Zero Events, False Comfort: Why Less Oxygen Before Extubation Is Not Yet Proven Safer

A randomized trial published in Pediatric Research has attracted attention for a deceptively simple finding: when children about to have their breathing tubes removed were given 50 percent inspired oxygen instead of 100 percent, lung ultrasound detected far less atelectasis, the collapse of small air sacs that commonly follows anesthesia. The result has been widely read as evidence that dialing down the oxygen before extubation is not just gentler on the lungs but also safe. A new correspondence in the same journal, authored by Edgar Fabián Manrique-Hernández and Jorge L. Alvarado-Socarras of Hospital Internacional de Colombia and Fundación Cardiovascular de Colombia, argues that this reading goes well beyond what the data can support. Their central point is statistical, and it is one that applies far beyond this single trial: the absence of observed adverse events in a small group of patients is not the same as the absence of risk.

The original trial, conducted by Gupta and colleagues, randomized children undergoing surgery to receive either 50 percent or 100 percent inspired oxygen in the period before extubation, the moment when the endotracheal tube is removed and the child resumes breathing independently. The rationale for the comparison is rooted in well-established respiratory physiology. High concentrations of inspired oxygen wash nitrogen out of the alveoli, the tiny gas-exchange sacs at the end of the bronchial tree. Nitrogen is poorly soluble and acts as a mechanical splint that keeps alveoli open. When it is replaced by oxygen, which is rapidly absorbed into the bloodstream, the alveoli lose their internal support and can collapse, producing atelectasis. This phenomenon, known as absorption atelectasis, is especially relevant in children, whose lungs have smaller airway calibers, less collateral ventilation, and a higher closing capacity relative to their functional residual capacity, making them more vulnerable to collapse during and after general anesthesia.

Consistent with that mechanism, the trial found that the children preoxygenated with 50 percent oxygen had a markedly lower incidence of immediate ultrasound-detected atelectasis than those given 100 percent oxygen. Lung ultrasound, which has become an increasingly popular bedside tool in pediatric anesthesia because it avoids radiation exposure, can visualize the loss of lung aeration as characteristic artifacts and consolidated regions. The finding is physiologically plausible and aligns with earlier work, including a prospective randomized study by Kim and colleagues showing that lower oxygen concentrations during induction of general anesthesia reduce atelectasis formation in children. On the question of lung aeration, the evidence base is coherent and mechanistically grounded.

The complication arises when the trial’s results are extended from a surrogate imaging outcome to a clinical safety claim. The correspondence authors emphasize that the study was neither designed nor powered to compare clinical safety between the two oxygen strategies. In trial methodology, the distinction between superiority, equivalence, and non-inferiority designs is fundamental. A superiority trial asks whether one intervention is better than another on a specified outcome. A non-inferiority trial asks whether a new or alternative strategy is not unacceptably worse than the standard, typically with respect to safety or efficacy, and it requires a pre-specified margin, an adequate sample size, and formal statistical testing against that margin. Without those elements, any statement that a lower-oxygen strategy is as safe as the conventional one is an inference the data cannot carry.

The most striking illustration of this problem involves the desaturation events. In the group of 32 children who received 50 percent oxygen before extubation, the trial reported zero episodes of arterial oxygen desaturation, the clinically feared complication in which blood oxygen levels fall after the airway is unprotected. On its face, zero events sounds reassuring. But as the correspondence authors point out, invoking a classic methodological paper by Hanley and Lippman-Hand published in JAMA in 1983, whose title poses the question directly: if nothing goes wrong, is everything all right? The answer, statistically, is that zero events in a finite sample only sets an upper bound on the true underlying risk. The rule of three, a widely used approximation derived from the Poisson distribution, states that if zero events are observed in n patients, the 95 percent confidence interval for the true event rate extends up to roughly 3 divided by n.

Applied to this trial, the arithmetic is sobering. With 32 children and zero desaturation events, the upper limit of the 95 percent confidence interval for the underlying desaturation risk is approximately 9 percent. In other words, the observed data are entirely compatible with a scenario in which nearly one in ten children given 50 percent pre-extubation oxygen would experience a desaturation episode. That is not a demonstration of safety; it is a statement of ignorance bounded only by the sample size. A risk of 9 percent for a perioperative hypoxemic event in children would be considered unacceptably high by most anesthesiologists, particularly given that the APRICOT study, a large prospective multicenter observational study across 261 European hospitals published in The Lancet Respiratory Medicine, documented the substantial burden of severe critical respiratory events in pediatric anesthesia. Children are physiologically predisposed to rapid desaturation because of their higher oxygen consumption and lower functional residual capacity, leaving a narrow window between airway loss and hypoxemia.

This is why the correspondence calls for an adequately powered non-inferiority evaluation before any safety claim is made. Such a trial would need to pre-specify a clinically acceptable margin for desaturation risk, calculate the sample size required to demonstrate that the 50 percent oxygen strategy does not exceed that margin relative to 100 percent oxygen, and follow standardized reporting guidance such as the CONSORT extension for non-inferiority and equivalence trials published by Piaggio and colleagues in JAMA in 2012. The methodological literature on non-inferiority designs, including a review by Mauri and D’Agostino in the New England Journal of Medicine, highlights how easily these trials can be misinterpreted when their logic is conflated with that of superiority testing. A trial that shows a benefit on one outcome, such as reduced atelectasis, does not automatically confer a favorable profile on another, such as perioperative hypoxemia, especially when the second outcome is rare and the sample is small.

The distinction matters practically because preoxygenation before extubation serves a genuine protective function. Maintaining a high fraction of inspired oxygen in the minutes before tube removal creates an oxygen reservoir in the lungs, buying time if the child develops laryngospasm, breath-holding, or airway obstruction during emergence from anesthesia. These events are common in pediatric practice and can precipitate rapid desaturation. Reducing the inspired oxygen concentration to 50 percent diminishes that reservoir. The trade-off is therefore real: less absorption atelectasis on one side, potentially less reserve against hypoxemia on the other. Deciding where the balance lies is precisely the kind of question that requires a properly powered clinical endpoint trial, not extrapolation from imaging findings and zero-event counts. Until such a trial exists, clinicians adopting the lower-oxygen approach on the basis of the current evidence are making a choice about which risks they prefer, not implementing a validated safety equivalence.

The broader lesson extends well beyond pediatric anesthesia. In medicine, small trials that report zero adverse events are routinely cited as evidence of safety, a pattern that Hanley and Lippman-Hand dissected more than four decades ago and that persists in the era of rapid publication and social media summaries. Confidence intervals around zero-event rates are wide, and the upper bounds can conceal clinically meaningful risks. The correspondence by Manrique-Hernández and Alvarado-Socarras, published in Pediatric Research in October 2026, is a concise reminder that mechanistic plausibility, surrogate outcomes, and reassuring numerators must be separated from formal demonstration of comparative safety. The finding that 50 percent inspired oxygen reduces immediate ultrasound-detected atelectasis in children before extubation stands on solid physiological ground and merits continued investigation. Whether that benefit comes at an acceptable cost in desaturation risk remains, for now, an open question that only an adequately powered non-inferiority trial can answer.

Subject of Research: Statistical interpretation of a pediatric trial comparing inspired oxygen concentrations before extubation and atelectasis risk

Article Title: Reduced atelectasis does not establish comparative safety of lower pre-extubation oxygen

Article References: Manrique-Hernández, E. F., & Alvarado-Socarras, J. L. (2026). Reduced atelectasis does not establish comparative safety of lower pre-extubation oxygen. Pediatric Research. https://doi.org/10.1038/s41390-026-05572-6

Image Credits: AI Generated

DOI: 10.1038/s41390-026-05572-6

Keywords: atelectasis, extubation, pediatric anesthesia, oxygen therapy, non-inferiority trial, lung ultrasound, desaturation, preoxygenation, clinical trial design, Pediatric Research, absorption atelectasis, zero numerator