Evaluating Approaches to Improve Equity in Critical Care Resource Allocation in the COVID-19 Pandemic

To the Editor:

The coronavirus disease (COVID-19) pandemic has forced healthcare systems to develop strategies to allocate critical care resources when demand outstrips supply (1). The pandemic has also disproportionately impacted Black patients (2, 3), for whom baseline health disparities are well documented and largely driven by inequity in social determinants of health. Concerns about the potential for inequity in resource allocation were raised early in the pandemic, especially if morbidity limiting near-term survival was factored into allocation decisions. Two mitigation strategies to avoid inequity in allocation have been proposed: eliminating consideration of expected survival beyond 1 year and incorporating measures of social disadvantage such as the Area Deprivation Index (ADI) (2, 4, 5).

Few studies have empirically evaluated the potential impact of allocation frameworks on disparities, and none have assessed the impact of these proposed modifications. We analyzed the distribution of allocation scores by patient race and modeled the impact of proposed modifications at four academic hospitals in Atlanta, Georgia.

Methods

We obtained data on adult non-Hispanic White (NHW) and Black (NHB) ICU patients admitted between September 1, 2020, and January 8, 2021, to four academic hospitals. For patients with multiple admissions, only the first was included. This study was approved by the Emory Institutional Review Board as part of the COVID-19 Quality and Clinical Research Collaborative.

Allocation scores were derived from three components. 1) Sequential Organ Failure Assessment (SOFA): defined as the maximum score the day of ICU admission. 2) Underlying Conditions Question (UCQ): for every patient, attending physicians were asked to assess whether the patient’s expected 1- or 5-year mortality was >50% based on preexisting medical conditions (independent of acute illness causing hospitalization). 3) ADI: the ADI ranks census blocks by socioeconomic disadvantage, ranging from 1 (least disadvantage) to 10 (most) (6).

We considered four allocation frameworks, adapted from criteria commonly used in state triage guidelines (7). The first (“Original”) prioritized saving lives (defined by surviving the admission) and life-years (defined by survival after discharge). The allocation score is a combination of points derived from the SOFA score and the UCQ score and ranges from 1 to 8. SOFA scores were assigned point values as follows: 1 (<6), 2 (7–8), 3 (9–11), and 4 (>12). Patients with >50% estimated probability of death from preexisting conditions within 1 year received a UCQ score of 4, and those with >50% probability of death within 5 years received a UCQ score of 2. Lower allocation scores would receive a higher priority for resource allocation.

The second framework removes consideration of 5-year mortality (“Mortality”). Patients with >50% estimated probability of death from preexisting conditions within 1 year received a UCQ score of 4; all others received a score of 0. The third framework (“ADI”) subtracts 1 point from the overall score for patients living in a neighborhood with an ADI ⩾8. The fourth framework combines both “Mortality” and “ADI” modifications (“Combined”).

If allocation became necessary, decisions would be made for patients above score thresholds determined by the degree of resource scarcity. We described patient characteristics by race and modeled allocation score distribution using thresholds from 1 (most resource scarcity) to 7 (least resource scarcity). We tested for differences in proportion of scores above each threshold by race for each framework using chi-square tests.

Results

Among the included 3,246 ICU patients, 53.4% were NHB (Table 1). The proportion of NHB patients in the ICU was consistent with overall admissions to the hospital during the study period (53.3% NHB). NHB patients were younger, were more likely to be female and had a higher mean ADI than NHW patients. More than half had no underlying conditions expected to limit near-term mortality (55.7%); 49.7% had maximum SOFA <6.

Table 1.

Demographic Characteristics of ICU Admissions to Four Emory Hospitals between September 1, 2020, and January 15, 2021

	Non-Hispanic White (n = 1,514)	Non-Hispanic Black (n = 1,732)	Overall (N = 3,246)	P Value
Age, yr, mean (SD)	64.9 (16.1)	59.0 (15.3)	61.7 (16.0)	<0.001
Sex, n (%)				<0.001
F	671 (44.3)	902 (52.1)	1,573 (48.5)
M	843 (55.7)	830 (47.9)	1,673 (51.5)
ADI, mean (SD)	3.28 (2.49)	4.89 (2.35)	4.14 (2.55)	<0.001
ADI ⩾8, n (%)				<0.001
Yes	132 (8.7)	249 (14.4)	381 (11.7)
No	1,382 (91.3)	1,483 (85.6)	2,865 (88.3)
UCQ, n (%)				0.42
0	827 (54.6)	982 (56.7)	1,809 (55.7)
2	438 (32.3)	544 (31.4)	1,033 (31.8)
4	198 (13.1)	206 (11.9)	404 (12.4)
SOFA, mean (SD)	5.93 (4.27)	6.18 (4.30)	6.07 (4.29)	0.10
SOFA score category, n (%)				0.44
<6	771 (50.9)	841 (48.6)	1,612 (49.7)
6, 7, or 8	344 (22.7)	392 (22.6)	736 (22.7)
9, 10, or 11	222 (14.7)	280 (16.2)	502 (15.5)
⩾12	177 (11.7)	219 (12.6)	396 (12.2)
Allocation score, mean (SD)
Original	3.04 (1.96)	3.03 (1.87)	3.03 (1.91)	0.91
ADI	2.95 (1.99)	2.89 (1.90)	2.91 (1.94)	0.35
Mortality	2.39 (1.84)	2.40 (1.78)	2.40 (1.81)	0.87
Combined	2.31 (1.87)	2.26 (1.81)	2.28 (1.84)	0.48

Definition of abbreviations: ADI = Area Deprivation Index; SOFA = Sequential Organ Failure Assessment; UCQ = Underlying Conditions Question.

There were similar proportions of NHB and NHW patients with scores above each threshold (Figure 1). For each framework, NHB patients had a higher proportion of scores above 1 or 2. A higher proportion of NHW patients had a score above threshold at a threshold of 3 or higher. The absolute value of the difference between groups was small (0–2.2 percentage points). At a threshold score of 3, adjusting for ADI would result in 16 NHW and 49 NHB patients moving below the threshold into the higher priority group, whereas dropping the 5-year mortality consideration would result in 197 NHW and 236 NHB patients moving into the high priority group. There were no statistically significant differences by race for any framework or threshold. Results were similar when stratified by age above or below 65 (not shown).

Figure 1.

Proportion of ICU patients above specified allocation cutoffs, by patient race. ADI = Area Deprivation Index; SOFA = Sequential Organ Failure Assessment; UCQ = Underlying Conditions Question.

Discussion

Concerns exist that allocation frameworks aimed at saving the most life-years by prioritizing individuals without preexisting conditions limiting near-term survival could exacerbate systemic disparities in health and healthcare in the United States (2). In this analysis of ICU patients across a healthcare system, absolute differences in allocation scores by race were small and not statistically significant. Proposed modifications to allocation frameworks to improve equity did not meaningfully impact the racial distribution of scores.

A strength of this study is that it includes all diagnoses and real-world physician assessments of underlying conditions. Prior work has involved simulated assessments (8) or has focused on patients with specific diagnoses (9), but allocation systems are not diagnosis specific. This analysis is also not specific to any one critical care resource, although it is most directly applicable to ICU beds, a resource for which significant concern about scarcity exists.

This analysis does have limitations. We were unable to model real-time use of these systems to distinguish between discrete, sequentially presenting individual patients who may compete for the same resource at a given time. Related, we were unable to model tie-breaking criteria such as significant age differences and essential worker status that may have equity-promoting characteristics (4, 10). The UCQ was not mandatory, and 36.6% of patients were missing data, although data were not missing differentially by race. We also did not explore adjustment tools for socioeconomic disadvantage other than dichotomized ADI. A sensitivity analysis of different ADI weighting did not reveal a significant impact, but other adjustment methods or thresholds may be more impactful. Finally, results may not be generalizable to other patient populations. Corrections designed to improve equity could have a larger impact in populations with greater disadvantage or stronger associations between race and area-level disadvantage or disparities in underlying conditions scores.

In summary, we did not find significant racial differences in allocation scores using an allocation system based on ethical principles commonly used in state triage guidelines, and proposed equity-promoting modifications did not meaningfully impact racial distribution of allocation scores at four hospitals in Atlanta, Georgia. Our analysis underscores the need for empirical evaluation of allocation frameworks as they would be implemented and of proposed modifications to improve equity. It also reaffirms the need for continued research addressing resource allocation equity.