Children—and their airways—come in all shapes and sizes, and the most seasoned pediatric anesthesiologists can share harrowing stories of patients who were difficult to ventilate or intubate. In 2012, a special interest group in the Society for Pediatric Anesthesia created the Pediatric Difficult Intubation Registry, an international, multicenter, web-based registry of observational data on pediatric difficult airway management, with the goal of improving the care of these challenging patients. The Pediatric Difficult Intubation Collaborative has produced a panoply of clinically relevant findings, and their most recent research sought to answer how these patients fare when managed with sedation versus general anesthesia. When confronted with a potentially or known difficult airway, most pediatric anesthesiologists induce general anesthesia, albeit with the goal of maintaining both adequate depth and spontaneous respiration.1 In contrast, there are some who decry general anesthesia may “burn the bridge” of spontaneous ventilation and instead choose to walk the tightrope of sedation that is deep enough to diminish airway responses yet light enough to maintain spontaneous ventilation and reverse course if airway management is too treacherous.
In this issue of Anesthesiology, Sequera-Ramos et al. drew on data from the Pediatric Difficult Intubation registry with the aim of comparing outcomes associated with sedation versus general anesthesia for tracheal intubation in children with difficult airways. Sequera-Ramos et al. used propensity score matching to address selection bias and other confounders inherent in the analysis of retrospective observational, real-world clinical data. The primary study outcome was first-attempt success of tracheal intubation, which is a clinically important outcome because multiple attempts at intubation are associated with worse outcomes in pediatric airway management. Sequera-Ramos et al. observed similar rates of first-attempt success of tracheal intubation in the propensity score-matched sedation and general anesthesia groups (48.3 and 47.9%, respectively). Very few patients (4%, 75 of 1,839) underwent sedation rather than general anesthesia. Based on the similar rates of first intubation attempt success and complications in the two groups, the authors concluded their study did not support a preferred approach for tracheal intubation in children with difficult airways.
What can clinicians take from this study and its statement of equipoise? First, as the authors mention, clinicians should select an approach based on their skill and patient factors—keep calm and carry on. However, hold your horses! Nearly one-third of the 75 attempts in patients receiving sedation required conversion to general anesthesia for successful intubation. A failure rate of 30% in intubation attempts in the sedation group suggests that sedation may be inferior to general anesthesia, and this may be due to an inadequate depth of anesthesia. Indeed, the Pediatric Difficult Intubation Collaborative reported the use of neuromuscular blockade (presumably administered during a general anesthetic) was associated with improved success in some instances of difficult airway management.4 Perhaps this failure rate reflected mild to moderate sedation that was inadequate for intubation, while deep sedation may have been the same as “light” general anesthesia. This is potentially a major limitation of this study, especially as there was no monitoring of the depth of either anesthesia or sedation.8 It is possible that patients defined as receiving sedation received general anesthesia (and vice versa), resulting in crossover. A lack of data on administered medications and end-tidal volatile agent values precludes any conjecture on the subjects’ depth of sedation. Second, as for many retrospective analyses of data sets, there may be known and, of course, unknown confounding. One plausible confounder was severity of any deformity. There was no measure of how difficult the airway was, i.e., severe or mild retrognathia. A child with severe deformity may have been given sedation due to a clinician’s fear of the patient obstructing under general anesthesia. The authors acknowledged both limitations, and these limitations certainly do not mean that the study is uninformative.
This study highlights limitations that are present in many retrospective analyses of data sets, particularly if the data set was created to answer a question that is a little different from the question being asked. The intervention or population of interest may not be as well defined as hoped, and important confounding factors may not have been collected.
How might a researcher address this challenge inherent to retrospective data analyses? One relatively novel approach to determining causal inference from large observational databases such as the Pediatric Difficult Intubation Registry is to perform the analysis in a way that emulates a randomized experiment. This is known as a target trial. In a target trial, a researcher with a large observational data set “imagines” that they are performing a randomized controlled trial and defines the population, eligibility criteria, treatment strategies, outcomes, and timeframe as if it were a trial. An important aspect of the target trial is to aim for reducing all confounding, as if the treatment groups were randomized. A crucial step when trying to remove confounding is to ensure that all biologically plausibly principal factors are included in any adjustments. Ideally, a directed acyclic graph is created a priori and included in the publication. It is good practice to generate the directed acyclic graph to include all key factors regardless of whether they are in the data set. This forces the researcher—and reader—to assess what factors could be missing and what their impact might be on any conclusions of causation.
A target trial of sedation versus general anesthesia based on Pediatric Difficult Intubation Registry data would need to provide details of the proposed interventions: sedation versus general anesthesia with standardized definitions and evidence of compliance. Similarly, if Sequera-Ramos et al. had produced a directed acyclic graph before doing their analysis, then the degree of deformity (and other unmeasured but plausible confounding factors) may well have been elements in the directed acyclic graph, which would have forced them to consider the impact of not including them. Last, in a target trial, the conversion of 27.6% of sedation cases from sedation to general anesthesia would be accommodated in an intention-to-treat analysis, i.e., if you start with the intention to use sedation, you may have to end up converting to a general anesthetic, but that is accommodated in the assessment of whether or not starting with sedation is equivalent to starting with a general anesthetic.
There are increasing amounts of clinical data available for analysis. Often the aim is to determine causal relationships, and most of these published analyses include phrases describing the limitations around confounding and missing data. They are inevitably, and appropriately, circumspect about what to conclude in terms of changing practice. In these retrospective studies, it is often tempting to dismiss any conclusion of causation, but this is likely an overly conservative approach. Target trials and directed acyclic graphs are methods to inch closer to understanding causation in these retrospective data sets, and hopefully readers will see a lot more of them.
For the reasons outlined above, the study by Sequera-Ramos et al. does not provide definite evidence that sedation is equivalent to general anesthesia for managing difficult airways in children. Such evidence would require a prospective trial. Organizers of pro–con debates on pediatric difficult airway management may seem to be the only group that will rejoice at this indeterminate conclusion. However, the study does provide some evidence that general anesthesia may not be a poor choice and may be a good rescue modality. Collaborative sharing and analysis of clinical data are the present and future of generalizable, clinically relevant, trusted evidence, and the Pediatric Difficult Intubation group is an excellent example of productive, meaningful collaboration.