Avoid bias and discrimination

In this approach, medical criteria should be designed solely to maximise outcomes. Bias would be detrimental to those subject to triage, and would also interfere with outcome maximisation.

Insofar as social injustice occurs because of bias and discrimination, non-discrimination also addresses some social justice concerns. However, it does not address all of them: the consistent application of criteria that exclude people with health detriments that are due to the social determinants of health compounds existing health injustices. Neither does non-discrimination address the egalitarian or epistemic balancing concern that differences in probabilities must be substantial in order to dissuade us from treating everyone equally or from attending to social justice. These are not reasons to reject unbiased application of medical criteria, of course, but reasons to think that this solution is an incomplete response to concerns of justice.

Note that some concerns about discrimination are not about irrelevant criteria or inconsistent application of relevant criteria, but about a predetermined idea of persons against whom it is impermissible to discriminate. As Johnson points out,18 no one has proposed prioritising women over men for ICU admission, even where there has been evidence that men with COVID-19 are less likely to survive. The grounds for this (in balancing considerations or in social justice concerns) must be made transparent and consistent with other groups for whom discrimination is a threat.

Another disadvantage is that consistent application of outcome-maximising criteria creates a relatively homogenous treatment pool, such that poor prognosis becomes a self-fulfilling prophecy. As people with certain conditions are excluded from ICU, the opportunity for healthcare providers to refine their management of persons with these conditions is lost, along with the opportunity to improve outcomes.

We should modify critical care resource triage on the basis of considerations of justice, even at the cost of saving fewer lives

In this section, I review several possible balancing solutions that reflect a willingness to sacrifice outcome maximisation for justice considerations.

One way to give weight to social justice considerations is to refuse outcome-maximising criteria that further structural inequalities. That is, outcome-maximising criteria could be adopted only where they are neutral in distribution among categories relevant to justice, such as socioeconomic status, gender and racialised groups. Triage protocols based on life expectancy or chronic multimorbidities may fail this test in relation to concerns about health inequities, as would the use of quality-adjusted life years in relation to disabled persons. If no such criteria can be found, then we can move to random allocation by lottery or time.

Another balancing approach would be to allow outcome maximisation to outweigh social justice considerations and fundamental egalitarian concerns, but only where the distinctions we apply make a substantial difference and the categories are homogeneous enough to warrant discrimination among individuals. Note that the differences in odds of surviving critical care between groups are not 9–1.⁵ Preliminary UK data on critical care survival suggest differences of 10–14 percentage points between risk categories (including adjacent age categories).17

A third approach would be to use justice considerations to limit our choice of which outcome-maximising principle or principles to use. Some candidates are lives, life years, quality of life and fair innings. Adopting several outcome-maximising principles at the same time would further compound health inequities. For example, given a 10-to-30-year difference in life expectancy for persons of different socioeconomic status and different chronic disease patterns,30 including life years saved or quality-of-life considerations in addition to probability of survival-to-discharge in critical care triage, would doubly advantage persons who already enjoy a considerable health advantage. We should instead choose whichever single outcome-maximising principle is least detrimental to social justice and to egalitarian opportunity. Since saving the most lives limits each person’s claim to a maximum of one, it is theoretically more egalitarian than saving the most life years, a principle that allows one person’s claim to outweigh another person’s by much more.

Another approach would be to place the most weight on pure egalitarian concerns in order to give everyone the same opportunity to have their existing chance, within a reasonable range of chances. This could be operationalised through randomisation by lottery or time. The same approach could be chosen on the basis of epistemic considerations: in circumstances where judgments of individual probabilities are uncertain, risks of bias are high and/or categorical differences in possibility of benefit are small, it may be preferable to turn to random allocation.

Social justice considerations should play a role in how we operationalise random allocation. Using time to randomise (by wait lists or first-come, first-served) is more publicly acceptable than using a lottery, but it is an imperfect randomiser.⁶ Healthier persons of higher socioeconomic status are likely to be more successful in seeking care and navigating systems than disadvantaged persons, and will have more alternatives for care. It is feasible for healthcare systems and providers to make judgments about, for example, delayed care seeking, the availability of care alternatives, the burden of returning for reassessment or a limited accessibility for follow-up care. Such an approach would also tie our response to health inequities closely to dimensions in which healthcare providers and systems can support disadvantaged patients. Non-critical care resources can be called on to assist in improving outcome potential, for example, by addressing gaps in postdischarge care or supporting home care to delay or prevent admissions.19

We should modify how we pursue the goal of saving the greatest number in order to also achieve equitable outcomes

A solution that maximises both equity and efficiency without sacrificing either would be ideal. One way to address the limitations in the ‘avoid bias and discrimination’ approach discussed above has been proposed by Schmidt31 and is permitted by Emanuel et al 1: among those who meet the chosen threshold of likelihood to survive ICU and organ support, we could prioritise disadvantaged persons. We would then (in theory) save the same number of lives but save more lives of disadvantaged persons. This would be a form of affirmative action in medical resource allocation. Proponents point out that we may be able to save even more lives of the disadvantaged than we would on the balancing solutions canvassed above. However, this proposal faces two normative challenges from quite different perspectives.

First, if we take this approach, we will fail to remedy social injustice. If we select healthy disadvantaged persons, ones whose health is not detrimentally affected by the social determinants of health, then we will not interfere with the way that maximising outcomes compounds the health inequities caused by socioeconomic or racialised inequalities. We will save only those disadvantaged persons who have escaped the health effects of their disadvantage. This can be described as awarding resources to those who share a feature that is proxy for health injustice but is not itself health injustice.

Second, this approach violates competing principles of domain-specific fair distribution. If we take an affirmative action approach, we attract the objection that access to medical care should be based on medical, not social, criteria.⁷ We would also face feasibility challenges that support concerns about domain-specific fairness, even if we reject the controversial idea that appropriate allocation principles flow from the nature of what is being allocated. These include the preparation of healthcare providers and healthcare systems to discern the categories of social disadvantage in question.⁸ Data on race and ethnicity but not class are routinely collected in some health systems, such as in the USA, and not in others, such as Canada. These data are collected for monitoring health system performance and enabling epidemiological surveillance. They are unlikely to have the accuracy sufficient to inform life and death decisions for individuals. Consider the use of postal codes to support judgments of socioeconomic status based on neighbourhood income, which would privilege gentrifiers. If the role of racial, ethnic or socioeconomic markers in affirmative action were known, they would be subject to manipulation. Asking healthcare providers to distribute resources based on social inequities may increase rather than mitigate vulnerability to discrimination. Expecting people to triage against their own class identities may not be feasible. Addressing procedural justice concerns would readily swamp the efficiency gains that the proposal otherwise promises.

Furthermore, the proposal would achieve its goal only where triage protocols are categorical, such that there is a group above a threshold chance of benefit but still too large for the available resources, and further selection to maximise outcomes is not possible. (If further outcome maximising is possible using a scalar approach, but that has been rejected, then the proposal no longer demonstrates true outcome maximising.2) Empirically, this group must contain enough disadvantaged persons such that preferentially selecting these persons will address the fundamental concerns of social justice raised by medical criteria that are designed to maximise outcomes.

Proponents of affirmative action argue that it could achieve even greater benefits for those worse off than could the balancing solutions I have proposed in the previous section, such that rejecting outcome maximisation and affirmative action would constitute levelling down for the disadvantaged themselves. The argument is that admitting all of the disadvantaged persons who have escaped the health detriments associated with disadvantage could result in admitting more disadvantaged persons than would be admitted under a lottery that includes a broader group of both advantaged and disadvantaged persons who are on average less likely to survive.

However, my normative critique of outcome-maximising affirmative action would still apply under this so-called levelling down scenario. On this proposal, we select those who have not experienced the health detriments of the social determinants of health. This achieves, at best, an ambiguous case of levelling up. Its application in disability cases, for example, involves playing the interests of the ‘healthy disabled’ against those whose disabilities imply greater health detriments or shorter life expectancy. Insofar as it does this, it weakens the claim that those with greater health vulnerabilities or shorter life expectancy should see the persons who are preferentially selected as levelling up for a group with which they identify.

Because only those with fewer health detriments are treated, the problem of the self-fulfilling prophecy still applies. And like the ‘avoid bias and discrimination’ approach, it does not address the egalitarian sense that differences in probabilities of survival must be substantial to override a fundamental commitment to human equality.