Predictors of postoperative respiratory complications in children undergoing adenotonsillectomy

Abstract

Study Objectives:

Obstructive sleep apnea (OSA) is commonly treated with adenotonsillectomy (AT), bringing risk of perioperative respiratory adverse events (PRAEs). We aimed to concurrently identify clinical and polysomnographic predictors of PRAEs in children undergoing AT.

Methods:

Retrospective study of children undergoing AT at a tertiary-care pediatric hospital, with prior in-hospital polysomnography, January 2010 to December 2016. PRAEs included those requiring oxygen, jaw thrust, positive airway pressure, or mechanical ventilation. Relationships of PRAEs to preoperative comorbidities or polysomnography results were examined with univariable logistic regression. Variables with P < .1 and age were included in backward stepwise multivariable logistic regression. Predictive performance (area under the curve, AUC) was validated with bootstrap resampling.

Results:

Analysis included 374 children, median age 6.1 years; 286 (76.5%) had ≥ 1 comorbidity. 344 (92.0%) had sleep-disordered breathing; 232 (62.0%) moderate-severe; 66 (17.6%) had ≥ 1 PRAE. PRAEs were more frequent in children with craniofacial, genetic, cardiac, airway anomaly, or neurological conditions, AHI ≥ 5 events/h and oxygen saturation nadir ≤ 80% on preoperative polysomnography. Prediction modeling identified cardiac comorbidity (odds ratio [OR] 2.09 [1.11, 3.89]), airway anomaly (OR 3.19 [1.33, 7.49]), and younger age (OR < 3 years: 4.10 (1.79, 9.26; 3 to 6 years: 2.21 [1.18, 4.15]) were associated with PRAEs (AUC 0.74; corrected AUC 0.68).

Conclusions:

Prediction modeling concurrently evaluating comorbidities and polysomnography metrics identified cardiac disease, airway anomaly, and young age as independent predictors of PRAEs. These findings suggest that medical comorbidity and age are more important factors in predicting PRAEs than PSG metrics in a medically complex population.

Citation:

Katz SL, Monsour A, Barrowman N, et al. Predictors of postoperative respiratory complications in children undergoing adenotonsillectomy. J Clin Sleep Med. 2020;16(1):41–48.

BRIEF SUMMARY

Current Knowledge/Study Rationale: Children with obstructive sleep apnea are often treated with adenotonsillectomy, which can be complicated by perioperative respiratory adverse events (PRAEs). PRAEs have not been predicted from comorbidity and polysomnography simultaneously. Our aim was to identify clinical and polysomnographic predictors of PRAEs in children undergoing adenotonsillectomy with prior polysomnography.

Study Impact: This is the largest cohort to date that simultaneously evaluated comorbidities and polysomnographic variables in the complete group, as predictors of PRAEs after adenotonsillectomy. Cardiac disease, airway anomaly, and young age were independent predictors of PRAEs. In this medically complex population, medical comorbidity and age aid in identifying children at risk of PRAEs, which is particularly important in resource-limited settings where PSG may not be readily available.

INTRODUCTION

Obstructive sleep apnea (OSA) occurs in 1% to 5% of children and results in complete or partial airway obstruction associated with desaturation, hypercapnia, or sleep fragmentation.¹ Left untreated, OSA can result in impaired neurocognitive function, behavior disorders, decreased quality of life, and cardiometabolic disease. Children and adolescents with OSA are commonly treated with adenotonsillectomy (AT). However, this surgical procedure has a low but recognized risk of perioperative respiratory adverse events (PRAEs), particularly during the postoperative period, ranging from oxygen desaturation to apnea requiring ventilator support.^1,2 Death following ambulatory AT in North America is rare. Reported all-cause 30-day mortality rates range between 1 in 1,500 (admitted patients only) to 1 in 55,000 (combined ambulatory and admitted patients)^3–6 and appear to be related to apneic and/or bleeding events.^7–9

Existing studies evaluating risk of postoperative complications have not simultaneously considered multiple specific comorbid medical conditions and polysomnography (PSG), the gold standard test for evaluating OSA, in complete cohorts.^10,11 Young age, asthma, cardiac disease, neurologic disease, genetic syndrome, prematurity, gastroesophageal reflux, craniofacial syndromes, failure to thrive/lower body weight, and obesity, especially when associated with other comorbidities, have been identified as independent predictors of post-AT PRAEs.^10,12–19 The presence of OSA has also been shown in a meta-analysis to lead to a fivefold increased risk of respiratory compromise.²⁰ Desaturations to below 80% and increasing apnea-hypopnea index (AHI) have also been independently associated with increased risk of severe PRAEs.¹⁰

Criteria incorporating age, comorbidities, and OSA have been integrated into clinical care guidelines to identify those at highest risk of PRAEs and provide recommendations with respect to disposition planning. However, the guidelines are varied in their conclusions regarding recommendations for postoperative admission. For example, admission thresholds based on AHI range from greater than 10 to 24 events/h.^1,21 The American Academy of Otolaryngology Head and Neck Surgery Foundation guidelines recommend postoperative hospital admission for children undergoing AT who are younger than 3 years or have severe OSA (AHI greater than 10 events/h).²² The American Academy of Pediatrics has similar admission recommendations, but suggest that an AHI ≥ 24 events/h and children with certain medical comorbidities or a current respiratory infection warrant overnight monitoring.¹ The American Society of Anesthesiologists guideline includes age, OSA status, and comorbidity admission criteria and in addition considers the nature and type of surgery performed, anesthesia administered, and the need for postoperative opioids.²³ Given the uncertainty regarding optimal perioperative care including pain management for children undergoing AT, determining appropriate postoperative disposition has been identified as a priority in the literature.^21,24 Further, clinical practices with respect to hospital admissions after AT are highly variable, even across tertiary care pediatric centers.²⁵

The primary objective of this study was to identify specific clinical and polysomnographic predictors of any severe PRAEs that required intervention (supplemental oxygen, jaw thrust, positive airway pressure, or mechanical ventilation) in a large cohort of children undergoing AT who all had prior laboratory PSG. Our data will further inform clinical planning regarding optimal postoperative monitoring environment for children undergoing AT.

METHODS

Study population

This retrospective study included all children, aged 0 to 18 years, who underwent PSG, followed by AT, at the Children’s Hospital of Eastern Ontario, between January 2010 and December 2016.

Study design and measurements

Ethics approval for this study was obtained from the Children’s Hospital of Eastern Ontario Research Ethics Board (14/113×). Data collection was managed using a Research Electronic Data Capture database (Vanderbilt University, Nashville, Tennessee, USA).

Medical charts were reviewed for demographic characteristics, comorbidities, polysomnogram results, postoperative complications, reason and location for postoperative admission. Demographic variables included sex and age at preoperative assessment. Preoperative comorbidity information was collected for the following conditions: craniofacial (eg, Apert syndrome), genetic (eg, Down syndrome), congenital heart disease (eg, septal or atrioventricular canal defects requiring repair), airway anomaly (eg, micrognathia, midfacial hypoplasia, vocal cord paresis, laryngotracheomalacia), nasopharyngeal symptoms (eg, nasal congestion or rhinitis), lower respiratory disease (eg, history of asthma or wheezing unspecified), neurologic condition (eg,cerebral palsy), gastrointestinal condition (eg, gastroesophageal reflux), endocrine disease (eg, type I diabetes mellitus), allergy (seasonal, drug or food)/atopy, history of prematurity (born before 37 weeks’ gestation) and obesity (body mass index above the 95th percentile for sex and age). Preoperative PSG variables included the total AHI (events/h) and the lowest oxygen saturation nadir (%).

Postoperative respiratory complications of interest were chosen among severe PRAEs that required intervention: oxygen desaturation requiring intervention, or the need for airway or ventilatory support (ie jaw thrust, oral/nasal airway placement, bag and mask ventilation or endotracheal intubation). PRAEs were included if they occurred at any time prior to first hospital discharge, regardless of the physical location of a child, which may have been in the postanesthetic care unit (PACU), medical daycare unit, hospital ward, or intensive care unit.

Statistical analysis

The outcome of the study was defined as the occurrence of any of the types of PRAEs (combined). The relationships of severe postoperative complications to preoperative comorbidities or to PSG results were examined separately with univariable logistic regression. Our goal was to derive a preliminary prediction model. Given the large number of types of comorbidities recorded (12 types), we employed an initial step to screen for promising associations: only comorbidities with a value of P < 0.1 were considered for further inclusion in the prediction model.

A backward stepwise multivariable logistic regression procedure was then carried out, starting with all of the comorbidities and PSG parameters that met the initial screening step. Age was included in the model as an a priori choice. The predictive performance of this model was estimated with a receiver operating characteristic area under the curve. The stepwise selection process was then validated using a bootstrap resampling procedure with 500 repetitions, and an optimism-corrected area under the curve was computed.

All statistical analyses were performed using R version 3.5.2. Resampling validation was performed using the rms R package.²⁶

RESULTS

A total of 374 children were included in the study analysis, with a median age of 6.1 years (Table 1). Thirty-nine of the cohort (10.4%) were younger than 3 years at the time of surgery, 120 (32.1%) were 3 to 6 years old, 215 (57.5%) were 6 years or older. A total of 286 of the population (76.5%) had at least one identified comorbidity and 178 (47.6%) had ≥ 2 comorbidities. In addition, 34.6% of the study population were obese (body mass index [BMI] greater than the 95th percentile). Of the 121 children with obesity (34.6%), 76 (62.8%) had at least one other comorbidity. Of the 96 children with a genetic syndrome (25.7%), 93 (96.9%) had at least one other comorbidity.

Table 1.

Baseline demographics (n = 374).

Seventy-seven children were described as having cardiovascular disease, including 13 (17%) with active disease and 51 (66%) with a Down syndrome diagnosis. Most of these children (52, 68%) had undergone surgical treatment of underlying congenital heart disease or had a current lesion at the time of their AT; spontaneous resolution was reported in 25 (32%). For the children who had a surgical repair or who had current cardiac disease at the time of AT, categories of cardiovascular disease were as follows (%): shunt lesions (73%), composed of atrioventricular septal defects (37%), ventricular septal defects (23%) and patent ductus arteriosus (21%); miscellaneous lesions (38%), including mitral/tricuspid valvular lesions (50%) and reported systemic hypertension (40%); left and right ventricular outflow tract obstructive lesions and single ventricle physiology, which were present in fewer than 13% of patients. Seventeen patients (32%) had two or more coexisting lesions. Current or a past history of pulmonary hypertension was identified in 10 patients (19%).

Three-hundred forty-four of the cohort (92.0%) had sleep-disordered breathing, which was characterized as mild in 112 (32.6%) (AHI > 1 to 5 events/h), moderate in 84 (24.4%) (AHI > 5 to 10 events/h) and severe in 148 (43.0%) (AHI > 10 events/h). Two hundred fifty-six children (68.4%) had obstructive sleep apnea (obstructive AHI > 1 event/h).

Sixty-six children (17.6%) had at least one postoperative respiratory complication occur which required intervention (Table 2). One of the children (0.3%) required bag and mask ventilatory support. Eighteen of the children (4.8%) required airway support. Intubation was needed in one child (0.3%). Oxygen desaturation requiring intervention occurred in 58 of the children (15.5%). Overall, respiratory complications were significantly more frequent in children with underlying airway anomaly, craniofacial, genetic, prematurity, cardiac, or neurological conditions (Table 3).

Table 2.

Severe respiratory complications by comorbidity (n = 374 patients, n = 66 of whom had a least one severe complication).

Table 3.

Univariate logistic regression model of complications by comorbidities and polysomnogram measurements.

Polysomnographic variables considered as predictors of major postoperative respiratory complications were AHI and lowest oxygen saturation nadir, both of which were predictive of PRAEs. Higher AHI and lower oxygen saturation nadir were both associated with higher odds of PRAEs (Table 3).

Applying the backward stepwise selection among all PSG and comorbidity predictors with at least a value of P < .1 in the univariable models identified cardiac comorbidity, airway anomaly, neurological condition and history of prematurity as significant predictors of PRAEs, along with younger age, which was selected a priori (Table 4). Odds ratios for the associations of these variables with PRAEs are as follows: cardiac comorbidity 2.09 (95% confidence interval [CI] 1.11, 3.89), airway anomaly 3.19 (95% CI 1.33, 7.49), neurological condition 1.89 (95% CI 0.99, 3.53), history of prematurity 4.13 (0.93, 19.25), and age younger than 3 years: 4.10 (95% CI 1.79, 9.26; or 3 to 6 years: 2.21 (1.18, 4.15). The unadjusted area under the curve for the receiver operating characteristic curve constructed for our model was 0.74. When the area under the curve was corrected for overfitting, it was 0.68.

Table 4.

Final prediction model.

Most children in our cohort had preplanned postoperative hospital admissions (304/374, 81.3%). Of those without a planned postoperative hospital admission, 16/70 (22.9%) had an unplanned hospital admission (Figure 1). All children with unplanned admissions were admitted for respiratory reasons, with the exception of two cases (one with postoperative bleeding and one with a dental abscess). Figure 1 depicts where patients were located in the hospital when PRAEs occurred.

Figure 1. Disposition of patients and respiratory complications.

Disposition of patients and their respiratory complications. Boxes represent patients with planned or unplanned admissions and their ultimate location of care (rectangular, shaded cells). “N” represents the number of individuals in each cell. “n” represents the number individuals in each cell with at least one respiratory complication. The number (and percentage) of individuals in each location who had each type of respiratory complication is also listed. Note that two additional patients were admitted because of nonrespiratory complications: one with postoperative bleeding and one with a dental abscess.

DISCUSSION

In this study we evaluated a medically complex surgical population of children undergoing AT, who had undergone a preoperative PSG, for PRAEs prior to hospital discharge and their predictors. Seventeen percent of the study population experienced a major respiratory complication, an incidence similar to that reported in other medically complex populations.^10,27,28 In a multivariable model, the odds of a PRAE were twice as high in children with cardiac disease, three times higher in children with airway disease, and two to four times greater in children younger than 6 years. The odds of PRAEs were estimated to be higher for children with a neurological condition or history of prematurity, although in both cases their 95% CIs crossed 1. Although AHI and oxygen saturation nadir were associated with increased odds of PRAEs in univariable analysis, they were not found to be independent predictors of PRAEs.

The findings of our study are consistent with existing clinical care guidelines that suggest inpatient admission postoperatively for children younger than 3 years, cardiac disease, airway anomaly (craniofacial abnormality) or neurologic disease, because of higher risk of PRAEs.¹ Most of our findings related to comorbid conditions predicting postoperative respiratory complications were consistent with existing literature. Cardiac and craniofacial conditions have been associated with risk of postoperative complications in other studies.¹² Interestingly, genetic condition was not found to be an independent predictor of respiratory complications in our study. As previous studies have shown that having Down syndrome was significantly associated with PRAEs and higher rates of postoperative pediatric intensive care admission,¹³ perioperative planning and management in this population may have differed, which may have avoided some respiratory complications. The use of perioperative admission guidelines has been shown to reduce postoperative respiratory complications.²⁹ In our study, as reported elsewhere, obesity (BMI > 95th percentile) on its own was not found to be an independent risk factor for postoperative respiratory complications.³⁰ Morbid obesity (BMI > 99th percentile for age and sex) has been shown in other studies, however, to be an independent risk factor for PRAEs in this surgical population.^31,32 In our cohort, most of the children with obesity (63.1%) and those with genetic syndromes (96.9%) also had other comorbidities, which appear to be more strongly predictive of PRAEs. Similarly, although other studies have specifically identified age younger than 3 years^10,12 or 4 years¹⁷ as a risk factor for PRAEs; in our population, most of whom had comorbidities, age younger than 6 years was associated with increased risk of PRAEs in this cohort. Finally, a history of prematurity has been associated in some studies with increased risk of PRAEs, as was suggested in our cohort.^14,33

In our study, none of the polysomnogram or oximetry parameters considered were independently associated with increased risk of postoperative respiratory complications. Previous studies have varied considerably in terms of the population studied and the candidate predictor variables considered, making direct comparison of results challenging. Our finding differs from previous studies in medically complex populations, in which preoperative PSG oxygen saturation nadir predicted postoperative respiratory complications, independent of AHI, in children known to have severe OSA or increased medical complexity.^2,28,34 Previous studies have reached variable conclusions about whether AHI is independently associated with PRAEs, demonstrating a significant association in a more complex tertiary care population¹⁰ and/or at a higher AHI (> 20 events/h)¹⁰ but not in an older community population (aged 5 to 9 years) without medical comorbidities.³⁰ Another recent study suggested an AHI threshold above 25 events/h was predictive of postoperative complications, whereas oxygen desaturation was not, although that study had a higher proportion of individuals with severe OSA (59% versus 43%) and higher recorded AHIs than in our study.³⁵ Similar to the findings of our study, Konstantinpoulou et al³⁰ reported PSG oxygen desaturation nadirs were not predictive of PRAEs in an older population of children (aged 5 to 9 years), although their sample lacked comorbidities.

The findings of our study differ from existing guideline recommendations pertaining to the AHI and oxygen desaturation. Although our study found that AHI was a predictor of PRAEs in isolation, when considered along with other variables in a model, it did not strengthen predictions. It is possible that our older study population (median age 6.1 years; interquartile range 3.9; 9.3) associated with a high incidence of PSG-defined OSA (68%) and a limited range of reported AHI values (median 7.6; interquartile range 3.1, 16.1) limited the predictive power of AHI for PRAEs. In our study, desaturations < 80% were not associated with increased risk of respiratory complications, which differs from guidelines from the American Academy of Otolaryngology-Head and Neck Surgery and the American Academy of Pediatrics,^1,22 which recommend this threshold of oxygen desaturation to trigger preplanned hospital admission. It appears that in our medically complex population, the presence of comorbidities is a stronger indicator of risk of PRAEs than nocturnal respiratory events or gas exchange abnormalities.

Overall, our study results highlight the need to better define the complex interaction between comorbidities, age, nocturnal respiratory events, and gas exchange abnormalities in predicting risk for PRAEs after AT. As a result, the utility of a preoperative PSG or oximetry for risk stratification is questionable in this surgical population, but it may be unavoidable from an OSA diagnostic perspective to rule in or rule out the need for surgery.³⁴ Also, as many as one-third to one-half of the children in a less medically complex population⁷ (1995–2013 Closed Claims Project, personal communication Karen L Posner) who experience severe adverse events after AT lack obvious age or clinical risk criteria. It is possible that children’s patterns of intraoperative and postoperative sentinel respiratory events also need to be considered to better predict significant PRAEs.³⁶

Strengths and limitations

The main strength of this study was that data were collected in a large cohort of children undergoing AT, for which preoperative laboratory PSG information was available for the entire cohort along with clinical history. This allowed us to simultaneously study the demographic attributes, comorbid conditions, and PSG variables as predictors of postoperative respiratory complications. This study was limited in that it was conducted in a referred population at a single tertiary care pediatric center, where all patients had undergone PSG. Studying children in a tertiary care setting who have undergone PSG introduces potential generalizability issues for a more medically complex population. Our findings, however, remain pertinent to other pediatric tertiary populations. It should be noted that the frequency of respiratory complications was much greater than that observed in a large prospective study of healthy children,³⁰ but similar to that of another large study in a pediatric tertiary center.¹⁰ The inclusion of children with a range of comorbid conditions allowed us to evaluate those as predictors, along with the PSG variables studied. Unfortunately, information on race was unavailable and this known risk factor for respiratory complications postadenotonsillectomy¹¹ could not be assessed in our study.

Variations in perioperative interventions to mitigate risk of postoperative complications such as time of day of surgery or intraoperative and postoperative medications administered were not considered in this analysis, but these perioperative factors may also contribute to risk of postoperative respiratory complications. Previous studies have reported varying degrees of association between perioperative morphine dose and risk of respiratory complications.³⁷ Also, the risk environment for PRAEs occurring in the PACU is affected by the pediatric experience of the attending surgeon, anesthesiologist, and nursing staff, which may limit the external application of our findings in a community care setting.^33,38,39 Further, most comorbidities represent a spectrum of disease and therefore the specific comorbid phenotype of those identified to increase risk for a PRAE is not clear. Moreover, it is uncertain whether PRAEs requiring an intervention during the immediate recovery period are predictive of major PRAEs occurring after PACU discharge. Finally, criteria for preplanned hospital admission may have changed over the study period as new guidelines guiding disposition planning were published.²²

Although the prediction model was developed and internally validated using rigorous methods, the sample size did not permit inclusion of all comorbidity variables in a single model and a preselection of variables was made. Additionally, we were unable to explore obstructive AHI due to its collinearity with total AHI; maximum carbon dioxide was not included due to frequent missing data. In the future, it would be interesting to additionally explore oxygen desaturation index and/or time with oxygen saturation below 90% as potential predictors of PRAEs, as recent literature suggests that PSG-based gas-exchange abnormalities that consider depth and duration of oxygen desaturation may have greater predictive power for PRAEs compared with AHI.^10,30,40

Last, we have been able to describe the frequency of unplanned hospital admissions in our population and have determined that the majority are related to PRAEs; however, it should be noted that most children in our cohort had preplanned postoperative hospital admissions, limiting our ability to evaluate the role PRAEs may play in determining disposition. Further, we were not able to evaluate how PRAEs may influence time in the recovery room or overall hospital length of stay.

CONCLUSIONS

In conclusion, this is the largest cohort to date that simultaneously evaluated comorbidities and polysomnographic variables in a complete cohort, as predictors of postoperative respiratory complications after AT. Underlying cardiac disease, airway anomaly, and young age were identified as independent predictors for increased risk of PRAEs. PSG metrics and oxygen saturation nadir were not found to independently predict PRAEs, suggesting that medical comorbidity is a more important factor in predicting PRAEs in a medically complex population. Further refinement of a clinical decision tool to guide clinicians caring for medically complex children is needed to inform perioperative planning and disposition.

DISCLOSURE STATEMENT

Work for this study was performed at the Children’s Hospital of Eastern Ontario, 401 Smyth Road, Ottawa, Ontario, Canada, K1H 8L1. Dr. Katz has received a speaker honorarium from Biogen, unrelated to this work. The other authors report no conflicts of interest.

ABBREVIATIONS

AHI

apnea-hypopnea index

AT

adenotonsillectomy

BMI

body mass index

OSA

obstructive sleep apnea

PACU

postanesthetic care unit

PSG

polysomnography

PRAEs

perioperative respiratory adverse events