Sleep apnea (SA) is a major challenge in the postoperative period. As many as one-fourth of patients undergoing elective surgery may be affected.1 The prevalence among orthopedic patients undergoing joint arthroplasty may be especially high, given that obesity is a widespread comorbidity in this patient population.2 Despite the increasing level of concern that SA is associated with increased risk for postoperative complications,2–7 there remains a paucity of population-based information available in the literature regarding postoperative outcomes. Most available data are from relatively small samples and academic institutions, thus limiting external validity and applicability. Large-scale observational studies, using secondary administrative databases, are increasingly being performed, because they provide more robust information on the impact of specific diseases in a more representative care setting.
Given the combination of a high prevalence of SA among orthopedic surgery patients2 and the projection that by 2030 >4 million hip (THA) and knee (TKA) arthroplasties will be performed in the United States alone,8 the joint replacement population is an especially important group of patients in need of further investigation.
Despite some data suggesting an increased risk for postoperative pulmonary complications associated with SA among orthopedic patients,2 more detailed analysis of other important outcomes such as utilization of economic resources remains largely unexplored. Such information is important to assess and gain better insights into the clinical and economic impact of SA in patients undergoing surgery.
Therefore, we analyzed data on >500,000 patients from approximately 400 institutions. We hypothesized that THA and TKA patients with SA (1) were more likely to experience postoperative complications and (2) consumed greater hospital resources, as represented by an increased likelihood for a longer length of hospital stay and greater use of economic resources.
Database and Study Design
For this study, data from Premier Perspective, Inc.’s (Charlotte, NC) collected between 2006 and 2010 were used. This retrospective administrative database contains discharge information from approximately 400 hospitals9,10 and is compliant with the Health Insurance Portability and Accountability Act. Because data are de-identified, the study was exempt from review by the Hospital for Special Surgery IRB. Before distribution, rigorous quality assurance and data validation procedures are used by the provider to assure the accuracy of entries. This database has been used for other studies by our group.11,12
The study population consisted of all patients in the Premier Perspective database, undergoing primary THA and TKA, as identified by International Classification of Diseases-9th revision-Clinical Modification codes (ICD-9-CM) 81.51 and 81.54, respectively.
The presence of SA was determined by the presence of respective ICD-9 codes. Appendix 1 lists specific diagnosis codes included and their individual prevalence.
Patient, procedure, and health care-related characteristics analyzed were age, sex, race (Caucasian, African-American, Hispanic, other), admission type (emergent, elective, other), hospital size (<300, 300–499, ≥500 beds), hospital location (urban, rural), hospital teaching status, anesthesia type (general, neuraxial, neuraxial-general, unknown), indication for surgery (osteoarthritis, rheumatoid arthritis, other), type of surgery (THA, TKA), year of surgery, and comorbidity prevalence (myocardial infarction [MI], cerebrovascular disease, peripheral vascular disease, renal disease, chronic obstructive pulmonary disease [COPD], uncomplicated and complicated diabetes mellitus, uncomplicated and complicated systemic hypertension, [“complicated” as defined by the absence or presence of disease-related end organ complications], cancer, obesity, and pulmonary hypertension). Overall, comorbidity burden was assessed with the Deyo adaptation of the Charlson comorbidity index method for use with administrative data for surgical outcomes.13 In brief, the Deyo Index comprises a number of comorbidities. Each comorbidity is assigned a severity weight, and its presence contributes to an overall score. A higher score correlates with increased risk of adverse outcomes.
Individual major postoperative complications studied were pulmonary embolism, deep venous thrombosis, cerebrovascular events, pulmonary complications, sepsis, cardiac complications (excluding MI), MI, pneumonia (including ventilator-associated pneumonia and aspiration pneumonitis), infectious complications, acute renal failure, gastrointestinal complications, and mortality. To evaluate these major complications, a combined outcome variable (“combined complications”) was created to indicate having at least one of the complications listed above. A case with “pulmonary complications” had at least 1 indication of pulmonary compromise, pneumonia, or pulmonary embolism. For “cardiac complications,” cases had an indication of cardiac complications (except MI) or MI.
In addition, utilization of critical care, stepdown and telemetry services (each defined by specific billing records for these services representing distinctly different levels of care), blood transfusions, postoperative mechanical ventilation, and noninvasive ventilation were studied. Utilization of economic resources in U.S. dollars and length of hospitalization were compared as continuous variables. Due to their skewed distributions, they were also dichotomized such that entries exceeding the 75th percentile were defined as increased length of hospitalization or increased use of economic resources, respectively. This approach was used by our group in various other publications.11,12 Furthermore, using this cutoff was not influenced by the length of hospitalization or patient costs of SA patients. To account for potential bias in choosing a cutoff for dichotomization, sensitivity analyses using cutoffs ranging from 50% to 90% were performed in the univariable analysis, and similar results were found. ICD-9 CM codes and billing data provided by the Premier database were used to define the presence of comorbidities, complications, and other outcomes and are listed in Appendix 2.
The primary goal of our analysis was to compare different outcomes between patients with and without SA. All statistical analyses were performed using SAS version 9.3 (SAS Institute, Cary, NC).
Patient- and Health Care-Related Characteristics by Presence of SA
Groups with and without SA diagnosis were compared regarding patient and health care-related characteristics in the univariable analysis. Means (standard deviation [SD]) and percentages were described for continuous and categorical variables, respectively. Length of stay and economic resource utilization exhibited a skewed distribution and were presented using median and interquartile ranges. Due to the large sample size, a significant difference of P (< 0.05) for differences between 2 groups using traditional t tests, Wilcoxon rank sum tests, or χ2 tests were very likely to be detected but may not be clinically meaningful. Therefore, standardized difference (STD) was calculated to measure group balance.14 A STD <0.1 for a continuous and 10% for a categorical variable indicated a negligible difference in the mean or proportion of a variable between groups.15 Due to the large sample size, the SEs of STDs were very small; therefore, and to support clarity of presentation, they were not shown. Percentage of missing data was reported for all study variables, stratified by presence of a SA diagnosis.
Logistic Regression Analyses
Univariable and multivariable logistic regression analyses were performed to evaluate the association between patients with and without SA. Separate models were fitted for the binary outcomes: combined complications, pulmonary complications, cardiac complications, mortality, mechanical ventilation, noninvasive ventilation, use of blood product transfusion, intensive care utilization, stepdown/telemetry service utilization, increased length of hospitalization, and increased economic resource utilization. Covariates included for controlling purposes comprised age, gender, race, admission type, hospital size, hospital teaching status, hospital location, anesthesia type, indication for surgery, type of surgery, year of surgery, and individual comorbidities. The association between each covariate and outcome variable was performed by univariable analysis. Almost all associations had P < 0.05 and were entered into the multivariable model. Few covariates (e.g., gender and hospital characteristic) had P > 0.05 for the outcomes of mortality, and cardiac complication, but were included in the model due to the consensus that they were of clinical importance.16
In addition to the above-mentioned main effects, we evaluated the interaction terms of SA with age, gender, year, COPD, diabetes, obesity, hypertension, and complicated hypertension for each of the outcomes. These interaction terms were selected based on (1) clinical relevance; (2) STD >10% between SA versus non-SA status; and (3) sufficient frequency (> 5% prevalence of comorbidities in SA) to achieve adequate power and obtain valid estimates. All interaction effects were included in each model, and a backward approach was used for testing the significance of interaction effects while keeping all main effects in the model. The significance of interaction effects was measured using a P of 0.004 (Bonferroni-corrected P = 0.05/11 outcomes)17 to adjust for multiple outcomes. If an interaction effect had a P < 0.004, but the corresponding coefficient was very small (e.g., < 0.001), we considered it quantitatively unimportant and removed it from the model. For models with significant interaction terms, the interpretation of SA effect would be conditioned on the terms interacted with.
Missing data were excluded from analyses, but because 27.6% of cases had “unknown” anesthesia, they were treated as a separate category and modeled as a sensitivity analysis.
Crude and adjusted odds ratios (OR), Bonferroni-corrected 95% confidence intervals (CI) and P values were reported due to multiple comparisons. Two-sided P < 0.05 (conventional threshold of significance) was used to determine significance of variables. Ninety-five percent CIs of estimates were reported to enable readers to interpret the significance of the findings; this was done to alleviate the potentially undue effect a very large sample size might have on the P values.
Diagnostic of Models
To evaluate independence of individual predictor variables, the value inflation factor was calculated for each predictor variable to determine whether multicollinearity was present. The final models were validated using the Hosmer-Lemeshow (H-L) test.18 It evaluated adequate calibration of a logistic regression model so that the probability predictions from the model reflected the true occurrence of events in the data. The area under the receiver-operating characteristic curve19 (c-statistic) was used to measure the level of model discrimination between observed data at different levels of the outcome. Discrimination was classified as perfect, excellent, very good, good, moderate, and poor if the area under curves were 1.0, 0.9 to 0.99, 0.8 to 0.89, 0.7 to 0.79, 0.6 to 0.69, or <0.6, respectively.20 To evaluate whether populations differed among outcomes, the amount of patients with multiple outcomes was determined. Statistical analyses were performed using SAS software version 9.3 (SAS Institute, Cary, NC).
Sensitivity Analysis Based on Propensity Score Matching
The OR from a matched sample using the propensity score method was performed as a sensitivity of the results to different statistical approaches. All covariates used in the primary analysis above were included in the multivariable logistic regression with the outcome variable of SA versus non-SA to calculate propensity scores. One SA patient (case) was matched with 3 non-SA patients (controls) for statistical efficiency.21 The matched pairs were generated by comparing the predicted propensity scores between cases and controls using the SAS macro %onetomanymatch with 8 to 1 digit match without replacement.22 Based on the matched sample, the effect of SA on outcomes was tested for significance using the Cochran-Mantel-Haenszel (CMH) test. To account for matching samples, common odds ratio, an overall OR across pairs of matching samples, and Bonferroni-corrected 95% CIs were estimated. For comparison purposes, multivariable models with main effects only were performed and reported.
Characterization by Presence of SA
We identified 530,089 entries for patients undergoing THA and TKA between 2006 and 2010. Overall, 8.4% had a diagnosis code for SA. The prevalence of SA increased from 6.2% in 2006 to 10.3% in 2010 (Fig. 1). Compared with non-SA patients, individuals with SA were on average younger (SA: 63.4 ± 9.6 years vs non-SA: 66.2 ± 11.3 years, STD = 26.7%), more frequently male (53.9% vs 37.3%, STD = 33.63%), carried a higher overall Deyo comorbidity burden (1.00 ± 1.12 vs 0.59 ± 0.92, STD = 40.1%) and had a higher prevalence of most individual comorbidities (Table 1, Table 2). Missing data were limited to the categories of anesthesia type, race, admission type, and payor type (28.7%, 19.4%, 0.3%, and 2.8%, respectively).
SA patients exhibited a higher incidence of major postoperative complications including pulmonary, cardiac (non-MI), and renal outcomes. Analysis of the various subtypes of pneumonia yielded a higher incidence of postprocedural aspiration pneumonia and/or Mendelson’s syndrome (as defined by ICD-9 code 997.39) in the SA group, compared with the no SA group (0.9% vs 0.6%, P < 0.0001). SA patients more frequently used critical care, telemetry, and stepdown unit services, and more commonly received postoperative mechanical ventilation and noninvasive ventilatory support. SA patients received fewer postoperative blood transfusions compared with non-SA patients. Appendix 3 details the incidence of individual cardiac complications, which suggests that the higher incidence of atrial fibrillation among SA patients was responsible for the increased rates in this complication category. Median length of hospitalization and economic resource utilization was similar among groups. Table 3 lists postoperative complication and resource utilization rates.
Logistic Regression Analyses
Table 4 details the results of the univariable and multivariable logistic regression analysis. Table 5 details the effect of SA for the models with significant interaction modifications.
Postoperative Complication Outcomes
No significant interaction terms were found for the models analyzing outcomes of combined complications, pulmonary complications, cardiac complications, and mortality. A diagnosis of SA emerged as an independent risk factor for the outcome of combined complications, as well as pulmonary and cardiac complications separately, but not for mortality.
Hospital Resource Utilization Outcomes
Significant interaction terms were detected for the models assessing outcome of mechanical ventilation (e.g., complicated hypertension, COPD), noninvasive ventilation (e.g., complicated hypertension, COPD), utilization of critical care (e.g., gender, year), stepdown/telemetry services (e.g., year, complicated hypertension, obesity), and prolonged length of stay (e.g., obesity). None were found for the outcomes of the need for blood transfusion and increased economic resource. The details of ORs conditioned on the modifications are shown in Table 5, and the SA effect on outcomes will have to be interpreted in the context of these modifications. When considering interaction terms, SA was associated with increased odds for mechanical ventilation, noninvasive ventilatory support, utilization of intensive care unit, stepdown and telemetry services as well as prolonged length of stay and increased economic resources.
In the sensitivity analysis including “unknown” anesthesia as a separate category, the results were similar.
The value inflation factors were all <10, indicating that no multicollinearity was present. The ranges of c-statistics were 0.7 to 0.9 except for the model evaluating increasing economic resource utilization (c = 0.6), indicating good to very good discrimination for most outcomes. The percentage of patients with multiple outcomes was 0.44% and 22.89% for postoperative complications and resource utilization outcomes, respectively. Among all outcomes, 26.54% of patients had at least 2 of any of the outcomes evaluated, indicating differences between outcome populations.
Sensitivity Analysis Based on Propensity Score Matching
Of 32,789 SA patients in the sample, 28,177 were successfully matched to non-SA patients. The propensity score-matched samples were well balanced (STD< 10%) between groups in terms of demographic variables and comorbidities (Appendices 4, 5). The common ORs were similar to the ORs found in the analysis with main effects only (Appendix 6).
In this study, we were able to show that SA was associated with higher rates and odds of postoperative complications, utilization of resources, and length of stay.
We observed that SA was associated with a 47% increased odds for the combined outcome of postoperative major morbidity. Increased odds for adverse outcomes among SA patients have been described,2–7 but information on a wide range of outcomes beyond pulmonary complications in the setting of orthopedic surgery remains rare. While the exact mechanisms by which SA confers increased odds for complications remains unknown, a number of abnormalities have been described among SA patients that may lower the clinically relevant injury threshold for various organ systems to exhibit signs of dysfunction. For example, SA is associated with higher baseline levels of systemic and pulmonary inflammation,23 decreased pharyngeal sphincter function,24 and increased sensitivity to the respiratory-depressant effects of opioids.25 These and other pathologic states may contribute to the increased susceptibility of SA patients to perioperative insults, such as transfusion and ventilator-related lung injury, and aspiration. However, it must be noted that not all our findings corroborate with available literature. For example, we did not find differences in the rates of cerebrovascular disease and complications between the 2 groups. Previous research has suggested that the presence of SA may indeed increase the risk for stroke,26 without allowing for inferences to be made in the postoperative setting. A factor to be considered when interpreting our findings is the fact that only patients with a known diagnosis of SA are included in our cohort and that use of positive airway pressure therapy, which may reverse some of the pathophysiology predisposing to long-term adverse outcomes, may be more likely used in this population.
In addition, we identified lower rates of blood transfusions among SA patients in our study. Feasible explanations for this finding are not obvious but warrant further inquiry. One possibility includes higher starting hematocrit levels frequently found in SA patients.27
Recent literature has further suggested a lack of evidence for increased mortality among SA patients.28,29 While speculative, an increase in vigilance among clinicians may indeed lead to better detection of complications in this patient group perceived to be at risk, thus allowing for interventions to avoid this extreme outcome despite higher complication rates.
In addition to the increased odds for adverse medical outcomes, we were able to show an effect of the presence of SA on increased resource utilization. The argument can be made that at least some of the increased utilization of services is not an indication of higher morbidity but reflects planned use of monitored settings and perioperative positive airway pressure equipment in an attempt to reduce complications. However, despite the recommendation by the American Society of Anesthesiologists task force on perioperative care of patients with obstructive sleep apnea that patients with SA be observed in a monitored setting postoperatively and treated with positive pressure ventilation in certain cases,30 little data are available on the use of resources such as telemetry, stepdown, and intensive care units. If the utilization of these resources would have to be interpreted in this context, the conclusion to be drawn would point toward a surprisingly low use of perioperative monitoring and use of ventilatory assistance. Indeed, there remains a paucity of data regarding the adoption of guidelines for the perioperative care in current practice. Interestingly, a single published inquiry into the existence of perioperative policies among anesthesia departments in Canada concluded that only 28% had such provisions.31 While lack of proof that these interventions lead to improved outcomes among SA patients may be 1 reason, the additional use of economic resources associated with implementation of these practices on a wider level certainly is a contributing factor. It is therefore not surprising that in our study SA was associated with higher odds for this outcome.
Our study is subject to a number of limitations. As a consequence of retrospective database analysis, clinically important covariates are not obtainable. However, a very large sample size provides access to outcomes as seen in actual practice. A further limitation is the reliance on ICD-9 coding for the diagnosis of SA. Thus, it is not possible to correlate the diagnosis with the severity of SA. This also applies to the severity of various other comorbidities. It is also almost certain that the true incidence of SA is higher than that reported here, as only patients with a preoperative diagnosis code for SA would have been entered. This potential misclassification may have lead to an underestimation of the effects of SA on outcomes. As mentioned previously, we were unable to determine whether the utilization of higher levels of care and noninvasive ventilation were the result of a complication and thus represented treatment or whether they were used in a prophylactic manner. The inability to determine causal relationships makes it impossible to study whether these interventions are capable of modifying outcomes in our sample. Thus, the value of these data lies in the estimation of the magnitude of utilization of these resources. Furthermore, because cause and effect or mechanisms of adverse events cannot be established from these data, we are unable to conclusively explain some of the findings. It is also likely that postoperative complications impact on the outcomes of mortality and resource utilization. However, the goal of this analysis was to study the impact of factors that are known preoperatively and may be considered before surgery commences. Finally, all comorbidities and complications are based on the ICD-9-CM coding system or billing codes (Appendix 1). Although rigorous quality checks are being performed by the vendor before release, coding errors or inconsistencies remain a possibility. A problem encountered for some outcomes, such as thromboembolic events for example, is the fact that there is no differential coding for an old versus new diagnosis, and therefore, we cannot conclusively determine whether such diagnoses were preexisting. Unfortunately, a present-on-admission variable, as introduced by many databases to facilitate this kind of interpretation, is not available for >70% of entries within our dataset, making it highly unreliable. However, there is no indication that this potential bias would affect one of the groups more than the other.
In conclusion, the preexisting comorbidity is associated with higher ORs of perioperative complications (adjusted OR: 1.47; CI, 1.39–1.55), utilization of economic resources (adjusted OR: 1.14; CI, 1.09–1.19) and prolonged length of stay (adjusted OR 1.12 for nonobese SA patients; CI, 1.06–1.18) among THA and TKA recipients. Despite a higher rate of advanced monitoring among SA patients, the overall utilization of stepdown, telemetry, or intensive care units was still <17%, at least partially putting into question the adoption of guidelines and perioperative protocols for the treatment of SA. The subject of outcomes among SA patients requires further study to identify patients at risk and determine ways to prevent complications using evidence-based and accountable approaches.
Name: Stavros G. Memtsoudis, MD, PhD, FCCP.
Contribution: This author helped design and conduct the study, analyze the data, and write the manuscript.
Attestation: Stavros G. Memtsoudis has seen the original study data, reviewed the analysis of the data, approved the final manuscript, and is the author responsible for archiving the study files.
Name: Ottokar Stundner, MD.
Contribution: This author helped design and conduct the study and write the manuscript.
Attestation: Ottokar Stundner has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
Name: Rehana Rasul, MA, MPH.
Contribution: This author helped analyze the data and write the manuscript.
Attestation: Rehana Rasul has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
Name: Ya-Lin Chiu, MS.
Contribution: This author helped conduct the study, analyze the data, and write the manuscript.
Attestation: Ya-Lin Chiu has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
Name: Xuming Sun, MS.
Contribution: This author helped design and conduct the study and analyze the data.
Attestation: Xuming Sun has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
Name: Satya-Krishna Ramachandran, MD.
Contribution: This author helped design the study and write the manuscript.
Attestation: Satya-Krishna Ramachandran has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
Name: Roop Kaw, MD.
Contribution: This author helped design the study and write the manuscript.
Attestation: Roop Kaw has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
Name: Peter Fleischut, MD.
Contribution: This author helped analyze the data and write the manuscript.
Attestation: Peter Fleischut has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
Name: Madhu Mazumdar, MS, MA, PhD.
Contribution: This author helped design the study, analyze the data, and write the manuscript.
Attestation: Madhu Mazumdar has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
This manuscript was handled by: Franklin Dexter, MD, PhD.
1. Finkel KJ, Searleman AC, Tymkew H, Tanaka CY, Saager L, Safer-Zadeh E, Bottros M, Selvidge JA, Jacobsohn E, Pulley D, Duntley S, Becker C, Avidan MS. Prevalence of undiagnosed obstructive sleep apnea among adult surgical patients in an academic medical center. Sleep Med. 2009;10:753–8
2. Memtsoudis S, Liu SS, Ma Y, Chiu YL, Walz JM, Gaber-Baylis LK, Mazumdar M. Perioperative pulmonary outcomes in patients with sleep apnea after noncardiac surgery. Anesth Analg. 2011;112:113–21
3. Gupta RM, Parvizi J, Hanssen AD, Gay PC. Postoperative complications in patients with obstructive sleep apnea syndrome undergoing hip or knee replacement: a case-control study. Mayo Clin Proc. 2001;76:897–905
4. Liao P, Yegneswaran B, Vairavanathan S, Zilberman P, Chung F. Postoperative complications in patients with obstructive sleep apnea: a retrospective matched cohort study. Can J Anesth. 2009;56:819–28
5. Chung F, Yegneswaran B, Liao P, Chung SA, Vairavanathan S, Islam S, Khajehdehi A, Shapiro CM. Validation of the Berlin questionnaire and American Society of Anesthesiologists checklist as screening tools for obstructive sleep apnea in surgical patients. Anesthesiology. 2008;108:822–30
6. Gali B, Whalen FX, Schroeder DR, Gay PC, Plevak DJ. Identification of patients at risk for postoperative respiratory complications using a preoperative obstructive sleep apnea screening tool and postanesthesia care assessment. Anesthesiology. 2009;110:869–77
7. Kaw R, Pasupuleti V, Walker E, Ramaswamy A, Foldvary-Schafer N. Postoperative complications in patients with obstructive sleep apnea. Chest. 2012;141:436–41
8. Kurtz S, Ong K, Lau E, Mowat F, Halpern M. Projections of primary and revision hip and knee arthroplasty in the United States from 2005 to 2030. J Bone Joint Surg Am. 2007;89:780–5
10. US Department of Health and Human Services. OCR Privacy Brief: Summary of the HIPAA Privacy Rule. (Office for Civil Rights HCA, ed.). 2003 Washington, DC Office for Civil Rights, HIPAA Compliance Assistance
11. Stundner O, Chiu YL, Sun X, Mazumdar M, Fleischut P, Poultsides L, Gerner P, Fritsch G, Memtsoudis SG. Comparative perioperative outcomes associated with neuraxial versus general anesthesia for simultaneous bilateral total knee arthroplasty. Reg Anesth Pain Med. 2012;37:638–44
12. Memtsoudis SG, Sun X, Chiu YL, Nurok M, Stundner O, Pastores SM, Mazumdar M. Utilization of critical care services among patients undergoing total hip and knee arthroplasty: epidemiology and risk factors. Anesthesiology. 2012;117:107–16
13. Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol. 1992;45:613–9
14. Austin PC. Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples. Stat Med. 2009;28:3083–107
15. Normand ST, Landrum MB, Guadagnoli E, Ayanian JZ, Ryan TJ, Cleary PD, McNeil BJ. Validating recommendations for coronary angiography following acute myocardial infarction in the elderly: a matched analysis using propensity scores. J Clin Epidemiol. 2001;54:387–98
16. Mickey RM, Greenland S. The impact of confounder selection criteria on effect estimation. Am J Epidemiol. 1989;129:125–37
17. Bland JM, Altman DG. Multiple significance tests: the Bonferroni method. BMJ. 1995;310:170
18. Hosmer DW, Lemeshow SA. Goodness-of-fit test for the multiple logistic regression model. Commun Stat. 1980;A10:1043–69
19. Pepe MS The Statistical Evaluation of Medical Tests for Classification and Precision. 2003 Oxford, UK Oxford University Press:66–94
20. Merkow RP, Hall BL, Cohen ME, Dimick JB, Wang E, Chow WB, Ko CY, Bilimoria KY. Relevance of the c-statistic when evaluating risk-adjustment models in surgery. J Am Coll Surg. 2012;214:822–30
21. Ury HK. Efficiency of case-control studies with multiple controls per case: continuous or dichotomous data. Biometrics. 1975;31:643–9
22. Lori S. Performing a 1:N Case-Control Match on Propensity Score. SAS Proceedings. 2010
23. Carpagnano GE, Lacedonia D, Foschino-Barbaro MP. Non-invasive study of airways inflammation in sleep apnea patients. Sleep Med Rev. 2011;15:317–26
24. Sabaté JM, Jouët P, Merrouche M, Pouzoulet J, Maillard D, Harnois F, Msika S, Coffin B. Gastroesophageal reflux in patients with morbid obesity: a role of obstructive sleep apnea syndrome? Obes Surg. 2008;18:1479–84
25. Blake DW, Chia PH, Donnan G, Williams DL. Preoperative assessment for obstructive sleep apnoea and the prediction of postoperative respiratory obstruction and hypoxaemia. Anesth Intensive Care. 2008;36:379–84
26. Yaggi HK, Concato J, Kernan WN, Lichtman JH, Brass LM, Mohsenin V. Obstructive sleep apnea as a risk factor for stroke and death. N Engl J Med. 2005;353:2034–41
27. Choi JB, Loredo JS, Norman D, Mills PJ, Ancoli-Israel S, Ziegler MG, Dimsdale JE. Does obstructive sleep apnea increase hematocrit? Sleep Breath. 2006;10:155–60
28. Lockhart EM, Willingham MD, Abdallah AB, Helsten DL, Bedair BA, Thomas J, Duntley S, Avidan MS. Obstructive sleep apnea screening and postoperative mortality in a large surgical cohort. Sleep Med. 2013;14:407–15
29. Mokhlesi B, Hovda MD, Vekhter B, Arora VM, Chung F, Meltzer DO. Sleep-disordered breathing and postoperative outcomes after elective surgery: analysis of the nationwide inpatient sample. Chest. 2013;144:903–14
30. Gross JB, Bachenberg KL, Benumof JL, Caplan RA, Connis RT, Coté CJ, Nickinovich DG, Prachand V, Ward DS, Weaver EM, Ydens L, Yu SAmerican Society of Anesthesiologists Task Force on Perioperative Management. . Practice guidelines for the perioperative management of patients with obstructive sleep apnea: a report by the American Society of Anesthesiologists Task Force on Perioperative Management of patients with obstructive sleep apnea. Anesthesiology. 2006;104:1081–93; quiz 1117–8
31. Turner K, VanDenkerkhof E, Lam M, Mackillop W. Perioperative care of patients with obstructive sleep apnea-a survey of Canadian anesthesiologists. Can J Anesth. 2006;53:299–304