“Practitioners and leaders in the field should reflect on whether there are biases in our practices […].”
The droll line “Houston, we have a problem” by the commander of Apollo 13 announced some troubling observations of unknown significance and impact, and set in motion an urgent but deliberate search for understanding the problem’s meaning for the crew, and to engineer solutions. Although the outcome of the Apollo 13 mission is known, the “we have problem” phrase has become a clarion call for focused attention to a perceived problem, its meaning, and a reliable solution in a hurry. In this issue of Anesthesiology, White et al. observe in a large retrospective cohort that black patients receive perioperative antiemetic administration at a lower rate than white patients.1 This result should be shocking, but like most retrospective research, this study has too many limitations to make definite conclusions about causation, and therefore we are inevitably left somewhat unsure what to do with these data. However, these data cannot be simply dismissed, and given the implications, at the very least they should be seen as a call to do better, prospective research on the topic of nonbiologically based influences on the care we provide—hence the title of this editorial.
One editorialist remembers distinctly the initial viewing and shock at seeing figure 2 from the paper by White et al., essentially unchanged in its now published form, at the poster discussion session at the 2019 American Society of Anesthesiologists (ASA; Schaumburg, Illinois) meeting in Orlando, Florida. Shame, revulsion, and awe were competing emotions as this editorialist carefully lined up the image in the camera of his smart phone and snapped (several) images to ensure the result would not be lost. Shame at the realization that the author’s institution’s performance was included in the contributed data and that the author’s own individual performance might have contributed to the observation that anesthesiologists at least seem to treat their patients differently on the basis of race. Revulsion at the potential evidence of wholesale self-deception despite our pious profession of objectivity and patient-centeredness. Awe at the development of the tools to perform such important and paradigm-challenging research via the Multicenter Perioperative Outcomes Group, hard-fought by its dedicated members since its initial proof-of-concept publication2 to this extraordinarily provocative one. These emotions are the result of attribution biases, apprehending race as a social construct and believing that the results implied poor performance by anesthesiologists, and that the empowered researchers were committed to doing the right thing.
The uncomfortable counterpoint is that there might be a biologic basis for a link between race and postoperative nausea and vomiting susceptibility and that this biologic basis is implicitly recognized and accounted for by anesthesia care teams. This would mean that the observed treatment is, in fact, appropriately tailored to patient needs, but the information used by the care teams to refine their prophylaxis and treatment decisions is not available in the observational data. The current retrospective study did not collect data on actual postoperative nausea and vomiting rates, which limits the capacity to differentiate between these possibilities. This argument also assumes that there is a plausible link between race and postoperative nausea and vomiting susceptibility, a link that the authors consider invalid and that we shall touch on again.
During the review process, the authors and the peer reviewers realized that the Multicenter Perioperative Outcomes Group database had not been constructed to allow distinction between antiemetics administered for postoperative nausea and vomiting prophylaxis and those administered for rescue in the postanesthesia care unit. The authors of the study had developed their protocol prospectively through multiple rounds of peer review discussion of the Multicenter Perioperative Outcomes Group Perioperative Clinical Research Committee with the intention of studying associations between anesthesiologists’ postoperative nausea and vomiting prophylaxis practices and patient race, and these intentions are documented in the supplemental protocol accompanying the manuscript. Does this matter? The authors argue not, and that both treatment and prophylaxis would be subject to the same bias. This makes sense. But what if race had an impact on postoperative nausea and vomiting and hence treatment? This seems like a plausible idea, but the authors argue there is in fact no biologic reason to suggest this as a possibility. Any hypothesis that race has an impact on the pharmacology of the drugs we use or the biologic mechanisms underpinning any anesthesia outcome must be driven by what is known about the pharmacogenomics and physiology, and not seen simply as an excuse for disavowing social constructs of race. Nevertheless, some will argue that it would have been a more informative study if the investigators could have unpicked the distinction between prophylaxis and treatment, and if they had measured the actual incidence of postoperative nausea and vomiting. With the benefit of hindsight, it is clear that despite the best of intentions, many people probably missed this potential limitation in this Multicenter Perioperative Outcomes Group–derived dataset. However, the peer review process at the Journal in this instance was both rigorous and gracious, and the authors were able to directly address and refute this limitation, reporting an important observation of how patients at many of the contributing Multicenter Perioperative Outcomes Group centers appear to have been treated with respect to race. After almost 4 yr between when this result was first shown at the ASA in 2019 and today’s publication, this troublesome observation has completed an arduous peer review process and now confronts us. One regrets the time lost in that interval but admires the tenacity of the reviewers and authors to get this as right as possible.
When considering treatment as opposed to prophylaxis, another interesting facet of the study may appear. The database records antiemetics charted and not those administered. It is possible that race had an impact either on whether or not the drug was charted, or on whether or not the caregiver in the postanesthesia care unit decided to administer the drug, or both. The authors suggest that the anesthesiologist has ultimate responsibility, which includes taking responsibility for the actions (or inaction) of others in the perioperative care team, so all results attribute to the anesthesiologist. Also, it is possible that anesthesiologists might recognize more protective Apfel criteria3 in their black patients than in their white ones, but do a poorer job in documentation for their black patients, thus only appearing to undertreat black patients. But what does this say, then, about their diligence and potential bias in documentation?
For all observational studies that seek to understand causation, investigators should define their specific clinically relevant questions and use the appropriate tools to assess causality, for example by constructing Directional Acyclic Graphs4 and then assessing whether the available data are able to answer the specific questions. If the appropriate data are not available in the retrospective dataset, the researchers should consider rejecting the question or collect the data prospectively. Postoperative nausea and vomiting is a topic that has been extensively studied. Here the authors adjusted for numerous obvious confounders (including the Apfel postoperative nausea and vomiting risk factors) and give us some idea of how big an unrecognized confounder could be required to undermine their troubling observation that black subjects with the same indications for postoperative nausea and vomiting treatment are less likely to receive it than white subjects. The biologically plausible and well-described confounding factors are largely known, but there are also plausible social determinants such as insurance status, wealth or income, and socioeconomic status that may all influence the decision to provide prophylaxis. The authors could not adjust for these and accept this limitation. It cannot go unsaid, however, that if any of these did influence prophylaxis or treatment, then we still have a problem.
In summary, like in all observational studies, the influence of confounding cannot be completely erased by adjusting for known confounders, and even though the risk factors for postoperative nausea and vomiting are well- described, the possibility of hidden confounders still lurks. This Journal is always loath to conclude that more research needs to be done, but if ever there was a case to do so, then here it is.
What next? Although the observation in the paper by White et al. cannot definitively tell us about whether the differential treatment has social, biologic, or combined origins, the observation and the ambiguity over its basis raises important subsequent questions. If anesthesiologists actually differentially treat patients for postoperative nausea and vomiting (arguably not a life or death treatment), can we be assured that these same anesthesiologists somehow change modes and treat every patient equitably in situations when it really matters? If we accept that there is no biologic basis for the observed difference in antiemetic treatment (and we have no evidence to assume there is), we should be concerned that discrimination pervades all our activities, including both vigilance and treatment. The authors focused on black race as distinct from white race, but there was a third race category of not white, not black, and these individuals assorted with the black subjects in receipt of antiemetic treatment. Let us not neglect these people.
Although not the intent of the article, the authors have additional troubling results in their data. For example, there is an inverse relationship between ASA Physical Status classification and the likelihood of receiving antiemetics. Why is this the case? Is the presence of minor physiologic derangements (ASA Physical Status II) or serious conditions (ASA Physical Status III) such a distraction that anesthesiologists cannot find time or attention to avert or treat postoperative nausea and vomiting? Or is this evidence of a subconscious or conscious ableist orientation? Similarly, there was a strong inverse relationship between the presence of an advanced trainee and the application of antiemetics. This is almost surely an epiphenomenon of the selective assignment to advanced trainees to patients with higher ASA Physical Status driven by training goals, but what are we then teaching them?
In most of 40 centers staffed by well-intentioned, educated anesthesiologists, implicit bias could be at work, as assessed by differential treatment on a task almost fully in the control of the anesthesia team. The apparent differential treatment is so pervasive that it just cannot be ignored, or attributed to a few centers in certain parts of the country. Electronic medical records are widespread, and in postoperative nausea and vomiting prophylaxis and treatment, we have an objective measure that can be monitored and improved. Nudges and checklists are variably successful at improving postoperative nausea and vomiting prevention performance.5 In related domains, aversive stimuli might be more effective at changing performance.6–8 Payer incentives to achieve complete antiemesis risk stratification and equal treatment can be readily effected if needed.
Payment incentives to achieve process equity for postoperative nausea and vomiting management neglects the larger, more important concern that bias may be pervasive in our practice. Several lines of work are urgently needed. After careful consideration of the capabilities of the Multicenter Perioperative Outcomes Group database (and other similar registries), we should identify other areas where patient and surgical interactions have small impacts and assess whether more apparently biased practices can be found (or refuted). This will help us understand how pervasive the problem might be. We should, at the same time, eschew reflexive sweeping practice changes aimed at eliminating apparent bias before prospective studies have been conducted to properly risk-stratify all patients equitably, or perhaps even trials that apply appropriate treatments, with randomization between usual care and consensus best practice, and a blinded, objective assessment of prespecified but comprehensive outcomes, including unexpected ones. The recent experience of perioperative medicine contains too many examples of strong retrospectively identified associations followed by negative prospective trials to safely omit this step. Still, the crescendo focus on race, racism, and discrimination demands urgency. While we await the needed “pervasion” assessments and prospective results, action is required—although we must match the action to the underlying evidence. Practitioners and leaders in the field should reflect on whether there are biases in our practices, and whether such biases might affect other, even more significant medical treatment factors under our control. Without even a clear answer to this question, what should we do?
Recognition of bias and deliberate personal actions to combat it in one’s personal practice come to mind, if only on the basis of research from other fields. During routine anesthesia provision, we train ourselves to systematically scan and re-scan our environment throughout each case, typically guided by a checklist. At a minimum, and with minimal risk, anesthesiologists could add to the checklist, “This patient is or is not white; is or is not healthy. Am I treating them the same?”
The authors are members of the editorial board of Anesthesiology.