Evangelos Oikonomou: Decoding the Hidden Signals of Heart Disease

Transcript

Harlan Krumholz: Welcome to Health & Veritas. I’m Harlan Krumholz.

Howard Forman: And I’m Howie Forman. We’re physicians and professors at Yale University. We’re trying to get closer to the truth about health and healthcare. Our guest today is Dr. Evangelos Oikonomou. But first, we like to check in on current or hot topics in health and healthcare. What do you got going today, Harlan?

Harlan Krumholz: Well, this is a topic you turned me on to last week. And I was already sort of set on talking about the diet soda stuff, but I turned back to it because it continued to get a lot of attention. So here we go. There was a study published recently in JAMA, one of the top journals, that was following more than 130,000 men and women from the Nurses’ Health Study and the Health Professional Follow-Up Study. Now, these are two very well known, very highly respected epidemiologic studies, which means that there’s people who enroll, they collect information about them and they follow them over time, and that sometimes do some intermittent specialized testing. And they’re trying to understand factors that are associated with health.

And in this case, they were focusing on dementia. Now, this study included people that had been followed up to 43 years. This is a really long-term study. And over that time there had been more than 11,000 cases of dementia that were documented. So here was a headline finding that most people noticed, people who drank caffeinated coffee had a lower risk of dementia. That was the sort of headline that most people noticed. And then the way they did it was, if they compared the highest to the lowest quartile, the people who drank the most caffeinated coffee compared to those who had the least, the hazard ratio was 0.82, which meant that there was about an 18% lower risk of dementia in those who were drinking more than those who were drinking not very much.

Interestingly, tea followed a similar pattern, but decaffeinated coffee did not. Now, this elicited a whole bunch of discussion in all sorts of fora about “Is this about caffeine? Is it about coffee? Should people who don’t like coffee start drinking it?” What was also interesting was that the dose response, how much you drank, was what they call nonlinear. It didn’t increase with the more you drank. The lowest dementia risk appeared around two to three cups of caffeinated coffee per day, and that’s roughly about 300 milligrams of caffeine. Drinking more than that was not associated with more reduction in risk. So then again, the messaging here was “caffeinated coffee, just drink two or three cups a day,” and some people even taking it so far to say everybody should be doing this in order to stave off dementia.

So let me just linger for a second on the strengths of this study. It’s a very large study with long follow-up, nearly four decades. A very famous study in the sense that lots of integrity, it’s well done, federally funded, has produced a lot of great scholarship. Diet was assessed repeatedly, every two to four years, using a validated food frequency questionnaire. And they didn’t just look at dementia; they examined subjective cognitive decline and objective cognitive testing in a subset. So meaning it wasn’t just about diagnosis of dementia, but they also had some information in some people about actual cognitive function. They had so much information they were able to take into account potential confounders, things that would confuse the relationship, like education, smoking, physical activity, BMI—body mass index—hypertension, diabetes, depression, overall diet quality. And then, like I said, they distinguished caffeinated from decaffeinated, which all this was pretty much in advance from previous studies.

Now, what about these cognitive tests? Well, higher-caffeinated coffee was associated with a slightly higher score on this telephone cognitive testing—that’s how they did it—there are ways you can sort of test people’s cognitive function without seeing them face to face. But the difference in this score, they call it “tick score,” was 0.1 points between the higher and lower groups. Now that sounds like maybe that’s a big deal, very precise, 0.11. You need to know the score ranges from 0 to 41. So 0.11 is kind of bupkis here. It’s not really a big deal. It could be highly significant because they’re studying so many people, but the actual number itself isn’t very substantial. The typical gap between someone with mild cognitive impairment and normal cognition is at least two to three points, and so that’s much bigger than this 0.11. It’s tiny. So it’s a population level signal, but not something an individual would detect. But let me just get to these limitations.

First and foremost, observational. It’s hard to say for sure that there’s a causal relationship that actually decaffeinated coffee causes a lower risk of dementia. That’s very hard to get to. The dementia itself relied mostly on self-reported physician diagnoses and death records. But this worked in these large cohorts. It’s hard to really get much deeper. But honestly, there’s still a lot of noise in this. And so there’s concerns about that kind of reporting as opposed to a protocolized identification of people. This idea about whether they’ve captured everything that’s different between coffee drinkers and non-coffee drinkers is a big deal. Coffee drinkers differ in many ways. They adjusted for a lot. But it could be that unmeasured factors like lifelong cognitive engagement or subtle health behaviors are still different and hard to pick up.

And then the decaffeinated coffee findings are interesting. You can speculate about a whole bunch of mechanisms. But it’s hard to know why it wasn’t associated with risk. It could be confounding. It could be that people start having problems, start actually changing their coffee habits, so it was what we call reverse causation. And then this cognitive testing was only done in 20%. Still a lot of people, but not so much.

So look, here’s the takeaway, I think. In this study, moderate caffeinated coffee consumption, about two to three cups a day, was associated with lower dementia, but the effect size was small and especially in the objective cognitive testing. It doesn’t prove the coffee prevents dementia, but it suggests that it’s unlikely to harm cognitive health and maybe associated with lower risk. I think a bigger question, should a journal like JAMA be publishing something like this? Because at the end of reading it, it doesn’t take you anywhere. Because whatever you believed before, you probably can still believe. If you believe that it does help, there’s some evidence here to help. If you believe it doesn’t, there’s enough holes in this study that makes it so all you can say is “We need more evidence and better evidence to say it.”

And should it be published? Absolutely. But should in a top-tier journal like JAMA? The press and the public tend to think, like, “This is a breakthrough. Something just happened that really showed us something we didn’t know before”—and that’s not what happened here. It wasn’t a breakthrough. It’s not a landmark study. It’s just another observational nutrition epidemiologic study that shows an association, and most of us don’t exactly know quite what to do with it. But if you like coffee, you can go home and feel better about it, because you can [inaudible 00:07:27].{relisten}

Howard Forman: Yeah. So I’ll just quickly say, over the last decades, the evidence that coffee or caffeine causes harm has gone down tremendously. There’s practically nothing that says coffee causes harm. There’s studies, including this one, that show coffee may have benefit for a number of things right now. And so for me, who doesn’t drink coffee and has no caffeine in my diet other than the occasional chocolate, I do wonder, should I go back to drinking coffee? And I’m happy they published it, because I think more information is always good.

Harlan Krumholz: Well, I think you’re right to encapsulate like that. I mean, if I were seeing you, I would say “I don’t think there’s enough here to change your behavior if you’re happy with how you’re living.” If you were somebody who enjoys coffee and you were worried about the adverse effects, I think you could say, like you said, “Enjoy it. It’s not a problem.” But I don’t think that it’s something that would get me to get someone to change their... I’d rather encourage people to exercise, which also mostly a lot of observational studies there, some intervention studies. But you’re just saying, I’d rather get you to focus on healthy... what I think represents healthy diet, healthy exercise, social interactions. If you really want to help yourself, I don’t think that encouraging you to drink coffee if you don’t drink it, it’s not strong enough for me to make that recommendation. All right, Howie, let’s get on to Evan. He’s going to be a great guest.

Howard Forman: Dr. Evangelos Oikonomou is a cardiologist, physician scientist, and an assistant professor in the section of cardiovascular medicine at the Yale School of Medicine, where he is also the associate director of the cardiovascular data science lab. Dr. Oikonomou’s research focuses on leveraging artificial intelligence to transform cardiac diagnostics, including ECGs—echocardiograms—and patient health records into digital biomarkers to detect signs of heart disease earlier than traditional methods, with a goal of building scalable tools that fit into clinical workflows and redefine how patients are diagnosed and cared for.

Despite being only in his first official year as a faculty member at Yale, he has an extraordinary track record of highly cited research, has been multiply honored by domestic and international societies, and funded through some of the most competitive and prestigious NIH research awards. Dr. Oikonomou graduated as valedictorian from the National and Kapodistrian University of Athens’ School of Medicine and earned his doctorate in medical sciences from the University of Oxford. He then came to Yale for the Yale Physician Scientist Training Program, through which he completed his residency in internal medicine and fellowship training in cardiovascular medicine.

So I want to just first thank you for joining us and for talking with us today. I am a radiologist and every single day I am looking at hearts on CT studies. Not to mention chest X-rays, but let’s just talk about CT studies. And I’m aware of the fact that there is an awful lot of opportunistic opportunity available, opportunistic screening available for radiologists to say more about the heart, and yet we say relatively little in general. Your research gives us a lot of direction in how we could be doing so much more without exposing patients to any more radiation and perhaps very little effort. Can you speak a little about that as just a starting point for our listeners to understand your work?

Evangelos Oikonomou: First of all, thank you so much for having me today. I’m excited to share some of our work and how we are thinking about this domain of using artificial intelligence and emerging technologies to extract more information from all the data that we acquire every day in our clinical practice. And I think the example that you just gave, I think, is a great one. There’s a lot of imaging studies that are being done, being read every day, ordered for many different reasons, and quite often not specifically to interrogate cardiovascular health or cardiovascular disease.

And much of our clinical workflow relies on simply answering the one question that was asked to us or the question that…the patient’s presenting symptom or the reason why the study’s being ordered. But in that process, we acquire so much information that is just left on the table. And perhaps quite often, many years later, we go back and we look at, as you point out, scans that were acquired four or five years ago, and in some cases early evidence of disease or established disease was already there, but not really leveraged to improve the diagnosis of cardiovascular disease in a timely way that could really change the trajectory of those patients.

And what you’re describing is, as you point out, this concept of opportunistic screening. And much of that work I started working on when I was in the United Kingdom during my doctoral studies—and I was fresh out of medical school training at that point, where I had spent years and years just simply memorizing tables and diagnostic criteria and learning what’s normal and what’s abnormal. And then I came to realize that there’s just so much variation in cardiovascular health and disease and phenotypes. And I got very interested in using emerging technologies to better characterize that and quantify that in a way that is easy to understand and interpretable. It can change these patients’ pathways.

And one of these examples was using cardiac CT scans. We often do those to look at blockages in the vessels of the heart, but there’s a lot of more information there in terms of, “What does the fat around the heart look like? What does the heart muscle look like?” All that information is there, but we’re just, up until that point, we did not necessarily have a way to extract that information in a way that would improve diagnosis and identify patients would need more therapies.

Harlan Krumholz: Evan, it’s such a pleasure to have you on. For those who are listening, Evan is just emblematic of the kind of depth of talent that exists at Yale. And also in this era of AI, the degree to which people who are still early in their career are making truly foundational contributions. I’ve said to Evan, “There’s not really a layer ahead of you.” I mean, there’s not a generation of people who are in their 50s and 60s who’ve been doing this their whole life. If you do heart failure research, I mean, there’s now a whole established hierarchy of individuals who have established that field and work in it.

But when you tack over towards data science and AI, all the rules are still being set, the ways that the studies can be built and the kind of products that can be envisioned. And people like Evan, you say, you know “first year on faculty, assistant professor”—he’s quite senior in this field. I mean, the things that he’s accomplished, the contributions that he’s made, it already placed him among a very elite group of people who are spanning academia and AI. And it’s just been a great pleasure to watch your progress as you’ve been here through residency and cardiology fellowship and now as faculty.

I have a lot of questions to ask you, especially about one of the new things that you publish, but I just wonder if you give people a little sense, what was it like growing up in Greece? I mean, you’re here, you went to UK, you’re in the United States, you’ve now become a person of the world. What was your childhood like? What got you interested in medicine and what got you interested in leaving Greece? Greece is such a marvelous place to live.

Evangelos Oikonomou: Well, I was born in Greece 30, 35 years ago at this point, and I will say I was fortunate enough to have a wonderful childhood. And I owe that to my parents who, both of them, they were pharmacists at the time. So they owned this small pharmacy in downtown Athens, which they were running on their own for 35, 40 years. And so much of my childhood would spend with them, listening about all things related to the pharmaceutical industry, but also in some ways they functioned as their community’s first contact for any medical or healthcare-related question. So I remember I would spend time there in the summers and I really got to appreciate how much opportunity and how much unmet need there is in this space of healthcare and therapeutics. And eventually, seeing new medications arrive or new names every day.

I was always interested in science and engineering, and I guess in Europe you have to decide what you want to do in your life when you’re 16 or 17, so I had to decide whether I was going to study engineering or medical school when I was 16 or 17. And eventually, what drove me to medicine was I just didn’t know yet whether I wanted to work more in a research environment or in a healthcare environment. I wanted to combine everything, and medicine would really offer me the password to study all those things and explore all those things.

I did my medical school training between 2009 and 2015, and that was also during the period of the financial crisis in Greece. And I was fortunate enough to study at a university that provided a lot of resources. But towards the end of my training, I realized I wanted to spend more time dedicating my efforts towards research before I returned to my postgraduate clinical training. And I was fortunate enough to get the opportunity to do an away rotation at the University of Oxford, where I got to meet my mentor, Dr. Antoniades, who I ended up actually working with for four years as part of my doctoral training.

That was a story there. And I will say, up until the end of my medical school, I had not really done much research. I really enjoyed reading about medicine, practicing medicine, being in the hospital. And it was towards the final few months of my medical school training where I paused for a moment, I said, “Well, am I doing this right? Is that how I can contribute the most to this community? Should I take a moment to reevaluate everything that I’m studying and contribute to new knowledge?” And this is what motivated my subsequent steps.

Harlan Krumholz: What was the story that you’re about to tell?

Evangelos Oikonomou: The way I experienced research is that the more time I spent exploring something that was unexplored, the more I got to appreciate how much was out there and how much I got to enjoy working on something that was uncharted territory. And there’s obviously, as we all know, a lot of trial and error in this process. A lot of hypotheses that actually don’t turn out to be valid. A lot of time that is spent working on things that might not necessarily lead somewhere. But in doing that, that really motivated me as I started realizing that there were a few things, especially I started working on AI applications while the field was still nascent, and how can we phenotype cardiac CT scans? A simple thing that now we take for granted, but at the time, I think it was still emerging.

And as I started acquiring that knowledge and that, perhaps, skillset, that I felt, “Well, perhaps I can contribute something here, because I happen to be right here, right now, I’m working on this very specific question that no one else is probably working on or just a few people around the globe are working on it right now.” It felt like a responsibility of some sorts to pursue that further. And I never really thought I was going to become a data scientist, if that’s a word, or an AI engineer, but it was a process. And I felt the responsibility merging those worlds of clinical medicine and AI, especially as opportunities were given to me that perhaps I, like other people did not have access to.

Howard Forman: It may not be completely obvious to people, but you’re also a patent holder, you’re an innovator. The things that you’re developing have the potential to change lives in the very, very near future. And in many cases, they require commercialization. They require investors to build up the innovation that you’ve started into something that people can use on a day-to-day basis. I’m wondering how much you’ve thought about that at this point and what you think about the innovation ecosystem at Yale or even at Oxford and how those entities really do seek to advance the public’s well-being.

Evangelos Oikonomou: This is an excellent question, because I think it showcases how as academics, we often think about the publication, about the robustness of the work, about getting our work published, but the true impact doesn’t end there; that’s where it starts. So we need to develop new methods, we need to show that they work, we need to do that in a robust way. But if we really want to have those methods be translated into better outcomes for our patients, we need to commercialize those technologies. And that doesn’t have to be done—the same person cannot do everything. And I know in our roles every day we wear multiple hats, but it’s important to be mindful of one’s strengths and one’s limitations.

I consider myself as someone who can generate perhaps new ideas, can test them, can show what’s perhaps more useful, more meaningful than perhaps another alternative approach. But there’s a lot of individuals who are much more entrepreneurial than I am, who can bring it to the next step. And this is the long pipeline, especially as some of those technologies are coming closer to regulatory process that is needed, where it’s just impossible for one person to do everything and know everything. And that’s why it’s important to be part of an ecosystem like the one in Oxford or the one here in New Haven with Yale University, where the university will identify meaningful opportunities and provide support and access to the team of people that is needed to do those next steps. And it’s always a team of people.

And a patent is just one thing. And perhaps in 2026, a patent in this space doesn’t really mean as much as in other domains or a few years ago, it’s more about establishing robustness in the pipeline and in the work. And I think what people get to appreciate more is methodologically solid work that actually addresses a real unmet need. A patent on its own doesn’t really mean much. It’s more about addressing an unmet need, and then people will support that vision. And Yale University is an amazing environment that has the resources and has provided the resources for us. Much of the work, as Harlan knows, especially on AI applications in electrocardiography, has been supported by the university to bring that to the next step, which is almost universally commercialization so that it can reach patients in a safe way.

Harlan Krumholz: I really agree with you. I think Yale’s done a great job with this. John Soderstrom sort of developed this sort of text transfer. Josh Geballe has come now and I think really helped do Yale Ventures, infuse it with an aspiration to be the very best place, and he’s actually delivering on it. And the team is amazing and really trying to make sure that the academics have a pathway to that. I’m really glad that you were describing that, Evan.

I want to give the people a chance to hear a little bit about an article you just published, this TARGET-AI article, because I really think it’s just a remarkable contribution. You published it in The New England Journal of Medicine AI journal. And this is the way it works in publishing: they first received your article last June and you went through revisions and everything, finally comes out in January. So you’re probably onto a lot of new stuff since this. But maybe you can translate for people a little bit, why is this article important? What does it mean? This is, as you described in the article, a multimodal approach that integrates longitudinal electronic health records with ECGs to define patient phenotypes and support targeted structural heart disease screening opportunities. But that’s a mouthful, and for many people listening, they’ll need you to break that down a little bit. Maybe if you’re explaining it to somebody you met on the corner, how would you explain what this is, why it’s important, and why it’s such a seminal contribution? Because I believe it is.

Evangelos Oikonomou: Thank you. Let me break it down. The problem statement that we’ve defined is that, as we are all discussing, there’s just so many AI, in a buzzword, a lot of AI applications across everything in medicine, whether it’s AI for diagnosing underdiagnosed heart disease from electrocardiograms, from cardiac ultrasounds, from CT scans, opportunistically or in a targeted way. There’s just a lot of applications that are being developed. Most of those applications are being developed for a specific data type, like data modality. So we have an AI application for an ECG, like an electrocardiogram. We have an AI application for a cardiac CT scan.

And a lot of the papers that we read about talk about the diagnostic performance of those tools. How good are they at separating who has disease and who doesn’t, or how good are they at predicting who develops disease and who doesn’t? But I think over the last few years, our thinking around AI in clinical medicine has evolved, and it’s increasingly about not how we develop those technologies but how do we implement them?

Let’s say we have a tool that can diagnose cardiovascular disease, heart disease from an ECG, an electrocardiogram, an electrical recording of the activity of the heart. How do we use that? Do we use it for everyone that enters our clinic? Do we use it opportunistically for every ECG that we have in our system? And the major concern we have is that when we start doing that, if we start doing that, we end up with a lot of false positives, just because a lot of those things are rare. Even if the models have reportedly good diagnostic performance, if we’re trying to detect something that is only present in 1 out of every 10,000 people, the model to reach that diagnostic performance will have to call a lot more cases as abnormal than there truly are abnormal cases in that population.

And think of an example where you start deploying tools like that and what we call the positive predictive value: how many of the cases you call abnormal are truly abnormal? It could be less than 10%, which means that you end up referring 10 people for more testing and you end up with 9 who do not have the disease. And that means unnecessary testing, unnecessary patient anxiety, increased costs. So we say we have a model that can improve diagnosis, but it might actually overwhelm the system with unnecessary referrals, unnecessary testing.

So what we’re building is, we’re trying to build this approach that sort of, it’s an AI data-driven approach, but it thinks like we do as humans and as clinicians. What is the context where this ECG, this study was done? What do we know about these patients’ history and trajectory based on their electronic health record? What is what we call their computable phenotype? We have a lot of information for every patient, the medications they’ve been on, their labs, their diagnosis when new symptoms were developed, but those models do not actually account for that.

So we need two layers when we have an AI system: we need a way that tells us when to use the AI and then the AI system that will be used to screen for disease. So what we propose in this study and what we show is that we should be more mindful in terms of how we deploy those studies. And perhaps what we say is we can use AI to guide AI use, and sort of form this protective layer to prevent people from losing trust in our systems because they end up with a lot of false positives. We really need to be mindful about how we use those systems at scale, and this is part of the motivation behind this work.

Howard Forman: On the other end of the spectrum of some of the work you’re doing is this practical application of AI to replace human interventions, like with echocardiography. But I think about my own field—it’s 10 years since people said that we wouldn’t need any more radiologists, and there’s a shortage of radiologists now. And I still am optimistic that over time, radiologists will benefit from AI or maybe society will benefit from AI. But what have you learned from your work with echocardiography about how fast our learning curve is for machines to be able to interpret complex studies like that?

Evangelos Oikonomou: I think there’s a key question here, which is, and we’re increasingly talking about that, the human AI interface, if you allow me the word. We have AI tools, as we’ve been talking about, that can automate what humans are already doing, so reporting an echocardiogram using the same diagnosis or labels that we always use. There’s also AI that can detect what we call hidden labels, like things that have diagnostic and prognostic value. They’re there on the images, but we do not necessarily include them or measure them in a reliable way, and we do not include them in the report. But increasingly, we’re realizing that, well, I mean, AI can do both of those things, but AI on its own is not going to help people live longer or do better or feel better, it’s what we do with that information. So there’s increasingly more and more studies in the literature about how humans interact with AI.

And what I find very interesting is that in many cases, actually, the human plus AI partnership is actually worse than human performance alone or AI performance alone. And there’s a lot of reasons why that might be, and it’s also dependent on the context or what it is that we are looking at. But in many ways, there’s what we call automation bias. For example, humans, their performance might degrade because they have this inherent tendency to trust the AI prediction more than their own instincts. We do see a lot of heterogeneity there. Experts actually sometimes might actually do worse when they get access to AI, that might confuse them. Novices actually often do better. Again, depends on the context. But I think it does show us ways in which perhaps we should be more mindful in terms of where we are deploying AI.

Perhaps there’s more of a role for AI where there’s no other specialist or expert to read those studies. Perhaps as a first triage step. But it might not necessarily be the solution where we have experts, and it might even make things worse. And obviously, that’s a different question for every model, every setting, every context, every data type. But increasingly, we’re learning more and more about that. And it’s humbling to see that great diagnostic models do not always translate into better clinical accuracy or better outcomes for our patients.

Harlan Krumholz: Let me just end... this has been a terrific interview and I’m so glad you came on with us. Let me end with a question that gives you a chance to speculate a little bit. I can only imagine what the arc of your career is going to look like, both with regard to your contributions and what you’re going to see around you with regard to the new tools that are going to become available. But as you sit now, in this point in your career, just at the beginning, really, of all the exceptional contributions you’re about to make, what do you think cardiology is going to look like in 10 or 20 years? I mean, how is it going to be different and how different will it be from what we see today?

Evangelos Oikonomou: I don’t know.

Harlan Krumholz: And if you don’t—

Evangelos Oikonomou: The way I think about it is that a lot of the tools that we’re building, my hope is that they will equip and they will empower providers in the community, healthcare professionals in the community to deliver care at a higher level and have access to perhaps expertise that up until now was perhaps confined or restricted to very limited locations. And I think there’s a wonderful example published recently where, we’ve been talking a lot about large language models, and there was an interesting study showing that dedicated adapted large language models can help perhaps primary care physicians and providers and healthcare providers deliver expert care or general cardiologists provide expert care for very specific rare kinds of diseases, where they would probably normally refer those to an expert and perhaps the patients would need to wait several months to get access to specialist care.

And perhaps what these tools are allowing us to do is to consolidate that specialist knowledge into tools that are readily available to the workforce, the healthcare professionals that we have available in our communities, minimize delays and transitions of care, and as we are referring patients for specialist care. And perhaps my hope is that we will really empower primary care, especially as we’re thinking about AI applications in electrocardiography, point of care like ultrasound applications. We will empower those first layers and most important layers of care so that we can improve early diagnosis, and we don’t have to wait for multiple referrals for those patients that contribute to unnecessary diagnostic delays.

So to answer your question, I hope that cardiology, we democratize cardiology and it’s no longer something that specialist cardiologists only have access to. And perhaps there will still be a role for that, but especially for early diagnosis, that needs to happen at the first point of care or at the earliest points of care. And I hope that those technologies would help us achieve that.

Howard Forman: We are so lucky to have you joining us today. I mean, this has been fantastic. And it’s great to even see somebody early in their career who’s already done so much. And I just thank Harlan for inviting you, because you are the best of us.

Harlan Krumholz: Yep. Thanks so much, Evan. It’s great to have you on.

Evangelos Oikonomou: Thank you for having me.

Harlan Krumholz: Keep up the great work. That was terrific interview. I’m glad we got him on.

Howard Forman: I’m so glad you invited him. Yeah, he’s great.

Harlan Krumholz: So great to see stars like this. But now I’m getting to a point of the podcast that I always enjoy, Howie, what’s on your mind this week?

Howard Forman: So this is interesting. The Republic of Guinea-Bissau is a relatively small nation in West Africa. It’s two and a half times the size of Connecticut in landmass, slightly more than half the size of Connecticut in population. So pretty small for a nation, and particularly in Africa, where the nations are huge. Per capita income is a few thousand dollars a year, putting them in the lowest rung of income among nations in the world.

So why do I bring them up? They are at the center of a controversy with regard to a study proposed and funded by the U.S. CDC, in order to determine the safety of the hepatitis B vaccine when given at birth. And this is a topic we’ve talked about occasionally on this podcast. They are a unique place to try to do this randomized experiment because they currently do not do universal vaccination at birth, as most developed nations do, but they do have plans to implement this by 2028, given the proven efficacy and the high rate of hepatitis B in the population. Nineteen percent of people in Guinea-Bissau have hepatitis B.

The World Health Organization has recently come out with a statement on this proposal with several important points that I think our listeners may be interested in. One, there is considerable benefit and no documented harm from the vaccine, and withholding an effective medication would be unethical under most circumstances. A placebo arm, which is what is included in this study, is only generally acceptable when there is no alternative effective strategy. And in this case, we have decades of experience with an effective strategy. And while testing for safety, the basis for this study is acceptable, there is no safety signal to base this study on. In other words, we have considerable safety data at this point, and it is now a hypothesis without a basis to question that data.

They believe, this is the World Health Organization, that “the single-blind, no-treatment-controlled” structure of the study makes the findings less valid to interpret once we even collect data. And they are of belief at least that the study does not do enough to mitigate harm. In other words, they’re not even screening pregnant women for hepatitis B in this study. So with one in five citizens testing positive and 90% of newborns who get hepatitis will progress to chronic liver disease and death, this would appear to be a population where withholding an effective treatment is tantamount to a death sentence in a significant number.

And there’s many layers to this story. I think our history of performing studies on individuals without proper protections should be more than cautionary here. On the one hand, there is possible scientific gain from this study. This group has raised, and by the group, I mean the study group in Denmark that would be funded, this group has raised interesting observations in their prior work, both in support of and in opposition to certain vaccines. But they are not the final word. And their research that they’re doing is ongoing and may or may not be helpful in this specific regard.

On the other hand, you just can’t help but notice that this is a very poor nation with citizens potentially at risk for deception at a vulnerable time in their lives. That is purportedly for the advancement of science, but without the level of safeguards and transparency that would otherwise be expected at this time in history. And so I just bring all this up because right now, this is suspended, it’s not approved, and it may or may not go forward. But I think we talk about this in the press, and I don’t think the press is doing a particularly great job of overall covering it. It is very concerning to me. I don’t know what the right answer is. For additional context, Guinea-Bissau just went through a coup in November. It’s not even a stable government. So I have concerns, and I don’t think that it’s an obvious answer to what should or has happened here.

Harlan Krumholz: Well, what a thoughtful reflection on this. And I would just say one of my take-homes from this was I just didn’t realize what a health threat this was in Guinea-Bissau. I mean, like you said, I mean it’s extraordinary, I mean, how endemic disease are, let alone, now you’re talking about the coup, I’m thinking about the threats of violence. I mean, I don’t know what the life expectancy is.

Howard Forman: I didn’t look it up, but I’m guessing it’s pretty low.

Harlan Krumholz: This country’s got lots of health threats, both external and communicable. Yeah. Well, thank you so much for bringing it up. I think you’ve covered most of it.

Howard Forman: Yeah, no, I’m bringing it up partly because I don’t have an obvious answer. My personal belief is we shouldn’t be doing this study, we shouldn’t be funding this study. But I can see somebody else saying, “Look, there’s a lot to be learned, and they’re not getting the vaccine until ’28 otherwise.”

Harlan Krumholz: Well, this is the thing, what’s going to happen in the vacuum of the funding? Nothing, right?

Howard Forman: Well, of course, it’s at the same time that our own country has withdrawn funding for vaccines in a lot of these countries. We could do the opposite, we could say, “We’re going to throw some resources to save these people.” At very low cost, by the way. It’s not particularly expensive.

Harlan Krumholz: Yeah, but I don’t think that’s happening.

Howard Forman: No.

Harlan Krumholz: You’ve been listening to Health & Veritas with Harlan Krumholz and Howie Forman.

Howard Forman: So how did we do? To give us your feedback or keep the conversation going, email us at health.veritas@yale.edu, or follow us on any of the social media, again, including Instagram.

Harlan Krumholz: We appreciate your feedback. We love to hear from you. We often get back to people and we learn from you. So please, give us any suggestions. They’re very welcome.

Howard Forman: We appreciate it. Health & Veritas is produced with the Yale School of Management and the Yale School of Public Health. To learn more about Yale SOM’s MBA for Executives program, visit som.yale.edu/emba. And to learn more about the Yale School of Public Health’s Executive Master of Public Health program, visit sph.yale.edu/emph.

Harlan Krumholz: And if you want to know the secret to why we’re good, if you think we’re good at all, it’s because we’ve got some superstar Yale undergraduates, like Tobias Liu and Donovan Brown and Gloria Beck. And we’ve got this incredible producer, who behind the scenes actually makes it work, Miranda Shafer. And then of course, the last secret sauce here, I get to work with the best in the business, Howie Forman.

Howard Forman: It’s right back at you, and I agree with you about all that you said there.

Harlan Krumholz: Yeah. Talk to you soon, Howie.

Howard Forman: Thanks, Harlan. Talk to you soon.