Association of Unrecognized Obstructive Sleep Apnea With Postoperative Cardiovascular Events in Patients Undergoing Major Noncardiac Surgery

Key Points

Question  What is the relationship between unrecognized obstructive sleep apnea (OSA) and 30-day cardiovascular complications after major noncardiac surgery?

Findings  In this prospective cohort study that included 1218 at-risk patients undergoing major noncardiac surgery, the rate of a composite outcome of postoperative cardiovascular events (myocardial injury, cardiac death, congestive heart failure, thromboembolism, atrial fibrillation, and stroke) among those with OSA vs no OSA was 21.7% vs 14.2%, a difference that was statistically significant. However, the difference was significant only for the subgroup with severe OSA.

Meaning  Among patients undergoing major noncardiac surgery, severe OSA was significantly associated with 30-day cardiovascular complications.

Importance  Unrecognized obstructive sleep apnea increases cardiovascular risks in the general population, but whether obstructive sleep apnea poses a similar risk in the perioperative period remains uncertain.

Objectives  To determine the association between obstructive sleep apnea and 30-day risk of cardiovascular complications after major noncardiac surgery.

Design, Setting, and Participants  Prospective cohort study involving adult at-risk patients without prior diagnosis of sleep apnea and undergoing major noncardiac surgery from 8 hospitals in 5 countries between January 2012 and July 2017, with follow-up until August 2017. Postoperative monitoring included nocturnal pulse oximetry and measurement of cardiac troponin concentrations.

Exposures  Obstructive sleep apnea was classified as mild (respiratory event index [REI] 5-14.9 events/h), moderate (REI 15-30), and severe (REI >30), based on preoperative portable sleep monitoring.

Main Outcomes and Measures  The primary outcome was a composite of myocardial injury, cardiac death, heart failure, thromboembolism, atrial fibrillation, and stroke within 30 days of surgery. Proportional-hazards analysis was used to determine the association between obstructive sleep apnea and postoperative cardiovascular complications.

Results  Among a total of 1364 patients recruited for the study, 1218 patients (mean age, 67 [SD, 9] years; 40.2% women) were included in the analyses. At 30 days after surgery, rates of the primary outcome were 30.1% (41/136) for patients with severe OSA, 22.1% (52/235) for patients with moderate OSA, 19.0% (86/452) for patients with mild OSA, and 14.2% (56/395) for patients with no OSA. OSA was associated with higher risk for the primary outcome (adjusted hazard ratio [HR], 1.49 [95% CI, 1.19-2.01]; P = .01); however, the association was significant only among patients with severe OSA (adjusted HR, 2.23 [95% CI, 1.49-3.34]; P = .001) and not among those with moderate OSA (adjusted HR, 1.47 [95% CI, 0.98-2.09]; P = .07) or mild OSA (adjusted HR, 1.36 [95% CI, 0.97-1.91]; P = .08) (P = .01 for interaction). The mean cumulative duration of oxyhemoglobin desaturation less than 80% during the first 3 postoperative nights in patients with cardiovascular complications (23.1 [95% CI, 15.5-27.7] minutes) was longer than in those without (10.2 [95% CI, 7.8-10.9] minutes) (P < .001). No significant interaction effects on perioperative outcomes were observed with type of anesthesia, use of postoperative opioids, and supplemental oxygen therapy.

Conclusions and Relevance  Among at-risk adults undergoing major noncardiac surgery, unrecognized severe obstructive sleep apnea was significantly associated with increased risk of 30-day postoperative cardiovascular complications. Further research would be needed to assess whether interventions can modify this risk.

Quiz Ref IDObstructive sleep apnea (OSA) is the most common type of sleep-disordered breathing and is characterized by cyclical alterations between pharyngeal collapse and arousals during sleep.¹ Consequently, there are recurrent episodes of nocturnal hypoxemia, hypercapnia, endothelial dysfunction, hypercoagulability, and sympathetic overactivity.² In the general population, OSA is associated with higher risk of cardiovascular complications³ such as hypertension,⁴ myocardial ischemia, heart failure,⁵ arrhythmias, stroke,⁶ and sudden cardiac death.⁷

Quiz Ref IDGeneral anesthetics, sedatives, and postoperative analgesics are potent respiratory depressants that relax the upper airway dilator muscles and impair ventilatory response to hypoxemia and hypercapnia.¹ Each of these events exacerbates OSA and may predispose patients to postoperative cardiovascular complications. In this respect, perioperative mismanagement of OSA has led to serious medicolegal consequences.⁸ However, recent analyses of large database repositories showed conflicting results. Depending on the selected end points, OSA was associated with worse,^9-14 equivocal,^10,12,15 or better outcome^10,11 after surgery. Uncertainty remains whether unrecognized OSA adversely affects postoperative outcomes.

Based on preoperative overnight sleep studies, the Postoperative Vascular Complications in Unrecognized OSA (POSA) study was designed to determine the association between OSA and a composite of cardiac death, myocardial injury, heart failure, thromboembolism, atrial fibrillation, and stroke within 30 days of noncardiac surgery.

This was a multicenter, prospective cohort study of patients undergoing major noncardiac surgery. Ethics approval was obtained for all participating centers, and all patients provided written informed consent. We reported the trial objectives, design, and methods previously.¹⁶

We recruited patients who were 45 years or older and undergoing major noncardiac surgery (intraperitoneal, major orthopedic, or vascular). Patients were eligible for the study if they had 1 or more risk factors for postoperative cardiovascular events (ie, history of coronary artery disease, heart failure, stroke or transient ischemic attack, diabetes requiring treatment, and renal impairment with preoperative plasma creatinine concentration >1.98 mg/dL [175 μmol/L]). We excluded patients with prior diagnosis of obstructive sleep apnea or undergoing corrective surgery for OSA (eg, tonsillectomy, uvulopalatopharyngoplasty, tracheostomy), or anticipated to require prolonged (>2 days) mechanical lung ventilation after surgery.

Quiz Ref IDPatients underwent a preoperative overnight sleep study using a type 3 portable sleep monitoring device (ApneaLink Plus; ResMed).¹⁷ Sleep studies were performed either at home within the preceding month (34.1%) or in the surgical ward on the night before surgery (65.9%). In addition, we used a high-resolution pulse oximeter wristwatch (PULSOX-300i; Konica Minolta Sensing Inc) to monitor oxyhemoglobin saturation. Monitors were applied to the patients by experienced research staff at bedtime and were collected the following morning. Recordings were transferred to the coordinating center for subsequent analysis. We scored the sleep-associated apnea and hypopnea events according to American Academy of Sleep Medicine criteria (eAppendix 1 in the Supplement).¹⁸ Respiratory event index (REI) was calculated as the number of these events per hour of recording. Mild OSA was diagnosed when REI was 5 to 14.9, moderate OSA when REI was 15 to 30, and severe OSA when REI was greater than 30.^18,19 Based on the pulse oximetry signals obtained from the wristwatch, we also calculated the oxygen desaturation index (ODI) as the number of events (duration >10 seconds) per hour when there was a decrease in oxyhemoglobin saturation of 4% or more from baseline.²⁰

Before surgery, research staff interviewed all patients to record their baseline characteristics and risk factors for postoperative cardiovascular complications. Patients also indicated their race and ethnicity from a list of fixed categories, so that differences in OSA by race or ethnicity could be determined. In addition, we assessed patients’ risk for OSA using the STOP-Bang (Snoring, Tiredness, Observed Apnea, High Blood Pressure, Body Mass Index, Age, Neck Circumference, and Gender) screening tool (scores range from 0-8, with a score of 0-2 indicating low risk; 3-4, moderate risk; and 5-8, high risk).²¹

Patients, the attending surgical team, and research staff who collected the outcome data were blinded to the results of the sleep study, STOP-Bang questionnaire scores, and oximetry recordings until 30 days after surgery. After this time, we referred patients with abnormal sleep study findings to their local sleep clinic for further management of care.

All types of anesthetic techniques were permitted, and surgery was performed according to routine standard of care at each site. After surgery, electrocardiograms and venous blood samples (for measuring plasma cardiac troponin concentrations) were collected at 6 to 12 hours and then daily during the first 3 days after surgery. Additional echocardiograms and lung scans were performed, if clinically indicated, to ascertain the diagnosis of cardiac complications. During the first 3 postoperative nights, we recorded oxyhemoglobin saturation using the PULSOX-300i device. All patients were followed up regularly up to 30 days after surgery. Patients discharged home were contacted by telephone. The interview was conducted in a structured fashion. If patients or their relatives indicated that an event had occurred, we contacted the attending physicians or hospitals to obtain documentation.

The primary outcome was a composite of myocardial injury, cardiac death, congestive heart failure, thromboembolism, new atrial fibrillation, and stroke within 30 days of surgery. The prespecified secondary outcomes were unplanned tracheal intubation or postoperative lung ventilation, readmission to the intensive care unit (ICU), and infections. Details regarding the outcome definitions are listed in eAppendix 2 in the Supplement. All outcome events were evaluated by adjudicators blinded to the results of the sleep study.

We estimated that a sample size of 1200 patients was required to ensure a stable regression model with an anticipated primary event rate of 4%.¹⁶ Crude comparisons among patients with varying severity of OSA was performed using analysis of variance, Kruskal-Wallis test, or χ² test, as appropriate. We used Cox proportional-hazards models to determine the association between outcome events and OSA, except for unplanned tracheal intubation or postoperative lung ventilation and readmission to the ICU, for which we used logistic-regression analysis. The independent variables consisted of severity of OSA (severe, moderate, mild, or no disease) and factors previously shown to adversely affect outcomes.^22,23 These included age, history of coronary artery disease, congestive heart failure, stroke or transient ischemic attack, diabetes mellitus, chronic renal impairment, peripheral vascular disease, chronic obstructive pulmonary disease, abdominal or vascular surgery, and surgery for cancer. In addition, we included baseline variables that were unbalanced in patients with different severity of OSA—ethnicity, history of hypertension, and preoperative use of β-blockers. The proportionality assumption was evaluated by Schoenfeld residuals test. Collinearity was assessed by variance inflation factor, with a cutoff threshold of 10.²⁴ We also undertook a random-effects (frailty) Cox model to account for possible site-clustering effect.²⁵ The adjusted hazard ratios (HRs) for different severity of OSA were compared among groups using χ² test.

We used a general linear model to determine the association between nocturnal hypoxia and the primary outcome. In this model, severity of postoperative hypoxia was expressed as ODI. Other covariates included in the model were risk factors for the primary outcome as described above. We repeated the analysis to determine the association between the primary outcome and other measures of nocturnal hypoxia, including the lowest oxyhemoglobin saturation and the duration of oxyhemoglobin desaturation less than 80% and less than 90% recorded.

We also analyzed the primary outcome in prespecified subgroups of patients with the following characteristics: general or regional anesthesia, volatile-based anesthesia or propofol infusion, the number of postoperative nights with supplemental oxygen therapy, receiving (or not receiving) opioids after surgery, using (or not using) patient-controlled analgesia, and whether the surgery was considered minimally invasive. Subgroup analyses were performed using Cox models with the addition of corresponding interaction terms.

To better understand the basis of adverse outcomes, we performed post hoc analyses by repeating the Cox models using individual components of the primary outcome as the dependent variable. In addition, we conducted a sensitivity analysis to determine the validity of myocardial injury as an outcome measure. In this analysis, myocardial injury in the composite primary outcome was replaced by myocardial infarction according to the universal definition.²⁶ A post hoc comparison of the length of hospital stay in patients with different severity of OSA was also performed using log-rank test. In-hospital deaths were assigned with the longest length of stay.

For the association between outcomes and preoperative risk assessment for OSA based on the STOP-Bang screening tool, we repeated the primary analysis by stratifying patients as low-, intermediate-, and high-risk. We planned to conduct multiple imputations if there were more than 5% missing data on the outcomes or baseline variables included in the regression models. There was no adjustment for multiple comparisons; therefore, the results of the secondary analyses, subgroup analyses, and other analyses should be interpreted as exploratory.

All tests were 2-sided, and P<.05 was designated as statistically significant. Analyses were performed using Stata Release 13 (StataCorp) and R version 3.5.2 (R Project for Statistical Computing).

A total of 1364 patients were recruited from 8 hospitals in 5 countries between January 2012 and July 2017. We excluded 78 patients because sleep recordings (<4 hours) were unsatisfactory for analysis. Another 68 patients were excluded because surgery was canceled and could not be rescheduled within the subsequent month. Overall, 1218 patients who completed a preoperative sleep study and had undergone major noncardiac surgery were included in the current analyses (Figure 1). Among these patients, 67.6% had unrecognized OSA (REI ≥5), 30.5% had at least moderate OSA (REI ≥15), and 11.2% had severe OSA (REI >30). Details of preoperative sleep studies are reported in eTable 1 in the Supplement. All patients completed 30 days follow-up; no imputation of data was performed.

Table 1 summarizes patient characteristics, type of surgery, preoperative medications, and results of preoperative sleep studies. A total of 59.8% patients had at least 2 risk factors for cardiac disease. The most commonly performed surgical procedures were intraperitoneal (35.0%) or major orthopedic (29.9%), 42.1% of procedures were performed for cancer, and 28.3% were performed using a minimally invasive approach. Patients with OSA had a higher mean age and higher mean body mass index, and 63.7% were men. These patients had a higher rate of hypertension (85.9%), and 37.3% were taking β-blockers before surgery. The details of anesthetic administration and use of postoperative analgesia are presented in eTable 2 and eTable 3, respectively, in the Supplement. Perioperative anesthetic management was not different between groups. At least 1 postoperative measurement of cardiac troponin concentration was obtained for 95.8% of patients (eTable 4 in the Supplement).

The primary outcome occurred in 235 patients (19.3%) within 30 days of surgery. Among these patients, 17 (1.4%) died of cardiac cause; 205 (16.8%) had myocardial injury; 21 (1.7%) had congestive heart failure; 30 (2.5%) had atrial fibrillation; 10 (0.8%) had thromboembolism; and 5 (0.4%) had stroke. In patients with myocardial injury, 67 (5.5%) had ischemic symptoms, changes in electrocardiogram or cardiac imaging, and fulfilled the diagnosis of myocardial infarction.²⁶ Age, renal impairment, peripheral vascular disease, and OSA were independent risk factors for postoperative cardiovascular events (eTable 5 in the Supplement). There was no collinearity between variables. The Cox models showed no interaction between severity of OSA and age (P = .06 for interaction), preexisting renal impairment (P = .07 for interaction), and history of peripheral vascular disease (P = .56 for interaction).

At 30 days after surgery, rates of the primary outcome were 30.1% (41/136) for patients with severe OSA, 22.1% (52/235) for patients with moderate OSA, 19.0% (86/452) for patients with mild OSA, and 14.2% (56/395) for patients with no OSA (Figure 2). Compared with the reference group (patients without OSA), OSA was associated with higher risk for the primary outcome (adjusted HR, 1.49 [95% CI, 1.19-2.01]; P = .01). However, the association was only significant among patients with severe OSA (adjusted HR, 2.23 [95% CI, 1.49-3.34]; P = .001) and not among those with moderate OSA (adjusted HR, 1.47 [95% CI, 0.98-2.09]; P = .07) or mild OSA (adjusted HR, 1.36 [95% CI, 0.97-1.91]; P = .08) (P = .01 for interaction). There was no evidence for nonproportionality of hazards (P = .22) or site clustering (eTable 6 in the Supplement).

In the post hoc analyses, severe OSA was also associated with cardiac death (adjusted HR, 13.66 [95% CI, 1.63-114.19]), myocardial injury (adjusted HR, 1.80 [95% CI, 1.17-2.77]), congestive heart failure (adjusted HR, 6.55 [95% CI, 1.71-25.06]), and atrial fibrillation (adjusted HR, 3.96 [95% CI, 1.24-12.60]) (Table 2). In a sensitivity analysis that replaced myocardial injury with myocardial infarction in the primary outcome, severe OSA remained independently associated with postoperative cardiovascular complications (eTable 7 and eFigure 1 in the Supplement). OSA was also associated with infective outcomes, unplanned tracheal intubation, or postoperative lung ventilation and readmission to the ICU (Table 2). The association between OSA and postoperative cardiovascular events was similar across all subgroups of patients (P > .14 for interaction) (Figure 3). The associations in subgroup analysis were unchanged with varying severity of OSA (eFigures 2-4 in the Supplement). The median length of hospital stay in all patients was 5 days (interquartile range, 4-8) and was similar between different severities of OSA (P = .08 by log-rank test) (eFigure 5 in the Supplement).

Based on the preoperative STOP-Bang risk score questionnaire, 317 patients (26.3%) were rated as at high risk for OSA, 648 (53.2%) at intermediate risk, and 253 (20.8%) at low risk (eTable 8 and eTable 9 in the Supplement). Being a high-risk patient was significantly associated with increased rate of primary outcome (adjusted HR, 1.68 [95% CI, 1.11-2.54]), myocardial injury, and ICU readmission (eTable 10 and eFigure 6 in the Supplement). Being an intermediate-risk patient was significantly associated with ICU readmission and wound infection.

A total of 1131 patients (92.9%) received nocturnal oximetry monitoring during the first night after surgery, 1076 (88.3%) during the second night, and 983 (80.7%) during the third night (eFigures 7-9 in the Supplement). In patients without OSA, there was a significant increase in ODI after surgery (P < .001 for general linear model). In contrast, ODI in patients with OSA was reduced during the first 2 nights and returned to baseline on the third night after surgery. These changes were associated with supplemental oxygen administration (P = .009 for general linear model) (eFigure 7 in the Supplement). There was no difference in ODI, lowest oxyhemoglobin saturation, and maximum heart rate in patients with and without postoperative cardiovascular events (eTable 12 in the Supplement). However, the mean cumulative duration of oxyhemoglobin desaturation less than 80% during the first 3 postoperative nights for patients with cardiovascular complications (23.1 [95% CI, 15.5-27.7] minutes) was longer than for patients with no cardiovascular complications (10.2 [95% CI, 7.8-10.9] minutes) (P < .001 for general linear model) (eTable 12 and eFigure 10 in the Supplement).

Quiz Ref IDIn this study of adults undergoing major noncardiac surgery, unrecognized severe obstructive sleep apnea was significantly associated with increased risk of 30-day postoperative vascular complications.

Kaw et al²⁷ conducted a meta-analysis of 9 cohort and case-control studies (n = 2615 patients) that evaluated the association between OSA and postoperative cardiovascular complications. They reported an increased risk with OSA (odds ratio, 2.07 [95% CI, 1.23-3.50]), but there were few events (event rate, 2.6%), and the studies used less stringent criteria to diagnose OSA and postoperative cardiovascular complications. More recently, the Society of Anesthesia and Sleep Medicine reported a systematic review of 61 studies, including analyses of large-scale national databases,^9-15 to examine the association of OSA with perioperative outcomes.²⁸ Although a large number of patients were included (N = 8 969 583), there were substantial variations in outcome definitions and duration of follow-up, and the studies reported inconsistent results. In particular, it is unclear whether patients in the control groups of the 61 studies had unrecognized OSA, and those who had a preoperative diagnosis of OSA may have received extra treatment to modify perioperative outcomes. This heterogeneity precluded quantitative analysis of data.

In this study, a representative sample of patients undergoing major noncardiac surgery was included. Standardized preoperative sleep monitoring was performed to diagnose OSA, and patients were stratified according to disease severity. All patients completed follow-up, and postoperative monitoring of troponin concentrations was used to detect myocardial injury.

This study demonstrated that severe OSA was associated with increased risk of postoperative cardiovascular events. Despite a substantial decrease in ODI with oxygen therapy in patients with OSA during the first 3 postoperative nights, supplemental oxygen did not modify the association between OSA and postoperative cardiovascular event. Given that these events were associated with longer duration of severe oxyhemoglobin desaturation (<80%), more aggressive interventions may be required. Currently, positive airway pressure and oral appliances have been shown to overcome the collapsed upper airway and to relieve severe desaturation in nonoperative settings.^29,30 However, high-level evidence demonstrating the effect of these measures on perioperative outcomes is lacking.^31,32 Further clinical trials are now required to test if additional monitoring or alternative interventions would reduce the risk.

Quiz Ref IDIn contrast to the current guideline recommendations,³³ regional analgesia or avoidance of postoperative opioids were not associated with better outcome. These data are consistent with a retrospective analysis of an administrative database of 30 294 patients with documented OSA undergoing hip or knee arthroplasties with neuraxial block, general anesthesia, or both.³⁴ The study showed no change in postoperative cardiac or respiratory complications with neuraxial or general anesthesia. However, blood transfusion, requirement for postoperative mechanical lung ventilation, and ICU admission were decreased with neuraxial block. In the current study, patients undergoing major noncardiac surgery were recruited, few received regional blocks, and the majority required larger doses of systemic opioids for postoperative analgesia. This may have limited the statistical power to detect important interactions between OSA, anesthetic techniques, and postoperative analgesia.

This study has several limitations. First, electroencephalograms were not recorded in the preoperative sleep studies. Thus, it was not possible to track whether patients were asleep during measurement, and this may have underestimated the severity of OSA. Second, perioperative management was not controlled, but there was no difference in the administration of anesthesia and analgesics in patients with varying degrees of OSA. Although the surgical team was blinded to the results of the preoperative sleep study, recognition of minor respiratory events, such as episodes of apnea and higher level of sedation in the postanesthetic care unit and surgical ward, may have influenced perioperative management. This may include reducing doses of opioids or prolonging supplemental oxygen therapy. It is unclear how these interventions may affect perioperative outcomes. Nevertheless, the event rates reported in this study would represent the expected perioperative outcomes associated with untreated OSA in contemporary anesthetic practice for major noncardiac surgery. Third, the results should not be extrapolated to ambulatory procedures or minor surgery, for which anesthetic and analgesic techniques may have a larger effect on perioperative adverse events. Fourth, 54.7% of patients in this study were Chinese. Although Chinese patients with OSA have a lower body mass index and distinct differences in craniofacial anatomy compared with white patients,^35,36 it remains unclear how these differences might influence outcomes.

Among at-risk adults undergoing major noncardiac surgery, unrecognized severe obstructive sleep apnea was significantly associated with increased risk of 30-day postoperative cardiovascular complications. Further research would be needed to assess whether interventions can modify this risk.

Corresponding Author: Matthew T. V. Chan, MBBS, PhD, 4/F Main Clinical Block and Trauma Centre, Department of Anaesthesia and Intensive Care, Chinese University of Hong Kong, Prince of Wales Hospital, 30-32 Ngan Shing St, Shatin, Hong Kong Special Administrative Region, China (mtvchan@cuhk.edu.hk).

Accepted for Publication: April 4, 2019.

Author Contributions: Drs Chan and Chung had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Drs Chan, Wang, Seet, and Chung contributed equally to this article.

Concept and design: Chan, Wang, Seet, Hui, Chung.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Chan.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Chan, Wu.

Obtained funding: Chan, Wang, Seet, Short, Chung.

Administrative, technical, or material support: Chan, Wang, Seet, Tam, Lai, Chew, Cheng, Lam, Short, Hui, Chung.

Supervision: Chan, Wang, Seet, Tam, Lai, Chew, Cheng, Lam, Short, Chung.

Conflict of Interest Disclosures: Dr Chung reported receiving grants from the Ontario Ministry of Health and Long-term Care, Acacia Pharma, and Medtronics and holding a patent pending for the STOP-Bang questionnaire. No other authors reported disclosures.

Funding/Support: The study was funded through grants from the Health and Medical Research Fund (09100351), Hong Kong, National Healthcare Group-Khoo Teck Puat Hospital, Small Innovative Grants (12019, 15201), University Health Network Foundation (Ontario, Canada), University of Malaya, High Impact Research Grant (UM.C/625/1/HIR/067), Malaysian Society of Anaesthesiologists K Inbasegaran Research Grant and Auckland Medical Research Foundation, New Zealand. ResMed has supplied the ApneaLink devices and PULSOX-300i oximeter wristwatch in all sites as an unrestricted loan. These were returned at the end of the study.

Role of the Funder/Sponsor: The study funders/sponsors had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Group Information: The Postoperative Vascular Complications in Unrecognized Obstructive Sleep Apnea (POSA) Study Investigators: Steering Committee: Frances Chung (chair); Matthew Chan, Chew-Yin Wang, Edwin Seet. Event Adjudication Committee: Gordon Choi (chair), David Hui, Tony Gin. Investigators: Canada:Scarborough Health Network—Central Campus: Frances Chung, MBBS, Stanley Tam, MD, Sohail Iqbal, BSc; Hong Kong:Prince of Wales Hospital: Matthew Chan, MBBS, PhD, Gordon Choi, MBBS, David Hui, MD, Tony Gin, MD, Matthew Tsang, BSc, Beaker Fung, BSc, Angela Miu, BSc, Alex Lee, MSc; Tuen Mun Hospital: Benny Cheng, MBBS, Carmen Lam, MBBS, Sharon Tsang, MBChB, PhD, Chuen Ho Cheung, MBChB, Hoi Lam Pang, MBBS; Malaysia:University of Malaya Medical Centre: Chew Yin Wang, MBChB, Hou Yee Lai, MBBS, Carolyn C.W. Yim, MBBS, Alvin S.B. Tan, MBBS, Ching Yen Chong, BA, Jason H. Kueh, BSc, Xue Lin Chan, MBBS; Hospital Kuala Lumpur: Eleanor F.F. Chew, MBBS, Su Yin Loo, MBBS, Simon M.T. Hui, MBBS; New Zealand:Middlemore Hospital: Joyce Tai, MBChB, Stuart Walker, MBBS, Sue Olliff, BSc; Auckland City Hospital: Ivan Bergman, MBBS, Nicola Broadbent, MBBS, Maartje Tulp, MBBS, Timothy Short, MD, Davina McAllister, BSc; Singapore:Khoo Teck Puat Hospital: Edwin Seet, MBBS, Pei Fen Teoh, MBBS, Audris Chia, BSc.

Additional Contributions: We thank Andrew Forbes, PhD (Monash University, Australia), for his statistical advice and Ying Liu, MPH, Thomas Lo, MSc, and Beaker Fung, MSocSc (Chinese University of Hong Kong), who helped with figures preparation. None of these individuals received compensation for their contributions.

Study Coordination: This study was coordinated by the Chinese University of Hong Kong, Hong Kong Special Administrative region, China.