A physician is obligated to consider more than a diseased organ, more even than the whole man — he must view the man in his world.
- Harvey Cushing
In 2020, I designed a small online survey called “Distributive Justice in Healthcare”, inspired by the research by Ubel et al [1]. The idea was born out of the early COVID-19 pandemic — a time when ventilators, ICU beds, and even basic protective equipment became scarce overnight. Suddenly, triage became the norm, and difficult decisions had to be made about who gets access to these resources. I found myself wondering how I would make such decisions if I were in charge, which got me into the medical ethics literature.Be careful, when you get into practice, to cultivate equally well your hearts and your heads.
- William Osler
Distribution of scarce medical resources is most relevant in the context of organ transplants. The triage calculations that happened during COVID happen quietly in every organ transplant committee. Before anyone is even placed on the transplant list, they go through a detailed assessment not only of their medical condition but also of their behavioral and psychosocial factors. Such assessments raise two major problems: first, which questions to ask, and second, how to quantify the answers in order to rank patients on a list — because ultimately, that's what needs to happen.
This is essentially the ancient philosophical problem of comparing the value of different lives, which modern medicine attempts to quantify through quality-adjusted life years (and it sounds like a joke). But poor formulation doesn't make the question disappear. First, you must decide what quality even means, then, whether every year counts the same, and finally, in which contexts these measurements apply. How do you compare the next five years of a child with leukemia to those of a thirty-year-old mother of two? I'm not trying to make anyone emotionally uncomfortable here. The point is to show how quickly these comparisons become far from straightforward, and how the line between clinical reasoning and moral judgment gets blurred.
In fact, I found reports of ethical transplant committees admitting the impossibility of such evaluations after pages of deliberation:
(2014) “In conclusion, the Committee found that using psycho-social criteria for transplant listing decisions was problematic and ethically challenging. Quality of life determinations were impossible.” [2]
(2021) “Nonmedical criteria are thought by some to uphold the principle of utility by selecting candidates likely to be good stewards of a donated organ, who may have better post-transplant outcomes. Yet, ethical concerns pertaining to equity and justice remain with using non-medical criteria to evaluate potential candidate.” [3]
Regardless of how impossible these evaluations seem, they still happen. As a transplant candidate, you may be asked (ridiculous) questions like [4,5]:
“How stable are your relationships with family / support system? (1 = no strong interpersonal ties or highly unstable relations; 5 = stable, committed relationships, strong family commitment.)”
“Do you understand the transplant process, follow-up requirements and medications needed after surgery?”
Based on your answers, your “inability to understand relevant information and poor receptiveness to education” or “absence of adequate living environment – or reluctance to relocate to a more appropriate housing environment” might classify you as high-risk. These are quoted verbatim from actual assessment criteria [4,5].
It's hard to know where to stand on these questions. On one hand, no one wants to see an organ go to waste, so you gotta figure out who will make the best use of it. But having stable family relationships or having the capacity to grasp the transplant process (not in complex medical details, but still, might require more than you think), often depends on life circumstances, rather than personal responsibility. Why should you score lower on a psychosocial assessment because you were born into a toxic family and chose to cut ties with them?
The point of my survey was to see how people would distribute scarce medical resources when faced with extremely simplified scenarios. It worked as follows: each participant read short clinical stories where 200 patients suffered from the same disease, but were divided into two groups of 100 patients based on lifestyles that may or may not have caused their condition. The first group (Group #1) always had a hereditary or maternally transmitted cause — hemochromatosis for liver failure, A1AT deficiency for lung failure, or maternal transmission of HIV. Their lifestyle played no role in their disease, meaning they bore no responsibility for their medical problem. They also never showed any stigmatized behavior like alcohol use, smoking, or drug use. The second group (Group #2) was the stigmatized group. They had a history of either smoking, drinking, or drug use, though they recently gave up these habits to be eligible for treatment. Depending on the scenario, their past lifestyle might or might not have caused their disease. In some cases, their medical condition was self-inflicted — alcohol use led to liver failure, smoking led to lung failure, or sharing needles resulted in HIV. In other cases, their stigma was non-causal, meaning it was disconnected from the disease — they could be former smokers who contracted HIV from their mothers.
Depending on their condition, these patients needed either a liver transplant, lung transplant, or HIV treatment; otherwise they would die in one year.
Prognosis after treatment was also a varying factor: Group #1 could have better, equal to, or worse prognosis than Group #2.
In total, this gave us six distinct scenario types where we permute whether Group #2’s stigmatized behavior caused their disease or not, with whether their prognosis was better, equal to, or worse than Group #1.
The dilemma was the following: you had 200 patients in total (Group #1 + Group #2), but only 100 organs or treatment slots were available. As a participant, you could choose to allocate all, more, or equal resources to either group — and then explain why, either by choosing from a list of suggested reasons or writing your own. The suggestions covered various reasons conditioned on the scenario and the chosen allocation strategy. You can find some example questions here.
Through emotional blackmail and relentless nagging, I forced 75 people to complete the survey. To my surprise, a considerable number of people took the time to write detailed explanations of their reasoning. I want to thank everyone who participated, and I apologize for taking five freakin’ years to analyze the results (#postdocking 🫠).
Now, to the results.
The subtle difference between distributing randomly vs. equally
The alluvial diagram below traces each participants' allocation strategies across six different scenarios, ordered from left to right by increasing moral "push" against Group #2. We start with the easiest case for sympathy — Group #2 has a hereditary disease (not their fault) and better prognosis (they'll live longer with treatment). As we move rightward, the scenarios progressively challenge participants: equal prognosis removes the utilitarian advantage, then worse prognosis reverses it. The right half repeats the same progression but now Group #2's stigmatized behavior causes their disease, adding moral culpability to the equation. Each colored band represents an allocation strategy, and each ribbon tracks one participant's trajectory of strategies across scenarios.
If you chose to split half-half, there were in total 4 things you could pick as your reason: (1) Both groups have an equal chance of surviving 5 years after transplantation; (2) All human life is equally valuable; (3) Medical decisions should be made only upon the need of the patients, and should not depend on any other factors; (4) I would like to write my own reason.
Option (1) only applies to the scenarios where prognoses of both groups are identical, and in those cases, 47% of participants chose the utilitarian option of allocating based on equal prognosis, and 9% wrote down their own reasons. The remaining 44% was almost equally divided between options (2) and (3). Similarly for unequal prognosis scenarios, 37% chose option (2) and 46% chose option (3), with a good 17% writing down their own reasons. A few representative entries from both scenarios are below.
"I would randomise it. I think it is the only 'fair' way." (Group #2 causal stigma, better prognosis)
"Randomise it. It would be fair." (Group #2 non-causal stigma, worse prognosis)
“People's names should be put in a bowl and be picked randomly which approx. be %50 to %50. We don't have the right to choose.” (Group #2 causal stigma, equal prognosis)
“I would actually randomly distribute resources after pooling both groups.” (Group #2 non-causal stigma, equal prognosis)
What confused me here was why these participants didn't simply choose the options (2) or (3), and instead took the time to emphasize allocating randomly instead of allocating equally. In the latter, you first need to make a conscious decision about splitting the resources half-half, and then decide who gets that resource within each group. But pooling people together and picking names out of a bowl is different. It might approximately distribute the resources half-half if your population size is big enough, but it doesn't require you to decide on anything. You're not splitting resources between groups — you're refusing to see groups at all.
So does 'random' reflect a genuine belief in fairness through chance? Or do we need randomness because we believe we don't have the right to choose?
Maybe some of us turn to randomness to avoid responsibility – in the end, it allows us to circumvent the very decision the survey asked us to make in a seemingly-ethical way.
Negotiating the equal split
Below are more verbatim examples of participants’ reasoning when Group #2 had a better prognosis but a causal stigma – a scenario designed to create maximum ethical tension.
“Just looking at the survival chances, it would make sense to allocate everything to the drug users. However, I could not be OK with treating only people who are in a sense "responsible" for their disease and not treating the "innocent" ones.”
“Heroin consumers had the choice to do it, whereas the other group didn’t. This counts negatively for them. But the chances of successful treatment are higher, which counts positive for them. So at the end I count as equal.”
“It is unfair to deny drug addicts treatment if they have a higher rate of survival, but then it is unfair to deny the congenitally infected because it simply was not their choice.”
“One group made bad decisions but on the other hand they have better chances of survival.”
I think the last one sums it up pretty well. These participants are not avoiding the ethical tension by offloading it onto randomness. They're explicitly negotiating culpability against survival and finding them equal, concluding +1−1=0. This shows how difficult it is to truly quantify opposing determinants — and how the easiest solution is to reduce them to simple binaries that can cancel each other out. One example of such simple calculations is the quality-adjusted life years I mentioned above, which reflects the general tendency for simple discretized quantifications.
The relative role of prognosis vs. personal responsibility
It's clear from the results above that even an equal split can be chosen for vastly different reasons. But justifications are constructed after the choice. The next step is to look at how strategies are shaped across all scenarios based on the two determinants that define them: prognosis and stigma. Finally it's time for some mathematical modeling :)
I used a structural equation model (SEM) to model the allocation strategies as a function of prognosis and stigma. SEMs let you estimate the independent contribution of multiple factors simultaneously, accounting for how they might interact. The point of this model is to quantify how much each factor—prognosis difference, stigma being causal or non-causal, and their interaction—shifts participants' allocation strategy.
The model is specified as:
\begin{align} \mathtt{allocation\_level} = & \beta_0 + \beta_1(\mathtt{prognosis\_gap}) + \beta_2(\mathtt{stigma\_causal}) \\& + \beta_3(\mathtt{prognosis\_gap} \times \mathtt{stigma\_causal}) \end{align} where, \begin{align} \mathtt{allocation\_level} = \begin{cases} -2 & \text{if strategy is "}\mathbf{All}\text{ to Group }\mathbf{\#1}\text{"} \\ -1 & \text{if strategy is "}\mathbf{More}\text{ to Group }\mathbf{\#1}\text{"} \\ 0 & \text{if strategy is "}\mathbf{Equal}\text{ split"} \\ +1 & \text{if strategy is "}\mathbf{More}\text{ to Group }\mathbf{\#2}\text{"} \\ +2 & \text{if strategy is "}\mathbf{All}\text{ to Group }\mathbf{\#2}\text{"} \end{cases} \end{align} \begin{align} \mathtt{prognosis\_gap} = \begin{cases} -1 & \text{if Group #2 has }\mathbf{worse}\text{ prognosis} \\ 0 & \text{if both groups have }\mathbf{equal}\text{ prognoses} \\ +1 & \text{if Group #2 has }\mathbf{better}\text{ prognosis} \end{cases} \end{align} \begin{align} \mathtt{stigma\_causal} = \begin{cases} -1 & \text{if Group #2's stigma }\mathbf{is\ the\ cause}\text{ of their medical condition} \\ +1 & \text{if Group #2's stigma }\mathbf{is\ irrelevant}\text{ to their medical condition} \end{cases} \end{align} The coding for \(\mathtt{prognosis\_gap}\) and \(\mathtt{stigma\_causal}\) is deliberately chosen to mirror the moral negotiation we saw earlier: when Group #2 has a survival advantage through better prognosis (+1) but a moral disadvantage through causal stigma (-1) (or vice versa), these should cancel each other out, if people are truly balancing competing values in a binary fashion.
Below are the results of parameter estimation.
| Parameter | Estimate | SE | p | Std. β | Interpretation |
|---|---|---|---|---|---|
| \(\beta_1\) | +0.63 | 0.05 | <0.001 | +0.50 | Better prognosis for Group #2 → more allocation to Group #2 |
| \(\beta_2\) | +0.16 | 0.04 | <0.001 | +0.15 | Non-causal stigma for Group #2 → more allocation to Group #2 |
| \(\beta_3\) | -0.03 | 0.05 | 0.633 | -0.02 | No meaningful interaction |
| \(\beta_0\) | -0.29 | 0.04 | <0.001 | -0.28 | Baseline bias for favoring Group #1 |
Prognosis was the dominant force: improving Group #2's chance of survival by one level shifted allocation by 0.63 units toward that group \((\beta_1 = +0.63, p < 0.001)\). This was a strong utilitarian signal — people cared significantly about the future use of the allocated organ or treatment.
But personal responsibility had an influence too. When Group #2's stigma was non-causal rather than the reason for their medical condition, participants allocated 0.16 units more to them \((\beta_2 = +0.16, p < 0.001)\). The effect was significant, but roughly one-quarter the size of the prognosis effect. So moral blame shaped the allocation strategy, though not as strongly as survival did.
Interestingly, these two forces influenced the strategies independently. The interaction term was negligible in magnitude and non-significant \((\beta_3 = -0.03, p=0.633)\), meaning the prognosis effect remained constant regardless of whether stigma was causal (or vice versa). People didn't abandon utilitarian logic when moral blame entered the picture. In fact, the alluvial diagram below clearly demonstrates this utilitarianism: parts of trajectories in red show how almost all people switched strategies from “All to Group #1” to directly the opposite, “All to Group #2", when prognosis for Group #2 got better despite the stigma becoming causal.
More interestingly, even in perfectly neutral scenarios — equal prognosis, non-causal stigma — there remained a significant baseline bias for favoring Group #1 \((\beta_0 = -0.29, p < 0.001)\). This hints at the fact that non-causal stigma also influences our decisions.
Does non-causal stigma play a role?
The short answer is yes, and more than we might like to admit.
As discussed above, SEM results suggest a bias toward favoring Group #1 even when Group #2's stigma was not the cause of their medical condition. Looking closer at participants who favored Group #1 despite Group #2's stigma being non-causal (over all variants of prognosis differences), 25% of those allocating all resources to Group #1 and 28% of those allocating more resources to Group #1 explicitly chose the following reason:
"Group #2 are past [drug addicts/alcoholics/smokers] and they deserve treatment less than people who do not have destructive habits."
In fact, one participant was honest enough to write down their own answer:
“I don't like or trust drug addicts."
—even when Group #2's stigma was non-causal and their prognosis was better!
Long story short, we judge people by their past, even when it’s irrelevant to their disease.
Effect of prognosis in decision-making: utilitarians vs. egalitarians
When I was designing the multiple-choice reasons for allocation strategies, I included the option "Group x has a higher chance of surviving the treatment" only when Group x had better prognosis. It never occurred to me that I should include the same reasoning for cases when Group x had worse prognosis. But seems like I should have, because a few participants made that counterintuitive choice: they allocated more resources to the group with worse prognosis. And because I hadn't included that reasoning as an option, they had to write it down themselves:
"Group 2 has more chances to survive (80% vs 60%), thus I would allocate more funding to Group 1 to give a chance to more people." (Group #2 causal stigma, better prognosis)
"Since Group 1 has less chance to survive I would increase the chance to revive more people there." (Group #2 non-causal stigma, better prognosis)
"All people are equal to have the treatment. Since Group 1 has less chance to survive, I would increase the chance of having more surviving people in that group." (Group #2 non-causal stigma, better prognosis)
These participants interpreted prognosis differently. Rather than seeing it as a measure of expected patient-years saved, they interpreted it as a measure of inequality between two groups. So they were not utilitarian, but egalitarian (or perhaps equitarian). They interpreted worse prognosis as a moral imbalance that warranted correction rather than a measure of treatment inefficiency.
The harsh truth about hereditary diseases: they will go on to be inherited
Again, another reasoning I hadn't anticipated when designing the survey... I never thought participants would think about population genetics when deciding who deserves medical resources — but two did.
The first was rather blunt:
"Natural selection" (Group #2 causal stigma, equal prognosis, strategy: All to Group #2)
The second was more elaborate and discussed how unequal allocation could perpetuate the disease at the population level — and despite Group #2's causal stigma and worse prognosis, chose the 'Equal split' strategy rather than favoring Group #1.
Both participants were reasoning about hereditary disease not just as a medical condition affecting individuals, but as a genetic trait that propagates through generations. If we preferentially treat those with hereditary conditions, we may inadvertently allow those genes to persist, creating a larger burden of disease in future generations.
This borders on the logic of eugenics, though there is a vast difference between actively engineering a population’s genetics and reasoning about the long-term consequences of treatment allocation. Neither participant used that language, and I won’t impose it on them. Still, it illustrates how easily utilitarian reasoning about public health can drift into questions that have historically been associated with troubling policies.
Fifty shades of fairness
Fair meant maximizing survival. Fair meant treating everyone equally regardless of survival. Fair meant making use of advantage. Fair meant compensating for disadvantage. Fair meant holding people accountable for their bad choices. Fair meant refusing to hold people accountable because who are we to judge? Fair meant random selection because chance feels like the only truly fair way — or because sometimes choosing itself is a burden. Fair meant negotiation, adding and subtracting competing determinants. Fair meant thinking about future generations rather than those alive today. In the end, everybody thought they were being fair to some degree — either to the scarce resource, to the patients, or to future generations — when they picked their strategies.
Harvey Cushing reminded us to view "the man in his world", a principle that should hold even when that man is born into difficult circumstances. William Osler urged us to cultivate both hearts and heads, even when one can be egalitarian at heart but utilitarian in head. Yet this survey shows how we don't agree on how much the man's world matters, or how to balance heart and head when they pull in opposite directions.
Maybe that's why so many people reached for randomness in the end. It seems like the only option that lets us feel fair without having to define it.
References
[1] Ubel, P.A., Jepson, C., Baron, J., Mohr, T., McMorrow, S. and Asch, D.A., 2001. Allocation of transplantable organs: do people want to punish patients for causing their illness?. Liver Transplantation, 7(7), pp.600-607.
[2] Community Ethics Committee, 2014. Organ Transplant Recipient Listing Criteria.
[3] Procurement, O. and Network, T., 2015. General considerations in assessment for transplant candidacy. US Department of Health & Human Services. Published online.
[4] Harashima, S., Yoneda, R., Horie, T., Fujioka, Y., Nakamura, F., Kurokawa, M. and Yoshiuchi, K., 2019. Psychosocial Assessment of Candidates for Transplantation scale (PACT) and survival after allogeneic hematopoietic stem cell transplantation. Bone Marrow Transplantation, 54(7), pp.1013-1021.
[5] Maldonado, J.R., Dubois, H.C., David, E.E., Sher, Y., Lolak, S., Dyal, J. and Witten, D., 2012. The Stanford Integrated Psychosocial Assessment for Transplantation (SIPAT): a new tool for the psychosocial evaluation of pre-transplant candidates. Psychosomatics, 53(2), pp.123-132.
