Race has long been a factor in how doctors approach diagnoses— removing it has proved a challenge. Katie Palmer, Health Tech Correspondent for Stat News, joins host Krys Boyd to discuss the bias baked into medicine for decades, how it contributes to system disparities, and why the work to change it is so difficult. Her series “Embedded Bias” is written with co-author Usha Lee McFarling.
- +
Transcript
Krys Boyd [00:00:00] When somebody needs a kidney transplant and they don’t have a matching donor lined up, they join a waiting list. We understand those lists might prioritize people depending on the state of their health, But it doesn’t seem right that among the criteria that might play into when or even if someone ever qualifies for a donated organ is the color of their skin. From Kera in Dallas, this is Think. I’m Krys Boyd. What is especially insidious about disparities in outcomes based on a patient’s racial identity is that they can happen regardless of whether the professionals caring for those patients hold racist beliefs. Often they are baked into diagnostic formulas designed to help physicians choose the proper course of treatment, literally taught to young doctors in training and now perpetuated potentially by algorithms that were supposed to lead to better care, including sparing patients unnecessary treatments. Journalist Katie Palmer has been looking into this for Stat News, where she’s a health tech correspondent. Her series, written in collaboration with National science Correspondent. Usha Lee McFarling, is called “Embedded Bias.” Katie, welcome to Think.
Katie Palmer [00:01:07] Thanks, Krys.
Krys Boyd [00:01:09] Let’s start with something that sounds like it would be a relatively routine thing to diagnose, which is urinary tract infections in children. Doctors at some hospitals use these like mathematical formulas to work in some ways, I guess like a magazine quiz, right at a point for this, subtract a point for that. And a particular numerical score suggests a certain standard of care. Will you tell us a little bit about that?
Katie Palmer [00:01:34] Sure. I mean, for several years, starting in 2011, the American Academy of Pediatrics has included race as part of its guidelines for treating a urinary tract sorry, not treating, but diagnosing urinary tract infections. And really the youngest children, these are kids under two. And these are tricky to figure out because you can’t just test the urine. You can’t just get a sample normally by telling somebody to go to a bathroom. These are babies. So in order to get that specimen, you need to categorize the kid, which is no small feat with a young child. So you really want to make sure that the kid is at actual risk for UTI if you’re going to go to the trouble of getting that urinary specimen. So these calculators were designed to help doctors pick the highest risk patients to meet that bar. Black patients, curiously in the data, seem to have fewer urinary tract infections across the board. These are based on just epidemiological studies. And so that was incorporated as part of the risk factors.
Krys Boyd [00:02:31] So why would the fact of a child’s race or any patient’s race, specifically black versus other racial identities, possibly be part of this calculus? Does anybody have have a reason why these folks might be less likely to be diagnosed with UTIs than other folks?
Katie Palmer [00:02:49] That is the tricky part. Usually the epidemiological data doesn’t say why race appears to have an impact on disease risk. But doctors and researchers have had different ideas. You know, they want to have a biological justification for everything that they do. So in this case, some doctors are like, maybe is there this there’s this thing called sticky epithelium that makes white kids more likely to have a urinary tract infection. But really, there’s no true biological reason that we know of for sure that would make black kids less likely to have urinary tract infections than white kids.
Krys Boyd [00:03:23] So what is the practical effect of this race? Question In these diagnostic tools on clinical practice for these pediatric patients suspected of having UTI?
Katie Palmer [00:03:34] On the whole, it seems like it is likely to underdiagnosed a certain number of black kids. It’s really tough to say for sure because you can’t really prove a negative. You can’t show that a UTI that went undiagnosed was there and that you missed it. But that’s the concern with a lot of these race based calculators that you’re either overtreating or under treating patients on account of their race without a biological justification. And that could contribute to or perpetuate existing health disparities.
Krys Boyd [00:04:05] Can failing to catch and treat these infections early in life have repercussions later on?
Katie Palmer [00:04:12] Certainly for young kids, you know, having a missed urinary tract infection can lead to renal scarring, which can lead ultimately into kidney failure down the line.
Krys Boyd [00:04:21] Okay. So this UTI calculator did receive some pushback from doctors at Boston Children’s Hospital a few years ago. How did the doctors respond or how did the hospital respond, rather.
Katie Palmer [00:04:33] Yeah, in this case, this is one of the earlier pushbacks that we started to see against the use of race in clinical calculators that started at the hospital level, often with the individual practitioners who were like. There’s something a little funny about this. I’m just not comfortable incorporating a patient’s race in these determinations. So pediatricians who brought up the issue sent a memo to leaders at Boston Children’s, and that, in fact, sparked a larger inquiry looking at all of the clinical pathways and algorithms that the hospital had come up with. And they did a whole reassessment of their many calculators, and they found eight that used race, ethnicity or ancestry in ways that they thought deserved addressing. So some of them were removed. Some of them were modified. And today, Boston Children’s doesn’t have race in those eight guidelines, but there are still many, many calculators used across different clinical specialties that incorporate race in just the same way.
Krys Boyd [00:05:29] Yeah, I had no idea. So some may be specific to a particular institution that says this is our set of, you know, diagnostic tools that we use. There may be specialties that have recommendations. This is race is used as a diagnostic factor in a lot of different medical situations, isn’t it?
Katie Palmer [00:05:48] Yeah, we found well, not we. But researchers looking into this have found that there still dozens of the sort of broad specialty society driven and recommended tools where race is still included. And that doesn’t account for the hundreds and hundreds of other algorithms that are used at the local level by hospitals like Boston Children’s that incorporate the same sort of concepts into their sort of homegrown algorithms.
Krys Boyd [00:06:12] Okay. So people might be asking why we have anything like this. How are diagnostic algorithms intended to improve doctors care decisions?
Katie Palmer [00:06:22] Yeah, I mean, race and ethnicity have been kept better track of in clinical health research in order to better understand the disparities that exist between different populations. And that’s critically important. Now, the goal for understanding these health disparities is that we can help to ameliorate them, hopefully through better clinical care and even sort of social follow ups of patients in the sort of 90s began this broader push to increase the diversity of clinical research in an effort to understand health disparities so we can address them. What that enabled some people believe, though, is sort of the misuse of this racial data to encourage false beliefs about biological differences between races in ways that some see manifesting in these clinical calculators that are adopted across medical specialties.
Krys Boyd [00:07:13] Okay. So I mean, reading this series, I kept thinking about kids playing a game of telephone, right? Where somebody whisper something into the ear of somebody else. And over time, mistakes happen. And pretty soon the phrase being passed on is totally different from the original. How did early racist beliefs about black people’s like skin and muscle and bone density end up being telephoned into standard medical practice later on?
Katie Palmer [00:07:40] Yeah, this is the really funny way that sort of our nation’s history of slavery sort of intersects with medical research. You know, obviously doctors are not racist, but some of these beliefs of and sort of like pre-Civil War science are still embedded in some of those biological justifications that people try to come up with to sort of fill in the blanks when research is included in these clinical calculators. So the clearest example of this is probably the EGFR calculator, which is a calculator used to estimate a patient’s kidney function. It’s based on creatinine levels in the blood, which is a byproduct of muscle breakdown. So black people typically do have higher levels of creatinine. So the EGFR calculator basically says for two patients with the same creatinine levels, the black patient has a higher level of kidney function than a white patient. The problem with that, if you trace it through the health care system, that means black patients appear to have healthier kidneys. Therefore, they’re eligible for kidney transplants at lower rates. The way that this was justified in the past is that black people were said to have higher muscle mass on average to account for that higher level of creatinine. It seemed like it might have biological plausibility. Patients who went to see their doctors as recently in the last five years, when asked to just explain the disparity in EGFR results for African-Americans and non-African-Americans, were told by their doctors, it’s because black people have greater muscle mass. But that is clearly not the case on an individual by individual basis.
Krys Boyd [00:09:16] And nobody went back and actually did. I mean, maybe later they did. But but initially, when people were using this, nobody went in and studied whether this was really something that would be relevant to individual patients.
Katie Palmer [00:09:29] No, the earliest references to the sort of muscle mass justification were based on really old research, pretty small by modern standards, shoddy studies. And the the references in the original calculator papers were by no means saying, you know, this is why the only reason why we’re seeing these creatinine level differences between African-American and non African-American patients. But it was just it made it easier to perpetuate the use of race sort of unthinkingly in the continued estimates of kidney function calculators.
Krys Boyd [00:10:04] And we should note, in case it’s not clear, initially these differences were noted and they amounted to conveniently based justifications for enslaving black people in this country.
Katie Palmer [00:10:18] In some cases, that has certainly been the case. I think we would tie this example most directly back to spirometry and lung function estimates. Samuel Cartwright was a slave owner who estimated that lung function was lower for African-Americans, and that sort of justified their continued enslavement. And essentially they needed to be worked harder to maintain their limited lung function. And until recently, the pulmonary function testing in this country similarly sort of downgraded the lung function of black Americans.
Krys Boyd [00:10:57] So in the modern world, doctors don’t necessarily have to hold a lot of racist beliefs in order to have been taught these things and sort of absorbed them as truth.
Katie Palmer [00:11:09] No, certainly not. I mean, even doctors that we spoke with who were extremely active in reducing health care disparities and are devoted to racial justice in medical practice and sort of accepted things as a matter of course, because it’s just embedded so deeply in medical education and practice.
Krys Boyd [00:11:29] And I guess the idea that race can affect individual health risk, it starts to be taught as early as medical school. Right?
Katie Palmer [00:11:39] Yeah. I mean the muscle mass justifications, sticky epithelium, lower lung function, these are all just sort of generalities that get repeated sort of as a matter of course, as somebody is going through their rounds.
Krys Boyd [00:11:55] It’s widely accepted. Now, of course, that race is a is a social construct, not a biological reality. But when medical students are taught that race might affect risk as opposed to like income or health care access or some other social determinants of health, do those students tend to assume there must be some hard and tested science back somewhere to prove all this?
Katie Palmer [00:12:19] I would assume that they do. And that’s increasingly becoming the case. You know, we’re seeing a lot of the movement to change clinical calculators, to remove or replace race coming from medical students who are hearing these messages and, you know, realizing pretty quickly that they don’t add up. That’s certainly not uniform across the board. But a lot of the momentum is coming from these young students who dig a little bit into the ends of the data, see that there’s really not much there. And then therefore question the standard practice of making these predictions in part based on a patient’s race.
Krys Boyd [00:12:55] That’s really interesting that these young students are of a generation that has been taught to not necessarily just accept information that is handed to them, but ask whether there’s real science behind.
Katie Palmer [00:13:06] And it’s led to some really big power clashes. Honestly, you’ve got young, relatively less powerful medical students who are going to the top leadership in their hospitals and their their educators and saying, hey, guys, we got to change this. And that results in some really tense conversations sometimes.
Krys Boyd [00:13:23] Katie, we will talk about the factors that might play into apparently different risks for people of different racial groups. But as they have existed until recently, how have these race questions figured into algorithms used to diagnose illnesses? Is it is typically as simple as like add two points if a patient is white or subtract two points up if a patient is black or some other race.
Katie Palmer [00:13:51] It’s a mix. Sometimes you will have strict numeric calculators like that where you add on points here and there. More commonly it’ll be sort of a percentage. You know, if a person is of a specific race, typically black is the only one that we have data for to make a reasonable difference. That will be just sort of a correction factor, you know, modify their risk down by 15% or something like that.
Krys Boyd [00:14:17] That’s really interesting, too, that in this country, any data we do think we have based on race is limited to as if there are just two racial groups.
Katie Palmer [00:14:28] Yeah, it’s not quite that. It’s that’s where the disparities come out most frequently, in part because black and Caucasian are the largest populations of data that we do have to work with. So if researchers are looking for a signal in epidemiological data, they’re likely to emerge along black and non-black lines. That’s also a result of the extraordinary structural racism in our system that leads to those biggest health care disparities existing between those groups. But it also comes down to the categories that we use to collect the data. So in clinical research, the categories that we use to collect race and ethnicity are largely copies of census categories from the OMB. So the ones you will see on any checkbox when you’re doing your entry forms, American Indian or Alaska, Native Asian, Black or African American, Native Hawaiian or other Pacific Islander and white. So those are the data that usually clinical researchers have to work with when they’re looking for disparities that may exist across racial categories.
Krys Boyd [00:15:33] How long have these researchers looking at disparities across racial categories even had these data to work with? How long have they had information about racial identity? And in the past, were they just guessing as opposed to letting someone self-identify? And, you know, how how much can be factored in based on someone who has a multiracial identity?
Katie Palmer [00:15:56] Yeah. Before the 90s, we barely had any diversity in clinical research, and that’s when the NIH started really pushing for more diversity in our clinical trials, pushing for the collection of these data so we could at least have a better sense of health care outcomes beyond white men. It wasn’t just racial and ethnic diversity. We also just had extraordinarily limited data on women in clinical research. So big push to get more diversity across the board. Again, we were only collecting data, though, using these limited categories. And even though we’ve had these policies in place for increased diversity in clinical research for a couple of decades, several decades, we still haven’t seen that much progress. We’re far, far better than we were before. But there have been recent reports that NIH policies that require a certain level of diversity in certain clinical trials haven’t met their benchmarks.
Krys Boyd [00:16:50] So maybe there are social differences that happen to correlate with race that could account for all these differences that appear between different groups. Is there evidence, though, that including these racial factors on diagnostic algorithms, feeds a perception that there is some hard and fast difference between different human beings based on their racial identity?
Katie Palmer [00:17:15] That is the concern. You know, it can be argued whether or not individual calculators that include race results in immediate harm to patients, the way that it is clear that, for example, the kidney function calculator has resulted in different access to kidney transplants across races. What we can say for sure is that the incorporation of race in clinical calculators alongside other biological variables like cholesterol level, perpetuates the idea that race is a biological construct that is meaningful in disease risk. And so that’s the reason at a high level that many clinicians and researchers are pushing back against its continued inclusion in clinical calculators.
Krys Boyd [00:17:59] Has anybody looked into whether I’m going to use the word innocent here in air quotes, but an innocent belief in some kind of biological difference between racial groups can foster discriminatory care for patients depending on their race.
Katie Palmer [00:18:15] And then that is also a reason why, you know, individuals are pushing hard for this change. Again, nobody who is advocating for the removal of race from calculators is saying that it’s not important to collect race, that it’s not important to proceed with racially sort of informed care and just extreme judgment with the patient about their lived experience. But there is the concern that it just makes it easier to see a patient as. As a race and not a person. I spoke with the chief health equity officer of the American Academy of Pediatrics, who is doing a lot of this work to investigate all of their guidelines, including the guide guideline for misuse of race. And he obviously comes with a ton of perspective as a practitioner, but he also experienced this as a patient, having a doctor who sort of in his mind over generalized his cardiovascular disease risk as a result of his black race. The calculator that we use to recommend the use of statins to reduce cardiovascular disease risk also has a racial factor. This doctor does not have a particularly high level of risk based on all of his interpersonal life factors for cardiovascular disease risk. But at its most recent appointment in one of his numbers in which is race and out popped recommendation for a statin and he did not like being treated as if he were just a black man amidst this entire sea of individuals and having his care directed differently as a result.
Krys Boyd [00:19:57] It’s a really interesting experience because I would imagine most people, you know, regardless of their race, who are not part of the health profession, wouldn’t even know to ask questions about this, wouldn’t wouldn’t even realize that their recommendations might be influenced by their racial identity.
Katie Palmer [00:20:15] Absolutely. I mean, the EGFR kidney function calculator came under a lot of scrutiny first in part because in the lab reports that displayed these results, it was one of the few calculators where you could see very clearly as a patient your results as an African-American and your results as a non-African-American. In many of these other cases, including this cardiovascular disease calculator, you’re not seeing anything directly that would let you know that your race was being incorporated.
Krys Boyd [00:20:43] And this is important, you know, based on the experience of that African-American doctor that you told us about with some risk factors for cardiovascular disease, but not significant ones. This is a situation where, like some people may receive care they don’t actually need while others are denied care they do need. And it doesn’t always necessarily follow the sort of standard lines of racial privilege.
Katie Palmer [00:21:08] Right. And that’s why there is concern from many about removing race outright or quickly or in a kneejerk fashion from these calculators, because there is a risk if you remove race, that some patients will not receive the care that their need that they need. It’s not a hard and fast rule that just pulling race out will automatically improve health outcomes across the board. It’s far more nuanced than that.
Krys Boyd [00:21:33] And of course, you know, if we’re sitting here thinking, we’ll just make sure everybody gets maximum care that comes with risk factors to giving care the people may not really need can be painful. It can be expensive. It can be alarming. It’s it’s not a road we want to go down.
Katie Palmer [00:21:50] Absolutely.
Krys Boyd [00:21:52] So how has race factored into you mentioned, you know, who qualifies for donor organs like kidneys? Is this is this a factor of black patients being perceived as somehow less ill than white patients when their health numbers are precisely the same?
Katie Palmer [00:22:11] In the case of kidney function, yes. So the EGFR for the same exact creatinine level, which is, you know, a blood test, would say that a black patient has a higher kidney function than an equivalent white patient with the same creatinine level and all the same other risk factors that are incorporated in that calculator, including age, gender and others that I can’t remember off the top of my head. So in that particular case, yes, the it has been shown in a bunch of research that patients were essentially denied entry onto the kidney transplant waiting list because as a black patient, your kidney function will appear better for longer. So we’ll take you a longer time to qualify to get on that kidney transplant list that has recently changed. So EGFR was one of the first calculators to be, you know, to have race removed, and that has now been incorporated pretty widely across the United States. And the new EGFR equation has been incorporated into the kidney transplant waiting list calculators as well.
Krys Boyd [00:23:17] Do you know how long ago that happened? I’m curious as to how much of a practical effect has been seen already.
Katie Palmer [00:23:23] So in 2021, that was the year that the EGFR race free equation was officially recommended by the nephrology societies. And then in 2022 is when the organ transplant network started to incorporate the race free equation.
Krys Boyd [00:23:40] Katie, what do we know about why it can appear that race affects susceptibility to certain illnesses?
Katie Palmer [00:23:49] This is the tricky part. You know, we know that differences in disease risk are genetically and ancestrally linked in a way that appears as a correlation with race. The more empirical way to design clinical calculators would be to find those underlying genes test for them and generate an individual’s risk score based on their truly individualized genetic propensity for that disease. And that’s the promise of precision medicine. We’re obviously not there yet. The other side of the coin is that the life a person lives impacts their health care outcomes, their financial security, their access to healthy food, even the stress of experiencing racism. And we know that nonwhite populations struggle with these so-called social drivers or social determinants of health more and so much more that they could be driving the racial signals that we see. So it’s really these it’s a combination of these hard and fast genetic differences that we can clearly measure. And these far more diffuse social factors that are really what race is a proxy for in these calculators.
Krys Boyd [00:24:53] So we’ve been talking about what it might take to eliminate race as a flawed and outdated health risk predictor in certain shortcut diagnostic formulas. But as you noted, some doctors are not convinced this is the right way to go. And in some cases, that’s because these physicians are especially concerned about the welfare of their patients of color.
Katie Palmer [00:25:16] Yeah. The it’s really a question over how to do no harm. Some physicians and researchers believe that the best way to do no harm is to remove race whenever it’s possible. And the other side is really concerned that if we remove race too quickly without making sure that the clinical algorithms that we produce as a result are just as accurate and can identify health issues in susceptible communities at the same level of accuracy that some of those people will go untreated and therefore be harmed in another way.
Krys Boyd [00:25:47] Do physicians think there are ways to use race as a consideration to help them achieve a correct diagnosis without race being used as like a determining factor that dictates who gets medication or surgery or a transplant? Like how can race be placed in context without allowing it to contribute to biases in actual care?
Katie Palmer [00:26:10] You know, that really comes to how it comes down to how the algorithm is designed. There was a big review at the sort of promoted a by legislators of all of the race based algorithms out there looking to see what the harms and other potential impacts of these algorithms could be. And it found that there are many examples of algorithms that could improve health disparities if they are intentionally designed to do so. So if you want to design an algorithm that helps you surface patients who are at greater need as a result of all of these social direct drivers that we find are correlated with race, then sure, you can go ahead and create those. It comes and it becomes a little trickier when you’re trying to diagnose patients based on their race. And basically it’s just extraordinarily complicated. You have to do research looking at the actual health care outcomes. And we in many cases don’t have the right research to answer those questions effectively.
Krys Boyd [00:27:14] Some people might like to see legislation that limits the use of race in these diagnostic tools, but does that threaten like special programs that have been developed to target patients who might be a particular risk based on, if not specifically, their race factors that correlate to their race?
Katie Palmer [00:27:36] I don’t know of any legislation that specifically targets the removal of race from these kinds of equations. There certainly are implications for nondiscrimination rules. The Office of Civil Rights recently issued a final rule that made it very clear that health systems should not discriminate based on the outputs of their clinical algorithms. Where that was sort of baked in before. You shouldn’t discriminate against your patients in any way. They made it very clear we’re incorporating these calculators, including the race based ones in that calculus, looking for discrimination. But the mere presence of race in a calculator does not mean that it’s being used to perpetuate discriminatory care. So this is where a lot of health systems are having a difficult time right now trying to figure out, okay, which of our race based calculators could be directly causing harm and therefore discrimination, and which ones do we really need to get ahead of and try to change quickly before this rule kicks in in May of next year?
Krys Boyd [00:28:37] Your reporting on this in this series is is very detailed and very nuanced. It seems like a lot to dive into. And I know this is what you do for a living, but can you talk a little bit about the challenges of really getting your mind around this problem and understanding all the sort of exceptions that might prove the rule?
Katie Palmer [00:28:55] Yeah, it was an extraordinarily difficult topic to dive into. I was lucky to have the the partnership of my colleague, Usha Lee McFarling who reports all the time on health disparities. I report on health technology, so I had a lot more to learn coming into this. I found the most challenging to talk about because it is it’s really difficult to talk about race and racism, period. None of the researchers and clinicians that I talked to wanted to be seen as disparaging their colleagues if they hadn’t built race based guidelines in the past. Nobody here wants to do harm, like everybody is coming from the same page there. And it was equally difficult for me as a white reporter coming into this space. You know, I was very concerned about my biases that I brought into the process and just took a lot of time making mistakes and going back and forth and getting lots of different perspectives to understand what is driving people’s, you know, vision to remove and replace race in these calculators and why some people are still really pushing on the brakes.
Krys Boyd [00:29:55] Katie, what is the social deprivation index?
Katie Palmer [00:29:59] The social deprivation index is one of many ways that we’re trying to estimate how a patient’s lived experience can impact their risk of a lot of different things. Recently, the social deprivation index was used in a calculator to estimate cardiovascular disease risk. As a crude measure, you know, it uses a patient’s zip code to estimate their position on a lot of different sort of social categories from education level to income. But it’s trying to get at those social drivers and social determinants of health that we discussed earlier to see if we can figure out how to pinpoint a patient’s disease, risk more directly on those lived experiences rather than the crude proxy of their race.
Krys Boyd [00:30:49] This is really interesting. I was thinking about my own personal zip code, which has some people that live in multi-million dollar mansions and also has a fair number of people living in like one bedroom apartments and studio apartments who who may not have very many resources at all. And it occurred to me that, you know, the same zip code might not be a great way to describe people and their risk. And then I thought, well, why would I assume that race is a better way to describe people in their risk?
Katie Palmer [00:31:18] Yeah, they’re definitely flawed in similar ways. And nobody’s saying that zip code and the resulting social deprivation index is the best way to measure these risks and incorporate them into clinical calculators. It is sort of the first stab at it, though. We simply don’t have the best individual level data. Again, we’re always looking for individual level data as much as possible to help drive meaningful disease risk estimates and predictions. We don’t have that individual level data and say, you know, what’s the food that you eat? Like how close do you have a healthy grocery store with produce? What are your transportation options? We just don’t collect that kind of data and medical records so researchers don’t have it to incorporate when they design new calculators. But we do have a zip code and we can estimate some of those factors based on community level surveys. So that’s what researchers are experimenting with first.
Krys Boyd [00:32:14] Yeah. I was going to ask you about those surveys. Is there a lot of highly geographically detailed data being collected and organized in ways that would make it useful to health care workers?
Katie Palmer [00:32:28] Really not yet. You know, similarly to using the census categories for the collection of race, a lot of the things that we’re using and we’re seeing used in clinical research about social determinants sort of piggybacks on existing structures outside of health care to collect these data. So it’s the word that a lot of researchers like to use is whether these data were “fit for purpose.” And they’re really not fit for purpose yet. But there are some efforts sort of at the federal level to incorporate the collection of more of the social data at a patient level, you know, categories that have to be included in electronic health care records and are sort of mandated in order for health care systems to get reimbursed by Medicare, essentially. So there will be more of these data in health care, health care records going forward. And the hope is that those individual level social determinants data will be able to drive more individualized predictions in the future.
Krys Boyd [00:33:29] That seems like a good step, although presumably if we get these data that are fit for purpose, as you said, it would need to be revisited and updated periodically, right, because neighborhoods change economically and demographically over time.
Katie Palmer [00:33:45] Absolutely. I mean, one of the things that gets lost in the conversation about race and clinical calculators is that algorithms in general, predictive algorithms in health care fall out of date very quickly. The prevent calculator that does not include risk and instead is incorporating some of these social drivers. It’s not just removing race. It was a broad reexamination of cardiovascular disease risk because the previous set of calculators that included race also just dramatically over predicted risk of cardiovascular disease across the board because they were designed on patient populations before statins were widely used. So constantly there’s a moving target for these algorithms in health care, and a lot of them need to be updated for reasons beyond the fact that they include race.
Krys Boyd [00:34:33] Of course, we are now beginning to use artificial intelligence to help create diagnostic algorithms. I’m happy to be talking to Stat’s Health Tech Correspondent because there’s a worry, right, that I can’t learn properly if it’s taking in previous data from models that were flawed.
Katie Palmer [00:34:52] Yeah. I mean, it’s the it’s the same problem, but a little bit on steroids with artificial intelligence because you’re dealing with data that comes directly from electronic health records instead of sort of collected in carefully run population health research trials. And you’re there’s a lot of messy, messy data in electronic health records that can result in sort of spurious correlations if you aren’t very careful about removing them before you design your algorithms.
Krys Boyd [00:35:24] Will you remind us once again, I think we’ve heard this lesson, but it never it always bears repeating the difference between correlation and causation and why it’s so bad to get those two things confused.
Katie Palmer [00:35:38] So taking this back to the UTI calculator, the we saw in the epidemiological research over and over again that white kids were at higher risk of utilities than black kids. You know, a correlation is clearly there. We see that if you’re white, you’re at a higher risk of UTI than a black patient. But that doesn’t mean that you are immediately at risk of a higher UTI because of your white race. There is no reason to believe that biologically, nor that you’re at a lower risk because of your black race. There’s nothing inherently about the color of your skin that shouldn’t have any impact on the way that bacteria take seed in your urinary tract. There might be other biological explanations that could be causative there, but we don’t know the true extent that the true reason for those differences. So it’s inappropriate to jump to say race causes different rates of urinary tract infection, only that there is a correlation between the two.
Krys Boyd [00:36:39] It would seem these shortcuts, these algorithms to assist with diagnosis might be less necessary if we could get sufficient data about each individual patient, write an extensive health background and genetic testing and information about their environment and family history and maybe income. Is it too much to hope that we might get to a place where health care professionals collect and can interpret all that information for every patient?
Katie Palmer [00:37:06] A lot of people that I talked to. You don’t think it’s too much to dream? It’s certainly not right around the corner. But we’re getting to the point where, you know, whole genome sequencing is going to be a possibility for every single patient. And we might even be able to start doing it, you know, at the point of birth. So you can start screening for for rare genetic diseases right off the bat. That gets the genetic side of things out of the way. When it comes to collecting individual like environmental factors. I think we’re a little bit further away. It’s still a challenging prospect for doctors to engage in that level of detail with their patients entire lives. But I do think we’re moving in that direction. It’s just going to be a question of how long it takes.
Krys Boyd [00:37:52] Some of these things, of course, are just going to be really hard to quantify no matter how good the information is. You mentioned that, you know, lots of studies have shown that having a lot of, like community support can have a beneficial effect on someone’s health. And, you know, knowing one’s neighbors, it’s really hard to sort of check a box there and feed that directly into a calculator, isn’t it?
Katie Palmer [00:38:14] Absolutely. I don’t think anybody’s got an easy solution for that one.
Krys Boyd [00:38:18] If some of these algorithms that have used race as a diagnostic factor are being eliminated at particular institutions, are they developing more accurate tools to replace them? What’s happening instead?
Katie Palmer [00:38:34] That is the crux of the debate. Whether or not the new algorithms are equally or more accurate than the previous ones. No health care specialty wants to move forward and create an algorithm without race that has lower accuracy. The real goal is to make sure that you’re at least matching it or doing even better. And very importantly, also that when you’re checking the accuracy of those predictions, that you measure the accuracy across demographic groups that you care about to make sure that you’re not resulting in discriminatory outcomes.
Krys Boyd [00:39:08] So just eliminating race altogether, literally, you know, no one is blind to race, but if on no one’s health records were racial categories or identifications listed, this wouldn’t necessarily be better, right? Because we might miss noticing patterns that some groups need more help than others.
Katie Palmer [00:39:31] Absolutely. Nobody is arguing that we stop collecting racial and ethnic data and medical records and research. As you say, it is necessary to understand health care disparities at the research level. It’s also important for just period. The tricky part is making sure that it is used appropriately in research going forward. So we want to make sure that the race variables in clinical calculators are not used to again, perpetuate this idea falsely, that there are biological differences between races, and then if it is used, it’s being used to reduce disparities between different demographic groups.
Krys Boyd [00:40:20] Is anybody concerned that we might somehow lose the ability to talk openly about particular health conditions that are associated with people who have a particular genetic background or can be traced to certain parts of the world? I’m thinking of something like sickle cell disease.
Katie Palmer [00:40:40] Yeah, that’s. That was a frequently cited example when we were doing our report. No, nobody is advocating that we stop using race as a way to talk about diseases that are truly, dramatically more common in certain demographic groups than others. Again, the most relevant thing here is not the race of a patient. It’s their genetic ancestry. So the ultimate goal is to get as close as possible to that genetic ancestry to derive individualized care. But we know that sickle cell is highly correlated with one ancestral group, and we want to make sure that that group gets their necessary care. So in this case, race can be sometimes a helpful, if crude proxy to help drive that medical care. That is, again, why race can sometimes be a very useful thing to include in those individual patient records. So a doctor can bring up a risk factor that they think could potentially be of interest to their patients. It’s just important that it is not applied systemically across an entire population because that’s when the broad systemic disparities can result, especially when incorporated into a clinical algorithm that spits out literally a numeric risk estimate for an individual based on an entire population’s risk of disease.
Krys Boyd [00:41:57] And can we assume we will continue to need those clinical algorithms? Because physicians are busy. They may not specialize in every, you know, health condition. I mean, how important is it that algorithms exist at all?
Katie Palmer [00:42:11] I think they’re only going to become more important. None of these algorithms are meant to do jobs for clinicians. They’re always supposed to be used as part of, you know, shared decision making with the patient. It’s supposed to deliver a spectrum of risk that a patient and a clinician can have a conversation about together. And I think that those kinds of triage tools, those kinds of, you know, supportive algorithms are only going to proliferate, especially as artificial intelligence continues to be adopted across the health care system. The tools that we’ve talked about so far are mostly pretty simple calculators. You know, the same kind of simple algebraic equations that you’d see in a math class. As we get more and more variables incorporated into the next generation of artificial intelligence algorithms. However, it’ll get a little bit harder for an individual doctor to tell exactly what’s going into an algorithmic prediction. And that’s where some of the danger exists, some people argue.
Krys Boyd [00:43:18] Are there takeaways for those of us who are just clients of the health care system from these revelations of the ways that race might shape the way we’re diagnosed and what treatments were offered?
Katie Palmer [00:43:29] I think it’s always good to engage your doctor in as much conversation as they will allow you to have in the minutes that you have with them up to maybe a whole 15 whole minutes. We did publish a database along with a story which was put together by researchers at the University of Pittsburgh demonstrating almost 50 of these calculators that currently use race. And it might just be worth a scan to see, you know, for conditions that you may have, whether there are calculators in use that incorporate your race. And if that’s news to you and you think maybe some of these algorithms have been used in the course of your care, have an immediate conversation with your doctor about that and whether it has implications for the care you’ve been receiving.
Krys Boyd [00:44:15] Katie Palmer is a Health Tech Correspondent for Stat News, which published a series she co-wrote with National Science Correspondent. Usha Lee Mcfarling. It’s called “Embedded Bias.” Katie, thanks very much for sharing your reporting on this.
Katie Palmer [00:44:28] Thank you so much for the chance to talk about it.
Krys Boyd [00:44:30] Think is distributed by PRX, the public radio exchange. You can find us on Facebook and Instagram and listen to our podcast. Wherever you get podcasts, just search for KERA Think. Our website is think.kera.org, and that’s where you can sign up for our free weekly newsletter. Once again, I’m Krys Boyd. Thanks for listening. Have a great day.
