Can White Doctors Treat Black Patients?
Or: Against media hype
Note: this post was recently picked up in an article written by Katie Herzog for Bari Weiss’ substack. While I have included many disclaimers throughout this post, I want to be clear that I am 1) not a medical researcher, and 2) that you should read the paper yourself. The research itself has merit, and the biggest issue here is the way in which uncritical reporters overhype findings to satisfy their personal biases. The solution isn’t to throw our hands in the air and cry foul when research is flawed, but to build on that research in future work.
A recent article from David Hill titled “Does Being White Make Me a Worse Doctor for Black Children?” was published on BeingWell — A Medium publication for which I am an author. I found myself disagreeing with the conclusions of Dr. Hill and felt I should articulate my position on a platform that doesn’t limit me to 280 characters.
Before moving on, I strongly recommend reading Dr. Hill’s article so you know for yourself if I’m making a good-faith characterization of his arguments. You can find the article below:
A central point of Dr. Hill’s article is based on a recent PNAS paper titled “Physician–patient racial concordance and disparities in birthing mortality for newborns”. I don’t like academic jargon, so I’ll simplify the title a bit for you: “Do infants have better survival rates when birthed by a doctor of the same race?” The conclusions in this article are pretty stunning. If the authors are to be believed, black infants were 58% more likely to survive when their doctors were also black instead of white. This is a crazy result, and if true, would have disturbing implications — how many black babies have died unnecessarily as a result of poor racial concordance?
The first thing I ask myself is: ok, but is the sample size small? Small samples sizes can show crazy results by having a couple outlier data points. Nope, this study was done on 1.8 million births, so I can’t really criticize an overly-small sample size.
The next thing I wonder is if the effect is truly important. A 58% (relative) increase in mortality is pretty big, so it’s not as if the authors are using a large number of data points to find a tiny effect.
Finally, I have to ask myself: do I agree with how the authors conducted their study? No matter the effect size, population, or other fancy statistics, if the methodology is flawed there’s good reason to remain skeptical. It’s here that I find I have significant issues with how the authors chose to present their findings.
Before we dive into the methodology, a disclaimer: Racial bias exists and can manifest in a variety of ways. To criticize an individual paper on racial bias is not the same as dismissing racial bias broadly, or even specific to medicine. Even if this paper “raises awareness” about racial bias in medicine, it’s still false and shouldn’t be given any slack on those grounds.
Now for the methodology:
Physician-Patient Racial Concordance
The authors use the State of Florida’s Agency for Healthcare Administration data on births between 1992 and 2015. This data, notably, contains details about the race of patients but not of doctors. This seems like a pretty big deal! If the paper is meant to test the effect of racial concordance, I’d hope the authors knew what the race of the doctor was.
Here’s how the authors got around this problem: They used the doctors’ names and searched them up on various health websites like vitals.com and healthgrades.com. We can be pretty sure these really are the doctors in the dataset, because the researchers matched doctors’ license numbers, and affiliations. They then downloaded photos of every doctor. Here’s where bias may start to creep in: of the 9,992 doctors in their sample, only 8,045 had readily available photos. The remaining 2,000 or so of the sample had their names handed over to research assistants who then used these names and functionally cyber-stalked these physicians to find online photos. No mention in the appendix what happens to doctors that had names in the dataset but no photos online.
I have some issues with this — how do we look at photos and go “yep, they’re black!” Race itself is a fuzzy concept, and there’s a reason terms like “white passing” exist. The determining factor for whether or not the authors appropriately categorized the race of physicians is not based on the self identification of physicians, but on the consensus of the authors’ research assistants. If the hypothesis is that black doctors have more empathy or care more about black infants, thus conferring a higher level of care, then it seems important to know the experiences and self-identifications of the doctors being analyzed. To build intuition: have you ever thought a white person was middle eastern? Latino/Hispanic populations aren’t necessarily a distinct “race” from whites yet have their own category and the two could be confused with each other. This process is extraordinarily subjective. I don’t like that.
The authors are then left with a dataset of patient outcomes, patient race, and physician race. Here’s where we get into some important methodological critiques.
Before continuing, I’d like to point out that I owe an intellectual debt to Dr. Vinay Prasad for his excellent twitter thread detailing his critique of the study. Many of my points below were inspired by or taken wholesale from his own analysis.
- The researchers assume physician-patient pairing is quasi-randomized — or that which doctors patients get are not “chosen”, but dictated by the doctor who happens to be on call. Is this justified? The authors don’t get into it. It could be the case, for example, that wealthier black mothers choose black doctors at higher rates — and their wealth is a predictor of mortality in and of itself. Again, I don’t know what’s true here, and the author’s don’t do much to justify their claims
- Mortality of infants is linked to their attending physician. Medicine is a team sport, and there are teams of nurses, other physicians, and administrators who all play a role in the outcome of a patient. Do we have good evidence that an individual physician has such an outsize impact on their team?
- Do we trust that the physician listed is even the person with primary responsibility for patient care? One doctor in Vinay’s thread says that “one hospital where I worked all infants had the medical director listed. Another it was the admitting doc. Could be paediatrician after discharge.” In other words, we have no guarantee that the patient of record the authors used was even the person who looked after the birth. This isn’t something the authors can really control for! They simply don’t know.
- Finally, the biggest critique: are there systemic differences in which doctors look after infants with higher mortality risk? Is it the case, for example, that ICU pediatricians are disproportionately white? In this case, we’d expect white doctors to on average have higher mortality, because they’re on average looking at tougher cases. Selection bias matters!
So we’re left with a paper that has an extraordinary result, but questionable methodology that makes me extremely skeptical of its conclusions.
But really, can we expect the media to spend hours scrawling through the appendix and text of the paper? Surely they are experts at critical appraisal. Just kidding:
Even Fox News is getting in on the racial disparity action! I’ll note that this is not unique to this paper. The media tends to overhype scientific studies at a blistering rate — just look at their reporting on COVID-19 studies.
Here’s my (modest) proposal for a restatement of these headlines:
Black infants have higher mortality rates when their hospital records list as their physician of record (not necessarily the physician who treats a patient) a person that a team of research physicians found pictures of online and labelled as “white”.
Not nearly as sexy, eh?
Additionally, I’d like the point out that the researchers didn’t use their data to look at disparities across other racial groups. As far as I can tell, the only thing they looked at was black patients and black infants. Suppose we found that black infants had worse outcomes with middle eastern or east asian doctors as well? Would we be as quick as to leap towards a racial bias narrative? What if we found that white infants had worse outcomes when treated by south asian doctors?
This is just one study Dr. Hill cites, but is certainly a core piece, accounting for nearly half the word count of his article. He also cites other metrics like the implicit association test, which is itself a highly questionable method of getting at racial bias. I have little doubt that racial bias in medicine exists — I’d bet a lot of money it does — but in this specific instance, I think the claims being made are overblown.
Can White Doctors Treat Black Patients?
Yes. Next question.
I am always dumbfounded by people asking this sort of question. There’s a pernicious sort of racial essentialism that pops up on Twitter and other dunk-friendly social media: White doctors can’t treat black patients, white researchers can’t understand indigenous research subjects, [X majority group] can’t do [Y thing related to a minority group]. As if the property of having a racial/gender/sexuality group affiliation necessarily changes behaviour. A white doctor or black doctor is perfectly capable of feeling empathy for those who don’t share their particular skin tone and to suggest otherwise strikes me as needlessly essentialist.
This focus on individual sins or biases seems to ignore the vast structural problems that cause a lot of inequities. For example, black people in general have worse health outcomes because blackness correlates with lower socioeconomic status. One paper from Annals of the New York Academy of Sciences highlights some of the ways these two traits interact:
- “compared to whites, college‐educated blacks are more likely to experience unemployment”
- “employed blacks are more likely to be exposed to occupational hazards and carcinogens even after adjusting for job experience and education”
- “76% of African American children and 69% of Latino children live under worse conditions than the worst off white children”
- “ the conditions created by concentrated poverty and segregation make it more difficult for residents to adhere to good health practices”
- “The higher cost, poorer quality, and lower availability of healthy foods in economically disadvantaged neighborhoods can lead to poor nutrition”
- “The heavy targeting of disadvantaged minority communities with advertising for tobacco and alcohol can encourage the use of these products”
The paper is full of example of how racial group membership is associated with a host of environmental challenges. In short, black people appear to be starting from a profound disadvantage when it comes to access to good health information and quality healthcare. This is also not by choice — remember that it’s only been 56 years since the abolishment of segregation in law, and it seems obvious that racist attitudes, policies, and institutions both present and historical have materially worsened the conditions of black Americans.
These structural inequalities are not the fault of any one person, and ascribing differences in outcomes to the biases of individuals is missing the point of what constitutes structural oppression and/or bias.
We’re living in a time when important conversations are being had about racial bias, particularly focused on the anti-black bias of whites — I think it’s uncontroversial to say this has become a very easy narrative to fall into, and from what I’ve seen of media coverage, this is exactly what’s happened.
I don’t know Dr. Hill and this is the first time we’ll have interacted, through however many degrees of separation. However, I feel comfortable saying that I expect him to be fully capable of treating patients of different races from him in the same way I’d wager he’s able to treat people of different sexualities, nationalities, or political leanings.
