You are here

  • Home
  • Archive
  • Volume 9, Issue 8
  • Obstructive sleep apnoea, positive airway pressure treatment and postoperative delirium: protocol for a retrospective observational study

Article Text

Article menu

Download PDFPDF

XML
Anaesthesia
Protocol
Obstructive sleep apnoea, positive airway pressure treatment and postoperative delirium: protocol for a retrospective observational study
  1. Christopher R King1,
  2. Krisztina E Escallier1,
  3. Yo-El S Ju2,
  4. Nan Lin3,4,
  5. Ben Julian Palanca1,
  6. Sherry Lynn McKinnon1,
  7. Michael Simon Avidan1
  1. 1Anesthesiology, Washington University in Saint Louis School of Medicine, Saint Louis, Missouri, US
  2. 2Neurology, Washington University in Saint Louis School of Medicine, Saint Louis, Missouri, US
  3. 3Mathematics, Washington University in Saint Louis, St. Louis, Missouri, USA
  4. 4Division of Biostatistics, Washington Univiersity in Saint Louis, St. Louis, Missouri, USA
  1. Correspondence to Dr Michael Simon Avidan; avidanm{at}wustl.edu

Abstract

Introduction Obstructive sleep apnoea (OSA) is common among older surgical patients, and delirium is a frequent and serious postoperative complication. Emerging evidence suggests that OSA increases the risk for postoperative delirium. We hypothesise that OSA is an independent risk factor for postoperative delirium, and that in patients with OSA, perioperative adherence to positive airway pressure (PAP) therapy decreases the incidence of postoperative delirium and its sequelae. The proposed retrospective cohort analysis study will use existing datasets to: (i) describe and compare the incidence of postoperative delirium in surgical patients based on OSA diagnosis and treatment with PAP; (ii) assess whether preoperatively untreated OSA is independently associated with postoperative delirium; and (iii) explore whether preoperatively untreated OSA is independently associated with worse postoperative quality of life (QoL). The findings of this study will inform on the potential utility and approach of an interventional trial aimed at preventing postoperative delirium in patients with diagnosed and undiagnosed OSA.

Methods and analysis Observational data from existing electronic databases will be used, including over 100 000 surgical patients and ~10 000 intensive care unit (ICU) admissions. We will obtain the incidence of postoperative delirium in adults admitted postoperatively to the ICU who underwent structured preoperative assessment, including OSA diagnosis and screening. We will use doubly robust propensity score methods to assess whether untreated OSA independently predicts postoperative delirium. Using similar methodology, we will assess if untreated OSA independently predicts worse postoperative QoL.

Ethics and dissemination This study has been approved by the Human Research Protection Office at Washington University School of Medicine. We will publish the results in a peer-reviewed venue. Because the data are secondary and high risk for reidentification, we will not publicly share the data. Data will be destroyed after 1 year of completion of active Institutional Review Board (IRB) approved projects.

  • obstructive sleep apnea
  • postoperative delirium
  • ehr data

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

https://doi.org/10.1136/bmjopen-2018-026649

Statistics from Altmetric.com

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

    Strengths and limitations of this study

    • Our granular database includes routine structured preoperative screening for obstructive sleep apnoea, processed laboratory results and verified comorbid diagnoses.

    • We have limited information on the severity of most comorbidities, creating the possibility for substantial residual confounding.

    • Our database includes near-universal and standardised nurse-driven delirium evaluations at multiple time points as well as clinician diagnoses.

    • Compared with prior studies, the large sample size will allow for more aggressive confounder adjustment using linked structured medical histories, intraoperative records and administrative data.

    • Selection bias and confounding by indication are important limitations, which we will address using advanced statistical methods.

    Introduction

    Delirium is described in the Diagnostic and Statistical Manual of Mental Disorders, 5th Edition as a disturbance in attention, awareness and cognition that develops over a short period of time and tends to fluctuate in severity over the course of a day.1 It is a common postoperative complication with important costs. The reported incidence of postoperative delirium in older adults ranges from 10% to 70%, depending on context.2 Patients with postoperative delirium require longer intensive care unit (ICU) stays,3 experience greater institutionalisation and death after discharge,4 and report decreased quality of life (QoL).5 As a result, postoperative delirium is associated with a substantial increase in healthcare costs.6 7 Delirium has been proposed as an indicator of quality of care in older adults,8 and will affect an increasing proportion of patients as the population ages.

    The current literature contains suggestive evidence that obstructive sleep apnoea (OSA) is a common9 10 and independent risk factor for postoperative delirium.11–15 In a small prospective study, Flink et al reported that OSA is an independent predictor of postoperative delirium in older adults undergoing total knee arthroplasty with an OR of 4.2.11 A prospective study of 92 patients undergoing cardiac surgery found that a preoperative apnoea hypopnea index of 19 or higher was associated with increased risk of postoperative delirium (OR, 6.4; 95% CI, 2.6 to 15.4).15 A large observational study found that patients with undiagnosed OSA had worse postoperative outcomes than those with diagnosed OSA.16 An exploratory 114-patient randomised trial of preoperative positive airway pressure (PAP) found no impact of the intervention on delirium, but did find that OSA severity predicted postoperative delirium.12 A retrospective study14 and case report13 also offer support for the relationship between OSA and postoperative delirium. Several plausible biological explanations for this relationship exist, including hypoxia, chronic inflammation and disruption of normal sleep architecture as mediators.17 18 However, the studies linking OSA and postoperative delirium have been small, and it is important to confirm or refute the association in a larger and more diverse sample.

    We have previously investigated perioperative risks conferred by OSA. In the Barnes-Jewish Apnea Prevalence in Every Admission Study,19 a cohort of 14 962 elective surgery patients, we found a 12.9% (n=1939) prevalence of previously diagnosed OSA. Depending on the screening instrument, roughly 10%–40% of patients without a diagnosis were identified as high risk for OSA.20 We validated a new diagnosis in about 80% of tested patients screening as high risk.21 Therefore, the true overall prevalence of OSA was about 20%–25%. Both a history of OSA and a positive OSA screen were associated with admission to the ICU postoperatively.19 Patients with known OSA, but not those screening high risk, had longer ICU stays. Patients screening high risk had significantly higher 1 year mortality than those with low risk scores.19 However, delirium was not routinely assessed at that time. Others have found that these patients are at increased risk of serious pulmonary,22 23 cardiac14 24 and neurological18 postoperative complications.

    The gold standard therapy for OSA, PAP, reduces hypoxic events, reduces markers of chronic inflammation, and improves sleep.25–27 American Society of Anesthesiologists (ASA) practice guidelines28 recommend the optimisation of PAP therapy prior to surgery. Unfortunately, adherence to prescribed PAP therapy is low. It is estimated that 30% of patients who have been prescribed PAP never initiate therapy,29 and many eventually discontinue therapy or have suboptimal adherence.30 At our preoperative assessment clinic, approximately 50% of surgical candidates with OSA report adherence with PAP therapy. Similarly, Guralnick et al found that only 33% of adult surgical patients with moderate or severe OSA used PAP for ≥4 hours per night.31

    Our proposed retrospective cohort study has two coprimary hypotheses: (i) the presence of OSA (diagnosed or suggested by high-risk screen) increases the incidence of postoperative delirium and (ii) adequate treatment of OSA with PAP therapy reduces the risk of postoperative delirium. Secondary hypotheses are (i) high-risk screenings for untreated OSA in the preoperative period are independently associated with increased risk for postoperative delirium and (ii) untreated OSA in the preoperative period is independently associated with decreased postoperative QoL.

    Methods

    Data sources and setting

    The cohort will include all adults admitted postoperatively to either our general surgical or cardiothoracic ICUs (SICU, CTICU) between August 2012 to August 2018 who have any postoperative delirium assessments and a pre-anaesthesia evaluation (where our primary exposure is reported). Data from electronic medical record databases at Barnes Jewish Hospital will be obtained and combined. This will include the preoperative anaesthesia assessment, preoperative laboratory values, the day-of-service inpatient record with home medications reconciliation, the intraoperative anaesthesia record, the inpatient record (providers’ notes, nursing assessments, laboratory values, vital signs, medication administration record) and administrative records. Although detailed socioeconomic data will not be available, we will use administrative data on insurer, race, ethnicity and link home addresses to census-level socioeconomic measures. For some of the patients, we will also use data from our ongoing SATISFY-SOS registry study, which tracks the intermediate term postoperative health and well-being of unselected surgical patients (NCT02032030).32

    Based on typical admissions rates to our SICU (~3200 patients per year) and CTICU (~1200 patients per year), we estimate conservatively that the final dataset will include >10 000 patients. SATISFY-SOS is a prospective registry study; we estimate that about 2500 patients will be available for this analysis, based on enrolment and survey completion rates.

    Main outcomes and exposures

    The main outcome will be the incidence of postoperative delirium. Several years ago, our institution implemented routine delirium assessment in our ICUs and trained all ICU nurses to administer the Confusion Assessment Method for the Intensive Care Unit (CAM-ICU).33 Patients in the SICU and CTICU are now assessed twice daily for delirium. Scoring on the Richmond Agitation and Sedation Scale (RASS) is also assessed regularly and recorded, typically at the same time as the CAM if it is being performed. Patients will be coded as delirious if they have any positive delirium assessment during their ICU stay. Each episode will be characterised as hyperactive (RASS >0) or hypoactive (RASS ≤0).34 Secondary exploratory analyses will examine for differences with delirium type. Although delirium occurs outside the ICU, at our institution it is assessed is a non-systematic fashion. To avoid selectively recorded data and ascertainment biases related to the decision to perform a CAM on the wards, we will only analyse ICU assessments. Note extraction for chart diagnoses is not possible with this dataset and billing diagnoses do not specify a chronicity.

    Previous OSA-related data from our preoperative assessment clinic (table 1 and figure 1) and published literature19 20 35 were used to generate the estimated numbers of patients in each category in figure 2. We routinely screen with the STOP-BANG (Snoring, Tiredness, Observed Apnea, High Blood Pressure, Body Mass Index>35 kg/m2, Age>50, Neck Circumference>40 cm, Male Gender) criteria to determine OSA risk.20 We shall implement recent modifications of the STOP-BANG instrument (eg, including Age, BMI and Neck Circumference as continuous rather than dichotomous variables) that have been shown to improve its predictive value and specificity.27 36 37 PAP adherence is patient reported and documented in the preoperative assessment. Patients will be categorised as ‘adherent’ if they report ‘routine PAP use’. We will investigate if patients with in-hospital PAP use are more similar to those with good adherence in terms of outcomes and covariates. Hours of PAP use in the ICU are recorded in the Electronic Health Record (EHR); however, this outcome is a mixture of treatment for obstruction and other causes of respiratory failure and is causally dependent on intraoperative factors and postoperative mental status, so we do not intend to use it as a covariate or outcome.

    View this table:
    Table 1

    Adherence with treatment for obstructive sleep apnoea

    Figure 1
    Figure 1

    This figure shows data from 14 962 patients at our preoperative assessment clinic who did not carry a prior obstructive sleep apnoea (OSA) diagnosis.19

    Figure 2
    Figure 2

    This figure shows a predicted breakdown of patients based on previous data from our preoperative assessment clinic. Approximately 1300 (13%)19 of the approximately 10 000 patients in the study cohort will carry a diagnosis of obstructive sleep apnoea (OSA), of whom about half (~650) will have reported non-adherence to home Pap therapy (group B). Of the remaining 8700 patients, based on the current Stop-Bang criteria, about 870 (≥5 out of 8 positive criteria) are very likely to have moderate or severe undiagnosed OSA (group C).20 Approximately 3480 patients (3 or 4 positive criteria) might have undiagnosed OSA (group D), and ~4350 patients (<3 positive criteria) are unlikely to have undiagnosed OSA (group E). OSA, obstructive sleep apnoea; STOP-BANG, Snoring, Tiredness, Observed Apnea, High Blood Pressure, Body Mass Index>35 kg/m2, Age>50, Neck Circumference>40 cm, Male Gender.

    SATISFY-SOS tracks intermediate-term postoperative outcomes; patients complete postoperative surveys (approximately 1 month and 1 year after surgery) that includes the Veterans Rand 12 Item Health Survey (VR-12), a validated measure of QoL. QoL will be a secondary outcome in this subset of patients; based on our prior work with SATISFY-SOS, 1500 to 2000 responses (versus ~1 00 000 extracted the EHR) will be available for analysis.38 We will not link to delirium or other assessments from independent studies conducted during this period at BJH (ENGAGES, PODCAST).

    Covariates

    Our models will include demographics (age, race, ethnicity, sex) as well as census-tract level economic variables. In prior work we identified several predictors of delirium: average volatile anaesthetic dose, units of blood products transfused intraoperatively and ASA physical status.39 EuroSCORE, a measure of severity of comorbidities, was also found to be predictive; however, it is only used for cardiac surgery, and we will substitute the Charlson comorbidity index.40 Other predictors will include preoperative use of sedating medications, alcohol and other intoxicants, surgery performed, baseline laboratory values (haemoglobin, creatinine, haemoglobin A1C, INR, bilirubin, albumin), baseline pain score, history of cognitive impairment, and preoperative psychiatric diagnoses. We will categorise procedures into a small number of ‘types’ and use existing calibrations between surgery code and mortality.41 Based on our prior data19 the most common surgical types with be orthopaedic (~20%), general (~10%), urologic (~10%), gyecologic (~10%), otolarygologic (~8%), cardiothoracic (~6%), and neurosurgical (~6%). Our prior data do not contain good estimates of the ICU admission rates for these specialties; however, we can anticipate a substantial enrichment of cardiothoracic surgeries (at least 25%) based on the total admission rate to the CT-ICU vs SICU and a substantial decrease in neurosurgical cases as many patients are excluded from CAM measurement. Several intraoperative and postoperative variables will also be used: duration of surgery, duration of cardiopulmonary bypass, total intraoperative vasopresor and inotrope (norepinephrine, epinephrine, dopamine, dobutamine, phenylephrine and vasopressin) doses, intraoperative urine output, intraoperative fluids transfused, duration of coma, mechanical ventilation, use of sedatives, opioids, hypnotics and organ dysfunction scores.42 SATISFY-SOS patients will additionally have multidimensional preoperative measures of anxiety, pain, functionality, stroke, visual impairment and cognition.43–45

    Primary analysis plan and bias reduction

    In our dataset there are no plausible sources of exogenous variation in OSA exposure or CPAP adherence to eliminate bias due to unmeasured confounders. For the primary analysis, we will use a propensity-score-based approach and semiparametric regression adjustment to reduce bias due to measured variables. We will create propensity scores for OSA diagnosis or high STOP-BANG using non-parametric regression. We will use the fitted propensity score and covariates in a flexible regression method based on an ensemble of decision trees (Bayesian Adaptive Regression Trees (BART)46); this two-stage approach has been shown to be valid and robust,47–49 accounting for the uncertainty in the mechanisms of exposure and allowing nonlinear effects, interaction terms and heterogeneity of treatment effects.50–53 As a sensitivity analysis we will compare the average treatment effect on the treated from our primary analysis with propensity score matching based estimates of the same with greedy 1:1 matching.54–56 Treatment effect estimates will be reported with 99% credible/CI. We will compare the above method to logistic regression with all variables entered linearly for the propensity and adjustment model. We will calculate a c-statistic as well as other overall fit statistics to assess the fit of this final model and will use the model to calculate odds ratios (with 99% CI) associated with each predictor. In the final regression model, statistical significance will be assumed for p value <0.01. Fitted rates in each group and the absolute risk difference (average treatment effect on linear scale) with credible interval will also be reported.

    Because some variables are plausibly on the causal pathway connecting OSA and CPAP adherence and postoperative delirium (eg, postoperative opioid and anxiolytic use could be less in those with untreated OSA because they have OSA, leading to less delirium) simply treating them as confounders would produce biassed estimates57 and we will initially exclude them and examine for mediation if the overall association is notable.

    Secondary analyses

    We will use a similar regression method to report variables associated with PAP adherence and in-hospital initiation of PAP. We will use a similar technique to estimate the effect of PAP on delirium given an OSA diagnosis. QoL outcomes will be handled with a similar regression model. We will also conduct exploratory analyses. For example, we will investigate possible mechanistic associations with delirium, if relevant data (eg, oxygen saturation data) are available. We will also investigate whether outcomes are different between those who carry a diagnosis of OSA and those who screen positive for OSA. We plan to explore stratifications according to OSA severity.

    Missing data and loss to follow up

    We expect that some data will be missing in the proposed study, especially as we plan to combine multiple data sources. Depending on the types, patterns and frequencies of missing variables, we will select accepted statistical approaches in order to minimise omission of patients from the analyses. Multiple imputation has been shown to be robust to the violation of normality assumptions and has produced appropriate results in similar contexts. We will conduct sensitivity analyses to evaluate the robustness of our results with and without imputation. There will be no imputation for the main risk factors of interest (OSA diagnosis or treatment) or for the primary outcome of the study (incident delirium).

    For our primary outcome, loss to follow-up will be a negligible problem as patients are rarely discharged while still at risk for new onset delirium. For the SATISFY-SOS cohort, efforts to minimise true loss to follow-up have been described elsewhere.32

    Power analysis

    Based on the estimated numbers in each group in figure 2, this study will be adequately powered (>80%) for the three most relevant comparisons (ie, delirium incidence in Group A vs Group B; Group A vs Groups B+C; and Groups A+B+C vs Group E). For example, for the comparison between the smallest groups (Group A vs Group B), with one sided alpha <0.05, there is >80% power to detect a 6% difference (from 26% observed in ENGAGES58 to 20%) in delirium incidence.59 We will not adjust the p values for multiple comparisons. However, when assessing variables for independent associations with delirium, we shall use a more stringent alpha value <0.01.

    Ethics and dissemination

    The conduct and reporting of this observational study will follow STROBE guidelines.60 Once the investigation has been completed, we intend to publish the results in a peer-reviewed publication. We also intend to present the results of this work at professional conferences for the anaesthesiology community. The nature of the dataset (high resolution clinical histories linked to administrative records) makes de-identification a serious risk, and we do not plan to publicly share the data. Encryption will be used for any web-based information transmitted. The data will be stored on private protected network storage. Access will be restricted to research team members in a role-specific manner. Individual patient identifiers will be destroyed after the linking process is complete. Because the data are purely secondary, no formal data sharing is planned unless investigators obtain a separate approval for its access with Washington University’s IRB. Primary outcomes will be prespecified, as will analytical techniques. Additional not prespecified analyses will be treated as hypothesis-generating.

    Patient and public involvement

    No explicit patient or public comment was sought in the design of the study. Patient-centred research has previously identified ICU delirium as a life-changing event with major consequences to quality of life; examples of patient experiences can be found at icudelirium.org. Because this is a retrospective database study, no attempt will be made to directly contact patients with the findings.

    Discussion

    This large observational study will clarify if there is an independent link between OSA and postoperative delirium in the ICU. It will also show if this hypothetical increased risk is mitigated by treatment with PAP. It is important in science to replicate previous findings,61–63 which in the case of this study is the reported association between OSA and postoperative delirium,11 12 although this time in a broader surgical population. Because of its large size, this study will be useful for comparison between and among groups based on other risk factors.

    This study will have important strengths compared with the existing literature, most notably the very large and granular database including routine structured preoperative screening for OSA, and postoperative delirium detection in the ICU setting. The sample size will allow for a more aggressive confounder adjustment compared with smaller studies. The population will be diverse in both comorbidities and surgery performed, allowing a more tailored identification of patients who benefit from PAP and greater generalisability. As with other large retrospective studies, purely statistical error will be small in magnitude. We have a relatively high quality assessment of medical confounders due to our experienced preoperative clinic and a well-implemented assessment of delirium reducing measurement error in key variables. We have largely prespecified our analysis, reducing the potential for ‘analyst degrees of freedom introducing spuriously high confidence after multiple comparisons. The statistical approach should provide a strong predictive model and reduce the degree of ‘overfitting’ compared with common techniques like stepwise selection.64–68

    There are important limitations to the approach we are taking in this observational study. Foremost is selection bias. Patients who seek and adhere to treatment are different in many difficult to observe ways from those who do not. For example, PAP diagnosis and adherence (conditional on severity) is likely associated with socioeconomic status, care of other chronic conditions and coping strategies. Presence of OSA or non-adherence to prescribed PAP could induce surgeons to not offer highly invasive procedure options (reducing surgical severity), cause patients to present later (increasing surgical severity) or cause patients with an otherwise lower burden of morbidity to be more aggressively admitted to the ICU where they are eligible for delirium assessments. OSA severity is likewise associated with both PAP diagnosis and adherence, making the net direction of confounding difficult to predict. Differing from selection bias, downstream indirect effects of OSA such as additional supplemental oxygen, higher usage of telemetry monitoring and avoidance of sedating drugs may be protective. Although our preoperative clinic assessments are routinely thorough medical histories, we will have limited information on the severity of most comorbidities, leaving residual confounding. Most comorbidities are reported simultaneously, meaning that we will not be able to distinguish between confounders and mediators; simply adjusting for them may increase or decrease bias. Our intraoperative measures suffer the same difficulty.

    The common problem of missing data can reduce the statistical power of a study and can produce biassed estimates and invalid conclusions if severe. There are measurement errors for both the primary exposure and outcome which will decrease the validity of the associations. These analyses rely on subjective patient reporting of OSA history and PAP adherence. The STOP-BANG screening while reasonably accurate, is imperfect and may create false positives. We will try to confirm the diagnosis of OSA in our study subjects with the data available to us. Unfortunately, objective measures of PAP adherence from the actual PAP devices will not be available. Because patients tend to overestimate their own adherence,69 70 we expect that using self-reported adherence will tend to underestimate its influence on postoperative delirium rather than suggest a falsely positive association. We will attempt to obtain information from the electronic health record on in-hospital use of home PAP devices, since this may signify home adherence with PAP therapy. Treatment with alternative modalities, such as mandibular advancement devices, is not being assessed. OSA severity may be a key parameter which will be unable to obtain; others have found that apnoea-hypopnoea indices greater than 1571 or 3072 associated with postoperative complications. We have undertaken substantial efforts to standardise assessment of delirium in our ICUs as described above; however, there is doubtless error due to busy nursing staff and subjective elements in the assessment. Because PAP and OSA symptoms could influence delirium assessment, these errors may be informative and create additional bias.

    The most rigorous way to answer whether treatment of OSA prevents postoperative delirium would be to conduct a prospective randomised, controlled trial of perioperative PAP in patients already diagnosed with OSA who are scheduled for elective surgery. However, given the established benefits of PAP in these patients, it would be unethical to randomise patients (especially those already prescribed PAP) to a non-treatment arm. Therefore, a large observational study is likely to be the most appropriate initial design for addressing this question.

    Evidence of an independent risk association between untreated OSA and postoperative delirium would strongly warrant further investigation. An important question for future prospective study would be whether efforts at diagnosing OSA in the immediate preoperative period could mitigate postoperative delirium and its sequelae. We believe that this would be feasible, since we have already demonstrated within our institution that it is practical to identify patients with probable undiagnosed OSA using simple, economical screening methods.19 This study will further identify patients likely to benefit from focused interventions.

    If we find that PAP non-adherence and untreated OSA are independent risk factors for postoperative delirium, this would inform two key priorities. First, it would reinforce the importance of promoting adherence to perioperative PAP therapy. Second, it would provide a strong impetus for conducting a randomised controlled trial in elective-surgery patients with undiagnosed OSA, which we could not ethically implement in patients who already carry a diagnosis of OSA. We hope to use the foundational work proposed in this observational study to guide the design of such a trial, with the goals of reducing postoperative delirium and improving associated outcomes for the large number of patients at risk due to OSA.

    References

    1. 1.↵
      1. Meagher DJ,
      2. Morandi A,
      3. Inouye SK, et al
      . Concordance between DSM-IV and DSM-5 criteria for delirium diagnosis in a pooled database of 768 prospectively evaluated patients using the delirium rating scale-revised-98. BMC Med2014;12:164.doi:10.1186/s12916-014-0164-8
    2. 2.↵
      1. Whitlock EL,
      2. Vannucci A,
      3. Avidan MS, et al
      . Postoperative delirium. Minerva Anestesiol2011;77:448–56.
      OpenUrlPubMedWeb of Science
    3. 3.↵
      1. Lat I,
      2. McMillian W,
      3. Taylor S, et al
      . The impact of delirium on clinical outcomes in mechanically ventilated surgical and trauma patients. Crit Care Med2009;37:1898–905.doi:10.1097/CCM.0b013e31819ffe38
      OpenUrlCrossRefPubMedWeb of Science
    4. 4.↵
      1. Witlox J,
      2. Eurelings LSM,
      3. de Jonghe JFM, et al
      . Delirium in elderly patients and the risk of postdischarge mortality, institutionalization, and dementia: a meta-analysis. JAMA2010;304:443–51.doi:10.1001/jama.2010.1013
      OpenUrlCrossRefPubMedWeb of Science
    5. 5.↵
      1. Koster S,
      2. Hensens AG,
      3. Schuurmans MJ, et al
      . Consequences of delirium after cardiac operations. Ann Thorac Surg2012;93:705–11.doi:10.1016/j.athoracsur.2011.07.006
      OpenUrlCrossRefPubMedWeb of Science
    6. 6.↵
      1. Milbrandt EB,
      2. Deppen S,
      3. Harrison PL, et al
      . Costs associated with delirium in mechanically ventilated patients. Crit Care Med2004;32:955–62.doi:10.1097/01.CCM.0000119429.16055.92
      OpenUrlCrossRefPubMedWeb of Science
    7. 7.↵
      1. Leslie DL,
      2. Marcantonio ER,
      3. Zhang Y, et al
      . One-Year health care costs associated with delirium in the elderly population. Arch Intern Med2008;168:27–32.doi:10.1001/archinternmed.2007.4
      OpenUrlCrossRefPubMedWeb of Science
    8. 8.↵
      1. Shekelle PG,
      2. MacLean CH,
      3. Morton SC, et al
      . Acove quality indicators. Ann Intern Med2001;135:653–67.doi:10.7326/0003-4819-135-8_part_2-200110161-00004
      OpenUrlCrossRefPubMedWeb of Science
    9. 9.↵
      1. Young T,
      2. Peppard PE,
      3. Gottlieb DJ
      . Epidemiology of obstructive sleep apnea: a population health perspective. Am J Respir Crit Care Med2002;165:1217–39.
      OpenUrlCrossRefPubMedWeb of Science
    10. 10.↵
      1. Peppard PE,
      2. Young T,
      3. Barnet JH, et al
      . Increased prevalence of sleep-disordered breathing in adults. Am J Epidemiol2013;177:1006–14.doi:10.1093/aje/kws342
      OpenUrlCrossRefPubMedWeb of Science
    11. 11.↵
      1. Flink BJ,
      2. Rivelli SK,
      3. Cox EA, et al
      . Obstructive sleep apnea and incidence of postoperative delirium after elective knee replacement in the nondemented elderly. Anesthesiology2012;116:788–96.doi:10.1097/ALN.0b013e31824b94fc
      OpenUrlCrossRefPubMedWeb of Science
    12. 12.↵
      1. Nadler JW,
      2. Evans JL,
      3. Fang E, et al
      . A randomised trial of peri-operative positive airway pressure for postoperative delirium in patients at risk for obstructive sleep apnoea after regional anaesthesia with sedation or general anaesthesia for joint arthroplasty. Anaesthesia2017;72:729–36.doi:10.1111/anae.13833
      OpenUrl
    13. 13.↵
      1. Lee JW
      . Recurrent delirium associated with obstructive sleep apnea. Gen Hosp Psychiatry1998;20:120–2.doi:10.1016/S0163-8343(98)00006-1
      OpenUrlCrossRefPubMed
    14. 14.↵
      1. Gupta RM,
      2. Parvizi J,
      3. Hanssen AD, et al
      . Postoperative complications in patients with obstructive sleep apnea syndrome undergoing hip or knee replacement: a case-control study. Mayo Clin Proc2001;76:897–905.doi:10.1016/S0025-6196(11)62108-3
      OpenUrlCrossRefPubMedWeb of Science
    15. 15.↵
      1. Roggenbach J,
      2. Klamann M,
      3. von Haken R, et al
      . Sleep-Disordered breathing is a risk factor for delirium after cardiac surgery: a prospective cohort study. Crit Care2014;18.doi:10.1186/s13054-014-0477-1
    16. 16.↵
      1. Fernandez-Bustamante A,
      2. Bartels K,
      3. Clavijo C, et al
      . Preoperatively screened obstructive sleep apnea is associated with worse postoperative outcomes than previously diagnosed obstructive sleep apnea. Anesth Analg2017;125:593–602.doi:10.1213/ANE.0000000000002241
      OpenUrl
    17. 17.↵
      1. Mirrakhimov AE,
      2. Brewbaker CL,
      3. Krystal AD, et al
      . Obstructive sleep apnea and delirium: exploring possible mechanisms. Sleep Breath2014;18:19–29.doi:10.1007/s11325-013-0846-z
      OpenUrl
    18. 18.↵
      1. Kaw R,
      2. Golish J,
      3. Ghamande S, et al
      . Incremental risk of obstructive sleep apnea on cardiac surgical outcomes. J Cardiovasc Surg2006;47:683–9.
      OpenUrlPubMed
    19. 19.↵
      1. Lockhart EM,
      2. Willingham MD,
      3. Abdallah AB, et al
      . Obstructive sleep apnea screening and postoperative mortality in a large surgical cohort. Sleep Med2013;14:407–15.doi:10.1016/j.sleep.2012.10.018
      OpenUrlCrossRefPubMed
    20. 20.↵
      1. Chung F,
      2. Subramanyam R,
      3. Liao P, et al
      . High Stop-Bang score indicates a high probability of obstructive sleep apnoea. Br J Anaesth2012;108:768–75.doi:10.1093/bja/aes022
      OpenUrlCrossRefPubMedWeb of Science
    21. 21.↵
      1. Finkel KJ,
      2. Searleman AC,
      3. Tymkew H, et al
      . Prevalence of undiagnosed obstructive sleep apnea among adult surgical patients in an academic medical center. Sleep Med2009;10:753–8.doi:10.1016/j.sleep.2008.08.007
      OpenUrlCrossRefPubMedWeb of Science
    22. 22.↵
      1. Liao P,
      2. Yegneswaran B,
      3. Vairavanathan S, et al
      . Postoperative complications in patients with obstructive sleep apnea: a retrospective matched cohort study. Can J Anesth/J Can Anesth2009;56:819–28.doi:10.1007/s12630-009-9190-y
      OpenUrl
    23. 23.↵
      1. Hwang D,
      2. Shakir N,
      3. Limann B, et al
      . Association of sleep-disordered breathing with postoperative complications. Chest2008;133:1128–34.doi:10.1378/chest.07-1488
      OpenUrlCrossRefPubMedWeb of Science
    24. 24.↵
      1. den Herder C,
      2. Schmeck J,
      3. Appelboom DJK, et al
      . Risks of general anaesthesia in people with obstructive sleep apnoea. BMJ2004;329:955–9.doi:10.1136/bmj.329.7472.955
      OpenUrlFREE Full Text
    25. 25.↵
      1. Garvey JF,
      2. Taylor CT,
      3. McNicholas WT
      . Cardiovascular disease in obstructive sleep apnoea syndrome: the role of intermittent hypoxia and inflammation. Eur Respir J2009;33:1195–205.doi:10.1183/09031936.00111208
      OpenUrlAbstract/FREE Full Text
    26. 26.↵
      1. Giles TL,
      2. Lasserson TJ,
      3. Smith BH, et al
      . Continuous positive airways pressure for obstructive sleep apnoea in adults. Cochrane Database Syst Rev2006;25:CD001106.
    27. 27.↵
      1. Chung F,
      2. Chau E,
      3. Yang Y, et al
      . Serum bicarbonate level improves specificity of Stop-Bang screening for obstructive sleep apnea. Chest2013;143:1284–93.doi:10.1378/chest.12-1132
      OpenUrl
    28. 28.↵
      1. American Society of Anesthesiologists Task Force on Perioperative Management of patients with obstructive sleep apnea
      . Practice guidelines for the perioperative management of patients with obstructive sleep apnea: an updated report by the American Society of Anesthesiologists Task force on perioperative management of patients with obstructive sleep apnea. Anesthesiology2014;120:268–86.doi:10.1097/ALN.0000000000000053
      OpenUrlCrossRefPubMedWeb of Science
    29. 29.↵
      1. Guest JF,
      2. Helter MT,
      3. Morga A, et al
      . Cost-Effectiveness of using continuous positive airway pressure in the treatment of severe obstructive sleep apnoea/hypopnoea syndrome in the UK. Thorax2008;63:860–5.doi:10.1136/thx.2007.086454
      OpenUrlAbstract/FREE Full Text
    30. 30.↵
      1. Weaver TE,
      2. Grunstein RR
      . Adherence to continuous positive airway pressure therapy: the challenge to effective treatment. Proc Am Thorac Soc2008;5:173–8.doi:10.1513/pats.200708-119MG
      OpenUrlCrossRefPubMed
    31. 31.↵
      1. Guralnick AS,
      2. Pant M,
      3. Minhaj M, et al
      . Cpap adherence in patients with newly diagnosed obstructive sleep apnea prior to elective surgery. J Clin Sleep Med2012;8:501–6.doi:10.5664/jcsm.2140
      OpenUrlPubMed
    32. 32.↵
      1. Helsten DL,
      2. Ben Abdallah A,
      3. Avidan MS, et al
      . Methodologic considerations for collecting patient-reported outcomes from unselected surgical patients. Anesthesiology2016;125:495–504.doi:10.1097/ALN.0000000000001217
      OpenUrl
    33. 33.↵
      1. Ely EW,
      2. Margolin R,
      3. Francis J, et al
      . Evaluation of delirium in critically ill patients: validation of the confusion assessment method for the intensive care unit (CAM-ICU). Crit Care Med2001;29:1370–9.doi:10.1097/00003246-200107000-00012
      OpenUrlCrossRefPubMedWeb of Science
    34. 34.↵
      1. Pandharipande P,
      2. Cotton BA,
      3. Shintani A, et al
      . Motoric subtypes of delirium in mechanically ventilated surgical and trauma intensive care unit patients. Intensive Care Med2007;33:1726–31.doi:10.1007/s00134-007-0687-y
      OpenUrlCrossRefPubMedWeb of Science
    35. 35.↵
      1. Singh M,
      2. Liao P,
      3. Kobah S, et al
      . Proportion of surgical patients with undiagnosed obstructive sleep apnoea. Br J Anaesth2013;110:629–36.doi:10.1093/bja/aes465
      OpenUrlCrossRefPubMedWeb of Science
    36. 36.↵
      1. Nahapetian R,
      2. Silva GE,
      3. Vana KD, et al
      . Weighted Stop-Bang and screening for sleep-disordered breathing. Sleep Breath2016;20.doi:10.1007/s11325-015-1255-2
    37. 37.↵
      1. Chung F,
      2. Yang Y,
      3. Brown R, et al
      . Alternative scoring models of Stop-Bang questionnaire improve specificity to detect undiagnosed obstructive sleep apnea. J Clin Sleep Med2014;10:951–8.doi:10.5664/jcsm.4022
      OpenUrlPubMed
    38. 38.↵
      1. Fritz BA,
      2. Escallier KE,
      3. Ben Abdallah A, et al
      . Convergent validity of three methods for measuring postoperative complications. Anesthesiology2016;124:1265–76.doi:10.1097/ALN.0000000000001108
      OpenUrl
    39. 39.↵
      1. Whitlock EL,
      2. Torres BA,
      3. Lin N, et al
      . Postoperative delirium in a substudy of cardiothoracic surgical patients in the BAG-RECALL clinical trial. Anesth Analg2014;118:809–17.doi:10.1213/ANE.0000000000000028
      OpenUrlCrossRefPubMed
    40. 40.↵
      1. Charlson ME,
      2. Pompei P,
      3. Ales KL, et al
      . A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis1987;40:373–83.doi:10.1016/0021-9681(87)90171-8
      OpenUrlCrossRefPubMedWeb of Science
    41. 41.↵
      1. Sessler DI,
      2. Sigl JC,
      3. Manberg PJ, et al
      . Broadly applicable risk stratification system for predicting duration of hospitalization and mortality. Anesthesiology2010;113:1026–37.doi:10.1097/ALN.0b013e3181f79a8d
      OpenUrlCrossRefPubMedWeb of Science
    42. 42.↵
      1. Gosselt AN,
      2. Slooter AJ,
      3. Boere PR, et al
      . Risk factors for delirium after on-pump cardiac surgery: a systematic review. Crit Care2015;19.doi:10.1186/s13054-015-1060-0
    43. 43.↵
      1. Brunault P,
      2. Frammery J,
      3. Couet C, et al
      . Predictors of changes in physical, psychosocial, sexual quality of life, and comfort with food after obesity surgery: a 12-month follow-up study. Qual Life Res2015;24:493–501.doi:10.1007/s11136-014-0775-8
      OpenUrl
    44. 44.↵
      1. Dunn WR,
      2. Wolf BR,
      3. Harrell FE, et al
      . Baseline predictors of health-related quality of life after anterior cruciate ligament reconstruction: a longitudinal analysis of a multicenter cohort at two and six years. J Bone Joint Surg Am2015;97:551–7.doi:10.2106/JBJS.N.00248
      OpenUrlAbstract/FREE Full Text
    45. 45.↵
      1. Vainiola T,
      2. Roine RP,
      3. Suojaranta-Ylinen R, et al
      . Can factors related to mortality be used to predict the follow-up health-related quality of life (HRQoL) in cardiac surgery patients?Intensive Crit Care Nurs2013;29:337–43.doi:10.1016/j.iccn.2013.04.003
      OpenUrl
    46. 46.↵
      1. Chipman HA,
      2. George EI,
      3. McCulloch RE
      . Bart: Bayesian additive regression trees. Ann Appl Stat2010;4:266–98.doi:10.1214/09-AOAS285
      OpenUrlCrossRefWeb of Science
    47. 47.↵
      1. McCandless LC,
      2. Douglas IJ,
      3. Evans SJ, et al
      . Cutting feedback in Bayesian regression adjustment for the propensity score. Int J Biostat2010;6:Article 16. Article 16.doi:10.2202/1557-4679.1205
    48. 48.↵
      1. Spertus JV,
      2. Normand S-LT
      . Bayesian propensity scores for high-dimensional causal inference: a comparison of drug-eluting to bare-metal coronary stents. Biometrical Journal2018;60:721–33.doi:10.1002/bimj.201700305
      OpenUrl
    49. 49.↵
      1. Vansteelandt S,
      2. Daniel RM
      . On regression adjustment for the propensity score. Stat Med2014;33:4053–72.doi:10.1002/sim.6207
      OpenUrlCrossRefPubMed
    50. 50.↵
      1. Zigler CM,
      2. Dominici F
      . Uncertainty in propensity score estimation: Bayesian methods for variable selection and Model-Averaged causal effects. J Am Stat Assoc2013;109.doi:10.1080/01621459.2013.869498
    51. 51.↵
      1. Hill JL
      . Bayesian nonparametric modeling for causal inference. J Comput Graph Stat2011;20:217–40.doi:10.1198/jcgs.2010.08162
      OpenUrl
    52. 52.↵
      1. Hahn PR,
      2. Murray JS,
      3. Carvalho CM
      . Bayesian regression tree models for causal inference: regularization, confounding, and heterogeneous effects. SSRN Journal2017.doi:10.2139/ssrn.3048177
    53. 53.↵
      1. Linero AR
      . Bayesian regression trees for high-dimensional prediction and variable selection. Journal of the American Statistical Association2016;113.
    54. 54.↵
      1. Austin PC
      . Double propensity-score adjustment: a solution to design bias or bias due to incomplete matching. Stat Methods Med Res2017;26:201–22.doi:10.1177/0962280214543508
      OpenUrl
    55. 55.↵
      1. Austin PC
      . A critical appraisal of propensity-score matching in the medical literature between 1996 and 2003. Stat Med2008;27:2037–49.doi:10.1002/sim.3150
      OpenUrlCrossRefPubMedWeb of Science
    56. 56.↵
      1. Zakrison TL,
      2. Austin PC,
      3. McCredie VA
      . A systematic review of propensity score methods in the acute care surgery literature: avoiding the pitfalls and proposing a set of reporting guidelines. Eur J Trauma Emerg Surg2018;44:385–95.doi:10.1007/s00068-017-0786-6
      OpenUrl
    57. 57.↵
      1. Schisterman EF,
      2. Cole SR,
      3. Platt RW
      . Overadjustment bias and unnecessary adjustment in epidemiologic studies. Epidemiology2009;20:488–95.doi:10.1097/EDE.0b013e3181a819a1
      OpenUrlCrossRefPubMedWeb of Science
    58. 58.↵
      1. Wildes TS,
      2. Mickle AM,
      3. Ben Abdallah A, et al
      . Effect of Electroencephalography-Guided anesthetic administration on postoperative delirium among older adults undergoing major surgery: the engages randomized clinical trial. JAMA2019;321:473–83.doi:10.1001/jama.2018.22005
      OpenUrl
    59. 59.↵
      1. Erdfelder E,
      2. Faul F,
      3. Buchner A
      . GPOWER: a general power analysis program. Behav Res Meth Ins C1996;28:1–11.doi:10.3758/BF03203630
      OpenUrlCrossRef
    60. 60.↵
      1. von Elm E,
      2. Altman DG,
      3. Egger M, et al
      . The strengthening the reporting of observational studies in epidemiology (STROBE) statement: guidelines for reporting observational studies. Epidemiology2007;18:800–4.doi:10.1097/EDE.0b013e3181577654
      OpenUrlCrossRefPubMedWeb of Science
    61. 61.↵
      1. Ioannidis JPA
      . Why most published research findings are false. PLoS Med2005;2:e124.doi:10.1371/journal.pmed.0020124
    62. 62.↵
      1. PSYCHOLOGY OSC
      . Estimating the reproducibility of psychological science. Science2015;349.
    63. 63.↵
      1. Yong E
      . Replication studies: bad copy. Nature2012;485:298–300.doi:10.1038/485298a
      OpenUrlCrossRefPubMedWeb of Science
    64. 64.↵
      1. Kwon D,
      2. Landi MT,
      3. Vannucci M, et al
      . An efficient stochastic search for Bayesian variable selection with high-dimensional correlated predictors. Comput Stat Data Anal2011;55:2807–18.doi:10.1016/j.csda.2011.04.019
      OpenUrl
    65. 65.↵
      1. Swartz MD,
      2. Thomas DC,
      3. Daw EW, et al
      . Model selection and Bayesian methods in statistical genetics: summary of group 11 contributions to genetic analysis workshop 15. Genet Epidemiol2007;31 Suppl 1:S96–S102.doi:10.1002/gepi.20285
      OpenUrl
    66. 66.↵
      1. Russu A,
      2. Malovini A,
      3. Puca AA, et al
      . Stochastic model search with binary outcomes for genome-wide association studies. J Am Med Inform Assoc2012;19:e13–20.doi:10.1136/amiajnl-2011-000741
      OpenUrlCrossRefPubMed
    67. 67.↵
      1. Greenland S
      . Bayesian perspectives for epidemiological research. II. regression analysis. Int J Epidemiol2007;36:195–202.doi:10.1093/ije/dyl289
      OpenUrlCrossRefPubMedWeb of Science
    68. 68.↵
      1. Babyak MA
      . What you see may not be what you get: a brief, nontechnical introduction to overfitting in regression-type models. Psychosom Med2004;66:411–21.
      OpenUrlAbstract/FREE Full Text
    69. 69.↵
      1. Kribbs NB,
      2. Pack AI,
      3. Kline LR, et al
      . Objective measurement of patterns of nasal CPAP use by patients with obstructive sleep apnea. Am Rev Respir Dis1993;147:887–95.doi:10.1164/ajrccm/147.4.887
      OpenUrlCrossRefPubMedWeb of Science
    70. 70.↵
      1. Reeves-Hoche MK,
      2. Meck R,
      3. Zwillich CW
      . Nasal CPAP: an objective evaluation of patient compliance. Am J Respir Crit Care Med1994;149:149–54.doi:10.1164/ajrccm.149.1.8111574
      OpenUrlCrossRefPubMedWeb of Science
    71. 71.↵
      1. Suen C,
      2. Ryan CM,
      3. Mubashir T, et al
      . Sleep study and oximetry parameters for predicting postoperative complications in patients with OSA. Chest2019;155:855–67.doi:10.1016/j.chest.2018.09.030
      OpenUrl
    72. 72.↵
      1. Chan MTV,
      2. Wang CY,
      3. Seet E, et al
      . Association of unrecognized obstructive sleep apnea with postoperative cardiovascular events in patients undergoing major noncardiac surgery. JAMA2019;321:1788–98.doi:10.1001/jama.2019.4783
      OpenUrl

    Footnotes

    • Contributors CRK contributed to the statistical methods initial draft of protocol and critical revisions of protocol KE contributed to the overall study design and critical revisions of the protocol, YSJ, NL, BJP, SLM MSA contributed to the study design and critical revision of protocol.

    • Funding statement This work is funded largely by a grant from National Heart, Lung, and Blood Institute (award number 1R21HL123666-01A1.

    • Competing interests None declared.

    • Patient consent for publication None declared.

    • Ethics approval The Human Research Protection Office at Washington University School of Medicine in St. Louis has approved this study (IRB #201311088).

    • Provenance and peer review Not commissioned; externally peer reviewed.

    • Data availability statement No additional data are available.