This latter study would indirectly suggest an increased re-intubation rate in patients with sleep apnea. Since re-intubation was relatively uncommon in all the studies we included in our meta-analysis, our study may not have had sufficient power to adequately assess this outcome even though there were almost 1,800 patients included in the analysis.

Cardiac complications

We found an increased risk of cardiac complications in patients with OSA compared to those without. This was present even when only studies where sleep apnea was confirmed by polysomnography were included in the analysis. Reporting of cardiac events varied between studies and as a result some studies had very few cardiac events reported. Despite these issues, cardiac events were clearly increased in the sleep apnea group. However, patients with sleep apnea also are more obese, have greater incidence of coronary artery disease and diabetes all of which may influence the incidence of cardiac complications post-operatively. Most studies adjusted for these confounders when examining post-operative risk while some did not. We conducted a sub group analysis including results from the studies that matched patient characteristics or adjusted for them during analysis [18, 21, 22, 25, 27–30, 37, 42]. Results showed significantly increased incidence of post-operative cardiac complications in patients with OSA (OR = 1.87; 95 % CI 1.23–2.83; I 2 = 0 %; Fig. 7). Thus, sleep apnea does appear to increase the post-operative risk of cardiac complications.

Fig. 7
figure 7

Association between OSA and post-operative cardiac complications; no confounders

Full size image

Neurologic complications

We found an increased risk of neurologic complications including delirium, agitation, confusion, and excessive drowsiness in patients with OSA. Since, these symptoms were detected retrospectively in many of the studies; their true incidence was likely underestimated. None of the studies included reported any morbidity or mortality attributable to these neurologic complications. It is worth noting though that neurologic complications like delirium in the elderly surgical patient in the immediate post-operative period have the potential to significantly complicate the patients' post-operative course leading to a prolonged hospital stay or other adverse outcomes [45].

Unplanned ICU transfer

We found an increased risk of ICU transfer in patients with sleep apnea. There was a difference in reporting among studies with some studies reporting only unplanned ICU transfers [21, 23, 24, 29, 37, 41], while others reported all ICU transfers [18, 27, 31, 42]. However, in those studies in which ICU transfer was not specified as unplanned, it does appear that ICU transfers were really unplanned based on the descriptions provided in the studies. In one study, all patients were transferred to the ICU overnight who had severe sleep apnea (planned transfer) [23]. Since unplanned ICU transfer was the outcome reported, the number of unplanned ICU transfers would be underestimated in the sleep apnea group since there could be no unplanned transfers in those with severe sleep apnea. In a quality improvement study by Hallowell et al. [26] in patients undergoing gastric surgery for morbid obesity, there were a number of measures introduced for patients with sleep apnea that reduced ICU transfers starting in 2004–2005. This suggests that focused interventions in such patients and active surveillance for sleep apnea may lead to a reduction in post-operative complications. Length of hospital stay was significantly increased in three out of the eight studies where it was reported. Unfortunately, due to the way the data was reported, pooling of data to calculate the overall effect could not be calculated.

Limitations of study

The studies on which this meta-analysis is based were not of high methodological quality which lends caution to our conclusions. The criteria used to identify respiratory, cardiac, and neurologic complications varied substantially among studies which could make pooling of the data problematic. Despite this limitation, analysis did not show substantial heterogeneity among studies supporting our decision to pool the data. We evaluated for possible publication bias by looking at the funnel plots in meta-analysis. We cannot exclude the possibility of publication bias for two outcomes, namely, post-operative hypoxemia and unplanned ICU transfer. Another limitation is the retrospective nature of some of the studies which relies on reporting of complications in the chart by clinicians who may have different definitions for the complications noted. While there should not be a selective difference in reporting between patients with and without sleep apnea, the retrospective nature likely leads to systematic underreporting of complications compared to a prospective study using validated tools for daily measurement. This is especially important for subjective outcomes such as delirium, whereas re-intubation or unplanned ICU transfer should be less affected by this methodological concern. Another important point is that meta-analysis is by design a univariate analysis and not adjusted for confounding variables that are likely different between patients with and without OSA and may be influencing the observed association between OSA and post-operative outcomes. However, most of the individual studies on which the meta-analysis is based attempted to correct for confounders, such as body weight, diabetes, coronary artery disease, etc., and still found a significant effect for OSA on post-operative outcomes [18, 21, 22, 27–30, 37, 42]. Finally, length of stay is influenced by many factors other than medical acuity such as availability of sub-acute rehabilitation beds in patients after hip replacement that may mask real differences between patients with and without sleep apnea.

Implications for clinical practice

The increased risk of post-operative complications in patients with OSA suggests that identification of patients at high risk for OSA pre-operatively might be clinically worthwhile. Identifying treatment algorithms for patients with sleep apnea to reduce post-operative complication rates will require further investigation. However, increased vigilance and monitoring of such patients may lead to a reduction in post-operative complications [21, 32, 33, 46]. Whether specific treatment with CPAP in patients at high risk for sleep apnea naïve to CPAP therapy will improve post-operative outcomes remains to be investigated but a recent pilot study did not show beneficial effects [47]. Another consideration is the venue for surgical procedures. Ambulatory stand-alone centers are becoming increasingly utilized for surgical procedures that do not require post-operative inpatient care. If sleep apnea patients are operated at these sites then they must be equipped to handle acute airway collapse in the post-operative recovery room and be staffed to handle potentially longer and more difficult recovery periods [48]. The ideal setting and monitoring for patients with sleep apnea is also an area that requires further study.

In conclusion, patients with sleep apnea or at high risk for sleep apnea have an increased post-operative risk for respiratory, cardiac, and neurologic complications as well as an increased risk of ICU transfer.