Sociodemographic Disparities in Extracorporeal Membrane Oxygenation Use: Shedding Light on Codified Systemic Biases

Extracorporeal membrane oxygenation (ECMO) is a life-prolonging, resource-intensive intervention with limited availability that is used when conventional therapies have been unsuccessful and patients require temporary pulmonary and/or cardiovascular life support. Prior studies have demonstrated inequitable outcomes among patients already receiving ECMO, but less is known about inequities in patient selection for ECMO (1).

In this issue of AnnalsATS, Mehta and colleagues (pp. 1166–1174) report the results of a retrospective cohort study to measure sociodemographic differences in ECMO receipt in the United States (2). They used the Nationwide Readmissions Database, an all-payer administrative database containing all nonfederal discharges from contributing states, to compare 2,189,477 patients who received mechanical ventilation alone or ECMO between 2016 and 2019. They hypothesized that sociodemographic characteristics of patients in the pipeline from mechanical ventilation to ECMO would be stable in a perfectly equitable system. Using hierarchical logistic regression, they controlled for age, chronic illnesses, illness severity, and hospital and found that the odds of receiving ECMO were 27% lower for women compared with men, 45% lower for patients with Medicaid compared with patients with commercial insurance, and 37% lower for patients from the lowest-income-quartile ZIP codes compared with patients from the highest-income-quartile ZIP codes. Bolstering the validity of their findings, they found a dose–response relationship between area-based income and odds of receiving ECMO. The strengths of their study are the use of a national database to study an uncommonly used treatment, the use of e-values to estimate residual confounding, and thoughtful and numerous sensitivity analyses to interrogate potential sources of bias (although, as they acknowledge, this may increase the probability of a type I error).

Their study also highlights two challenges of working with administrative data. First, the ability to accurately estimate illness severity—or in this case ECMO eligibility—is limited. In a sensitivity analysis using seven state inpatient databases, the authors used the best available measure, an acute organ failure score that performs similarly to the Sequential Organ Failure Assessment score in predicting mortality risk. However, both administrative and physiologic risk scores are less well calibrated among patients with the highest predicted mortality risk (3), and as the authors note themselves, mortality risk and ECMO eligibility may not have a linear relationship (i.e., those with lowest and highest risks are likely to be deemed “too well” and “too sick,” respectively, to benefit from ECMO). Thus, it remains unclear if the observed disparities reflect mechanisms preceding or subsequent to ECMO eligibility.

Second, administrative data include limited and imprecise data about social drivers of health. For example, Medicaid enrollment is an incomplete proxy for underinsurance or low income. Eligibility is based on a restrictive federal definition of poverty, so many patients who are not eligible to receive Medicaid benefits still experience financial barriers to health (4). Somewhat surprisingly, the Nationwide Readmissions Database does not include race/ethnicity data, because not all states report these data. Area-based deprivation measures are gaining popularity because they provide valuable information about structural influences on health; yet these are also prone to ecological fallacy, whereby inferences about a geographic area may not be applicable to all individuals who reside in that area (5). The geographic unit of analysis is also consequential. Five-digit ZIP codes are a common geographic indicator in administrative data because they are easily derived from addresses and minimize privacy concerns. However, they represent postal delivery points rather than neighborhoods of like individuals and can include anywhere from 1 to more than 100,000 people or range in size from a few blocks to more than 10,000 square miles (6). Still, the authors are to be commended for not only including the socioeconomic indicators that were available to them but also conducting an intersectionality analysis to measure the compounding effects of structural disadvantage (7).

Despite these limitations of administrative data, the top-line message from Mehta and colleagues (2) is an important one: that gender and socioeconomic privilege may be beneficial in securing access to a costly, supply-limited, life-prolonging treatment. This is not unusual in the American healthcare system, with organ transplantation as an archetype (8). The authors posit several mechanisms for their findings that can be framed using the social–ecological model (Figure 1) (9). Patients’ differential preferences for life-prolonging care are unlikely to be a unifying mechanism, as socioeconomic disadvantage is often associated with more, not less, use of life-prolonging treatments (10). Differential access to ECMO-capable hospitals, either during patients’ index hospitalizations or through interhospital transfers, also does not appear to be a primary mechanism, as the observed disparities persist after including hospital random effects in analytic models and restricting the sample to ECMO-capable hospitals. Clinicians’ bias, implicit or explicit, certainly may contribute. There is robust literature describing disparities in how clinicians communicate with and make treatment decisions on behalf of patients (11). Recent work by Modra and colleagues has also defined a gender-based volume–outcome relationship, in which critically ill women experience greater mortality if they have an illness that is uncommon among women (12). This suggests that heuristics, or mental shortcuts, about what a “typical patient” looks like, may influence clinical decisions (13). Finally, we would add a fourth mechanism: that the American healthcare system maintains inequitable access to health resources (14). It is said that ECMO is not a destination but a “bridge” to either recovery from acute illness or mechanical organ support or transplantation. Therefore, during an evaluation of ECMO candidacy, it becomes necessary to prognosticate, a process fraught with bias, and even explicitly evaluate patients’ financial and social resources (15–18). Patients and families who have the most cultural health capital, financial means, and caregiving resources may thus be prioritized as optimal candidates for ECMO. Through such processes, the healthcare system necessarily codifies, values, and perpetuates socioeconomic privilege.

Figure 1.

A modified social–ecological model contributing to the disparities in extracorporeal membrane oxygenation (ECMO) receipt.

Mehta and colleagues (2) have identified an important disparity in the pipeline from mechanical ventilation to ECMO. We hope that future studies will elucidate mechanistic pathways and develop and test solutions. This study also raises enduring questions for our field: about who has access to resource-intensive, supply-limited, life-prolonging therapies and what role the healthcare system plays in gatekeeping access to health and, potentially, life.