Hello, my name is Cameron Sabbat and welcome to Cutting to the Case. Today we're joined by Dr. Robert M. Califf, cardiologist, clinical researcher, and two-time commissioner of the US Food and Drug Administration, serving under both presidents Obama and Biden. He led the FDA from 2016 to 2017 and again from 2022 to 2025.
Dr. Califf is one of the most influential figures in modern clinical research. He was the founding director of the Duke Clinical Research Institute, helping build Duke into a global center for large-scale clinical trials and has authored more than 1,300 peer-reviewed publications. He's also a member of the National Academy of Medicine. Between his FDA tenures, he worked on health strategy at Alphabet, including Verily and Google Health. He returned to Duke following his second FDA term and is currently affiliated with Duke's division of cardiology and DCRI. Thank you so much for joining today.
Absolutely. So, I wanted to start out with the question that's on everyone's mind, which is you've spent your career moving between clinical medicine, research, technology, and government. How has each of these perspectives changed the way you think about what actually improves patient outcomes?
Well, you know, it has been quite an experience to have worked in almost every sector. You know, the largest period of time I spent was as a coronary care unit director and a practicing cardiologist. That's still really my home base. And I guess I would say we live in a system, particularly here in the United States, which is fragmented.
Each element plays an important role and each sector is full of good people trying to do the best they can, but the way the system is currently configured, the whole is a lot less than the sum of the parts. So it's sort of a mixed picture for me. I have the experience of being inspired. You know, if you're in the patient care arena, the inspiration is direct when you interact with people who need their health problems attended to. When you're developing new drugs and devices, the amazing things that happen when something really works and can save lives or change quality of life. And of course in government, you're impacting large numbers of people with every decision.
And you know, in the tech sector, probably still underappreciated is the direct effect of social media and the way technology works on the way people think and make decisions about their health. So it's all exciting, all full of good people, all hardworking people, but we got a problem in the US right now about how it all fits together because we're not optimizing healthy longevity at this point.
Yes, unfortunately not. And I wanted to also ask about your service as FDA commissioner. You've served twice, nearly a decade apart. Do you know what was fundamentally different about leading the FDA the second time?
Well, the first time I basically was at a phase in my career, I thought of it as almost like the pre-retirement last act of my career was to go to the FDA as a civil servant initially and then President Obama asked me to be commissioner, which is a political appointment, and that was a wonderful experience, stressful, but everything I've done in my career has been stressful. That's not a problem for me. It's just a really amazing opportunity. And then I was given the chance to come back to Duke and start a data science center with some collaborators, but also to work halftime at Alphabet, with Google being the dominant revenue generating part of Alphabet. And I was having a wonderful time there. It was like being a kid in a candy store.
Then the pandemic hit and I got a call asking me to come back to the FDA. So it was totally different the second time around in terms of the issues that were dominant. It was right in the middle of the Omicron wave of the pandemic. So very, very intense.
What was the same was just being in a place where your mission when you get up in the morning is the health and well-being of the people of the United States. And the fact that in the FDA commissioner role and across the FDA you see everything that's going on in technology, and let's not forget about food, a really hot topic right now as we're speaking. So that was common, and the people just were really great both times, but the pandemic, you know, is a once-in-a-lifetime event and there I was in the middle of having the opportunity to help make decisions in very difficult circumstances.
I can't imagine. Yes. And you know, when the FDA evaluates a new drug, how do you think about the trade-off between getting potentially transformative therapies to patients quickly and making sure the evidence is strong enough?
Well, when we talk about, you know, I would say that drugs get the most publicity in terms of FDA function, but it's only a part of the FDA. And when we talk specifically about drugs, you know, there's a long history of laws and legal cases that set the evidentiary standards and it's not the same for every drug and there's a good reason for that. If you take a field like mine, cardiology, and you take the big part of cardiology like coronary disease, we have a number of effective treatments already. So the evidentiary standard is very, very high. You typically need a large outcome trial with long-term follow-up. But if you take a rare genetic disease with a very high mortality and no effective treatments, people have expressed their views, and this is parents and people with the diseases, that they're willing to take more risk.
And so over the years there's been an accelerated set of pathways put into law by Congress that have a different evidentiary standard. You know, it can be confusing to people because legally, technically the standard is the same, adequate and well-controlled clinical studies, but in the different circumstances the ability to use unvalidated surrogates, for example, in rare disease is something that you wouldn't do in a common disease that already has effective treatments. And so no matter how you look at it, although there is a standard in law, every case is different and in my view this is why the US system of evaluating drugs and devices has been the gold standard for the world because it's been done by civil servants who have no financial conflict of interest but have a great deal of knowledge. And built into this is the fact that the FDA does its own analysis of the data, which is unique for a regulatory agency around the world. So the whole world in many ways depends on the analyses that FDA does.
People have said, "Well, I know a better way of doing it, but I can't think of a better way than having people without a financial conflict and with a lot of knowledge looking at the data and making an informed judgment, which might look different in different circumstances, but typically I would say it happens that way for good reason." Just one other way to think about this. We're getting ready to enter college football season, or football season, where Americans regardless of political affiliation talk about sports. And I think the sporting analogy for the FDA in my view is the FDA is a referee. It's not the coach. It's not the team owner.
The FDA doesn't write the rules. The rules are written by the Congress and the executive branch of the government. The FDA applies the rules to individual cases. And so, I commonly hear people say, "Why doesn't the FDA do this or that?" Well, the answer is because the FDA can't pass laws. The FDA depends on elected people, elected by you, the public, to write the laws.
But it's also the case that if you think about the way referees function in sports, there are rules to the game, but the application of the rules depends on judgment. And often the referees will have points of emphasis. In basketball, it might be the block-charge call is a common one where it looks a little different depending on the year. And in the same way, the FDA has some leeway within the rules, but it's just critical to remember that if you want different rules, you got to elect different people.
Yes. Yes, I agree. And it all comes back to civic engagement, being able to elect. And I'm curious when you talk about evidence-based, the FDA is the referee. Have we become too dependent on randomized controlled trials as the gold standard? Are there areas where real-world evidence can give us equally actionable answers?
Well, let's talk a little bit about terminology here. So, I'm going to start out by saying I think we're not dependent enough on randomized controlled trials. We need to do more, not less. But we also need to take advantage of the fact that we now live in a datafied environment and we have tools like artificial intelligence, of course, computation of various types, and the ability now to look at the human experience over time through electronic health record claims data, digital information from wearable devices, cell phone information, all of that.
And you know, my career started with a database doing observational comparisons in coronary disease in the 1970s.
But I quickly learned that when you talk about cause and effect, for the most part, you need to have randomization for inference. But here's where people confuse evidence and data. Real-world evidence... real-world data refers to data collected everywhere except in a specific research clinic. And so it encompasses all these various kinds of data.
And real-world data has been used forever by the FDA. And when you go to design a clinical study, sometimes you'll have a research clinic where you get part of the data and the rest of it you may get from the electronic health record or other kinds of data to the construction of a study, the analysis of the data. And so I think the most underused method that we have today is pragmatic clinical trials, that is randomized trials done in the real world that are fully representative of the populations to be treated, and I think we're going to see a great increase in the use of that technology.
It's just an application of a method to real-world data. So people tend to say you got real-world evidence and randomized trials. They're polar opposites. That's not true. You have traditional randomized trials, you have pragmatic randomized trials, you have observational comparisons. There's a spectrum there. But randomization itself says nothing about the kind of data that you use. It's a method to be able to draw an inference.
And I, you know, I was sort of spoiled. I started out in the 1980s as a CCU attending right when we figured out what caused heart attacks. And you obviously can't take a heart attack patient and say, "Why don't you come back to the research clinic, you know, when we have an appointment and we'll collect all this data." We had to collect the data in the real world. But we learned quickly you had to randomize to get a reliable answer as to whether there was a treatment effect. That was a long-winded answer, but it sort of maybe in a way summarizes my work in almost every sector, even in clinical practice.
You know, it's important to distinguish when we really know what the best treatment is and when we're applying our best judgment in the absence of the knowledge that we'd like to have. And we used to have a saying back in those early days of doing the big randomized trials in heart attack: the relentless pursuit of evidence. And when we don't know, I think the best form of clinical practice is to participate in a clinical trial and get the answer.
And you know, you spent decades thinking about the American clinical trial system at large, too. You even helped build the Duke Clinical Research Institute and have spent a lot of time thinking about this. If you could redesign this whole system from scratch, what would you change?
Well, I mean for fun, let's compare it to Google, a tool that everyone in the world essentially has used at one time or another and still has billions of users. One of the things that happened at Google very early on was the use of A/B comparisons, which is a randomized comparison. So when you wanted to do something important at Google with its website, you had to design a study. Google had the advantage that it could randomize millions of people in an hour.
But the outcome was ad revenue over time. Hopefully not the outcome we're looking for in medicine, but it has some of the same characteristics that we need in medicine for clinical outcome trials. You're looking at a way of portraying a computer screen, A and B, and then you're following all the people that saw A. You randomly allocate who sees which one. And then you follow them over time and see how much ad revenue is generated. You have people that drop out, you have people that cross over. You have missingness of various types. All that needs to be accounted for. And so the view of Google was you shouldn't make an important business decision without the best form of evidence to what would be useful. And you can see the result of that. They've done pretty well.
So, it's sort of baffling that in medicine and health care, we're willing to just say, "Gee, I don't really know, but I'll give it my best guess based on what I know as a doctor." I'm all for that over just a complete best guess. But in other words, I'd rather have a good doctor than just someone who thought about it 10 seconds ago on a podcast. But the best of all would be a doctor armed with high-quality evidence.
It was sort of hard to think that way until recently because electronic health records were still so clumsy. But we're really getting into an era now with AI where data can be aggregated. It can be cleaned up much more quickly. And I'm hopeful that we'll see a system where we're constantly learning in real time and then applying clinical trials in that context in the real world when we don't know the answer to a question.
And if you really want to get far out, I mean, sort of a fun concept that's evolving now that we're going to need to get used to is sort of a tripartite system of AI agents. I mean, I personally now use OpenEvidence. I think like 80% of American doctors do. So that is AI giving us completely up-to-the-millisecond synthesis of information from journals and clinical practice guidelines that we can interrogate through asking questions or typing questions.
I have an agent combing my electronic health record now to find things that don't look right or need attention. And once that, you know, that's in prototype mode now, but once it becomes available, imagine a world where we have all the evidence and we have an agent looking at our record to see what the recommendations are and then we look at that across millions of people. We can begin to have a real-time evidence generation system that might even tell us when we need to do a clinical trial and then we can all work together on it. Now, I know the next question people say is, "Are you saying we don't need doctors?" No, I think we'll need doctors more than we ever have because the human condition is so complicated and multifactorial. We're going to discover all kinds of things that have been unattended or glossed over just because people are busy.
I don't know about you, but you're probably young enough not to have to worry about this, but at age 70, almost 75, my health record is, you know, hundreds of pages long. There's no human being that can actually go through all that and read it. But an agent can do it in a few minutes.
