Most pediatric patients with obesity and severe obstructive sleep apnea who were classified as high risk for tonsillectomy complications did not experience severe perioperative events, suggesting that preoperative polysomnography cutoffs may help identify lower-risk patients within this group, according to a retrospective study.
Current guidelines recommend overnight admission for patients considered high risk following tonsillectomy, including many with severe obstructive sleep apnea and obesity. However, in the new analysis, 88% of patients classified as high risk did not experience a severe perioperative event, underscoring the clinical heterogeneity of this population.
The researchers reviewed outcomes among 304 patients aged 2 to 18 years who underwent tonsillectomy with or without adenoidectomy at a tertiary pediatric hospital from January 2021 to January 2024. All patients had obesity and severe obstructive sleep apnea confirmed by preoperative polysomnography. Patients with tracheostomy, Down syndrome, mucopolysaccharidosis, neuromuscular disorders, or craniofacial abnormalities were excluded.
The primary outcome was a composite severe perioperative event, defined as intensive care unit (ICU) admission, hospitalization longer than 48 hours, or need for advanced respiratory support, including high-flow nasal cannula greater than 5 L/min, continuous positive airway pressure or bilevel positive airway pressure, or endotracheal intubation.
Overall, 36 patients, or about 12%, experienced a severe perioperative event. Twenty patients required ICU admission, 17 required advanced respiratory support, and 32 had hospital stays longer than 48 hours. The researchers noted that prolonged hospitalization may reflect institutional practice patterns as well as clinical severity, and therefore performed a respiratory-focused sensitivity analysis excluding length of stay from the composite outcome.
In that analysis, 23 patients, or about 8%, experienced a respiratory-focused severe event. Among the 87 patients who met the simplified clinical rule of an apnea-hypopnea index (AHI) below 25 events per hour and oxygen saturation nadir above 85%, 1 patient experienced a respiratory-focused severe event.
Patients with severe events had worse polysomnography findings than those without severe events, including median AHI of 41 vs 23 events per hour and oxygen saturation nadir of 70% vs 84%.
Using Bayesian logistic regression, the researchers found that oxygen saturation nadir was the strongest predictor of severe perioperative events, followed by class III obesity and AHI. The Bayesian approach incorporated prior published evidence into the model, which may provide more stable estimates in a modest-sized retrospective cohort.
Time spent below 90% oxygen saturation, or TB90, was the weakest predictor in the model, despite being identified as a strong predictor in prior studies of postoperative respiratory events. The researchers suggested this discrepancy may reflect the study's focus on identifying patients potentially safe for discharge rather than predicting any respiratory event, as well as the influence of Bayesian modeling.
Model-based stratification identified 101 patients, or about one-third of the cohort, as having a predicted severe-event probability of 5% or lower. This model-derived low-risk group was distinct from the simpler clinical rule subgroup of 87 patients who had AHI below 25 events per hour and oxygen saturation nadir above 85%.
For identifying patients at the very-low-risk threshold, defined as a model-predicted severe-event probability of 2.5% or lower, the AHI below 25 events per hour and oxygen saturation nadir above 85% rule had 85.7% sensitivity and a negative predictive value of 99.5%. At the 5% low-risk threshold, the same rule had lower sensitivity and a lower negative predictive value, reflecting the trade-offs inherent in how conservatively low risk is defined.
A more restrictive rule using AHI of 15 events per hour or lower and oxygen saturation nadir above 85% improved specificity but identified fewer low-risk patients.
Although class III obesity was one of the strongest predictors in the model, it was not included in the simplified two-variable rule. The researchers wrote that polysomnography measures may already capture much of the physiologic risk mediated by obesity, whereas adding obesity class could risk double-counting related risk. They also noted that a two-variable rule may be easier to apply in existing clinical workflows than a more complex risk calculator.
The findings may be relevant to health care utilization as well as clinical risk stratification. The researchers noted that routine admission of all high-risk patients carries an estimated annual cost burden of about $331 million and may limit surgical capacity.
The study should not be interpreted as support for immediate practice change. All patients in the cohort were admitted overnight under existing protocols, so the analysis did not directly test same-day discharge. In this context, same-day discharge refers to discharge from the postanesthesia care unit following standard recovery, not extended 23-hour observation.
The researchers outlined a staged evidence pathway: retrospective identification of candidate criteria, prospective observational validation while maintaining existing admission protocols, potential quality improvement implementation if the criteria are validated, and continued peer review at each stage.
The study was limited by its retrospective, single-center design and predominantly Hispanic cohort, which may affect generalizability. Other limitations included lack of preoperative continuous positive airway pressure data, possible underascertainment of comorbidities from electronic medical record problem lists, and variation in surgical technique.
"Many patients classified as high-risk for perioperative complications are, in fact, at low risk," the researchers wrote. "Prospective validation is warranted before implementation."
Disclosures: The researchers reported no funding. The study was presented as a poster at the Triological Society Combined Sections Meeting, held January 20 to 23, 2025, in Orlando, Florida. Stephen R. Chorney reported serving as an education consultant for Smith and Nephew. Kenneth Lee reported serving as chief medical officer of Qualio Oto. The remaining researchers reported no conflicts of interest. The researchers acknowledged using Claude, an artificial intelligence tool from Anthropic, for manuscript formatting, copyediting, and grammar review.
Source: The Laryngoscope