Oby Ukadike: Happy New Year. The new year is here. And we are excited for the conversations we will have in 2023. For now, we're still doing a recap of older episodes while we get back into the office and up and running. We hope you enjoy.

[music playing]

Oby Ukadike: From the campus of Harvard Medical School, this is ThinkResearch, a podcast devoted to the stories behind clinical research. I'm Oby, your host. ThinkResearch is brought to you by Harvard Catalyst, Harvard University's Clinical and Translational Science Center and by NCATS, the National Center for Advancing Translational Sciences.

First, we talk to Dr. Jena about combining health and economic data to come up with interesting questions.

[music playing]

Oby: So the last time you were here, we got into your background and how you started out in medicine and economics. You did a PhD at the University of Chicago. And then you did a medical degree. So is that--

Anupam Jena: That's correct.

Oby: Yeah. It's a long time ago. And you've done a lot of interesting work. And you look at some of the questions that we might not think about, some interesting ways to combine health data and economic data. And I wanted to ask you about some of the recent work that you've done since we last had you on the podcast, which is about two years ago now.

And one of the studies that you did was looking at patients being prescribed opioids for the first time in the ER based on who their physician is and this idea of a natural experiment. Could you tell us about that work?

Anupam Jena: Sure Yeah, absolutely. So I'm sure people who are listening could identify people in their own lives who are prescribed an opioid that they maybe didn't need. And some fraction of these individuals may be on an opioid long term. And it's not uncommon to hear a story about, let's say, a soccer mom who had a tooth extracted and was prescribed an opioid and now one year later is continued continue to be on an opioid.

And when we think about people who have chronic pain or who are dependent on opioids, there's a certain stigma of what that person looks like. But I don't think that stigma is entirely correct. And anybody can become addicted. That's why they're addictive substances. The question, though, is, how do you show this? In medicine, we do randomized controlled trials. But there's no randomized controlled trial that's going to randomize patients to opioids versus not and then follow them up a year or two later to see whether or not they remain on opioids.

So even though it makes a lot of sense that even a single opioid could lead to long-term use, it's actually difficult to show in a rigorous and elegant way. And the study that you're referring to, I think the key insight there was there actually happens to be a circumstance in which patients do get randomized to differing probabilities of being prescribed an opioid. And that happens in the emergency department. When you go to the ED as a patient, you don't know the ED doctor who's going to take care of you. And the ED doctor doesn't know you.

And what that means is that, if you look across a number of ED doctors within the same emergency department, there is likely to be substantial variation in their practice styles. We already know that's true across many other decisions that a doctor may make. But it hasn't been shown that doctors within the same ED vary a lot in terms of how likely they are to prescribe an opioid or what the strength of the dosing is that they may prescribe.

And in this study which was published in the New England Journal of Medicine a year or two ago, what we showed was that, within the same emergency department, the likelihood that a patient is discharged from the ED with an opioid varies by almost threefold depending on who they see. So we're not comparing patients who are going to different EDs. Some EDs may be in areas where opioids are more likely to be prescribed. We're talking about patients who come to the same emergency department.

There's a threefold difference in the likelihood that they're going to walk out with an opioid that is unrelated to why they came to the hospital in the first place. And the reason why is because the patients are essentially randomized to the ED doctor. And what's driving that variation is preferences of the ED doctor or prescribing habits. Now, I think it's somewhat interesting to show that there's this large variation. But what I think is most interesting is to say, well, if you're a patient and you are seen by a high-prescribing doctor, what is the likelihood that one year later you're going to remain on an opioid compared to if you had been seen by a low-prescribing doctor?

And what we found was that patients who were randomized to seeing these high-intensity prescribing doctors, they're about 20% more likely to be on a long-term opioid, like one year out, compared to patients who are initially seen by a low-prescribing doctor. And the summary is that, I think, adds some credence to the idea that a single prescription to someone who was not using opioids before they came to the emergency department, before they saw care, that a single prescription could lead to long-term use of opioids.

Oby: Right. And the example you gave at the beginning, the soccer mom who gets a tooth extracted and is still on an opioid one year out. Is that type of time frame, is that typical for the type of pain that you would see from--

Anupam Jena: No. In fact, I mean, depending on the surgery or what the reason is that you got the opioid, sometimes you don't need an opioid. Sometimes a week or so would be enough. But, certainly, a year out would not be typical. So what's almost certainly going on here is that someone was prescribed an opioid for short-term pain relief but continued to stay on the opioid long term for reasons that may be due to just the sense that the euphoria that the opioid generated. Or they may have other pain issues that they feel are being treated by the opioid.

But, either way, it wasn't for the actual reason that they're prescribed an opioid in the first place.

Oby: You talked about the patients being randomized. They're being, in effect, randomized because, when you go to the emergency department, you're not making an appointment. It's urgent. You just go wherever the ambulance takes you or the closest one that you can get to.

Anupam Jena: Yeah. That's absolutely-- and probably the correct term would be quasirandomized. So they're not they're not randomized by some investigator who's conducting a randomized controlled trial. They are quasirandomized by nature. By nature, this is how emergency department works. You don't pick your doctor. You don't make an appointment. And the doctor can't select certain patients who have higher propensities to be on long-term opioids. It really is effectively random.

[music playing]

Oby Ukadike: Our conversation with Dr. Blumenthal in 2019 left us wondering if we really had a penicillin allergy.

[music playing]

Oby Ukadike: How common are these allergies?

Kimberly Blumenthal: Well, our most commonly reported allergy in America and worldwide is penicillin. And so penicillin was invented-- was in use by the 1940s. And soon thereafter, there were reports of anaphylaxis or a life-threatening allergic reaction to penicillin. And because penicillin is one of our oldest drugs and because it was used to cure a lot of infections over a long period of time, the most common allergy out there is penicillin just by use of a drug.

And so that is 10% of Americans. So that's over 30 million Americans who have an allergy on their record of a penicillin allergy. Either they remember the allergy, or somebody told them about the allergy. And so it's not always a true allergy but about 1 in 10.

Oby Ukadike: That's funny. I'm allergic to penicillin. But not because I actually know. Because my parents told me.

Kimberly Blumenthal: I was going to say that. I don't like to blame moms because I am a mom. But there's a lot of, my mom told me I was allergic to penicillin. And I see that a lot. And that's hard to go through life not knowing if you are allergic or whether you're not because, with penicillin allergy, there are a lot of relatives of penicillins that are used for common infections. And they're old, trusted, safer than a lot of other antibiotics to use.

And so it's hard when you have to work under the assumption you're allergic to penicillin. But you don't really know.

Oby Ukadike: How did you get into this work? What's your background, how you ended up here?

Kimberly Blumenthal: So I'm internal medicine trained. And I did my internal medicine residency at Mass General Hospital. And during my residency, I saw the impact of allergies on care and just purely the idea that you're admitting a patient to the hospital. And you look at their medical chart to understand who they are before you go meet them. And you look at the chart. And you'd see so many allergies. So at least half of our patients had one or more drug allergy.

And, often, the reason why they have those drug allergies is because somebody thought that was a good medicine to prescribe them for something. It's not an arbitrary allergy that doesn't impact their decision making. So we would see cancer patients allergic to their chemotherapy. We would see patients having a heart attack allergic to aspirin, which they need. We would see then patients allergic to penicillin who needed penicillin or a relative. And so it was seeing how this changed decision making that made me want to learn about how we document and how we think about allergies and how and when we should be evaluating them further.

There are some tests for allergies and some tests not, and so where really the gaps are in our knowledge.

Oby Ukadike: Is there a way to find out if you're allergic to penicillin? I mean, I know the answer is yes. But how do you go about doing that even in this space where you talk about people like myself who say, oh, I'm allergic to penicillin because historically my parents have told me I'm allergic to penicillin?

Kimberly Blumenthal: Yeah, so with penicillin, and this is unique for the penicillin right now, we have an FDA-approved drug test, like a drug allergy test, for penicillin. That doesn't mean that allergists cannot test to a whole host of other drugs. The way that the test was invented was to identify the parts of penicillin, the breakdown products that we're allergic to and what happens in the body when someone who is allergic to penicillin has a reaction.

And so this test is a skin test and not a blood test, which would be more useful for sort of spreading across America. It's a skin test. And it's been developed. It's been around since 1960 and then commercially available just since 2009. So in my whole practice as an allergist, this kind of just came out. And now we have to figure out how to use it best in people like you, actually. So you're lucky in that you live in Boston. And we have allergy specialists who are trained to do skin tests.

We have so many of them compared to the majority of the country where nobody might know how to do this. And so it's a specialized procedure that just involves a little extra training but that any physician-- pharmacists are doing it in some states, nurse practitioners, physicians' assistants. So it's a test that really takes about 45 minutes to do and would tell you whether or not you have an allergy to penicillin. It's about 95% correct.

And then we get that extra 5% by giving you amoxicillin and watching you. And that's called an amoxicillin challenge. Where the terminology is the same as food allergy. We call these things challenges. We challenge you to have a reaction or not have a reaction. But we will capture, then, almost all reactions, either in front of us that are severe and then also, sometimes I have to tell my patients that, really, you can have a reaction that night or the next day. But usually it's more mild. But just please let me know because we will capture a lot of the reactions through that sort of testing sequence, the skin test and the challenge test.

[music playing]

Oby Ukadike: In the summer of 2019, Dr. Pandya returned to talk with us about tackling health care costs in America.

[music playing]

Oby Ukadike: You talked about the definition of value and the different factors you think about when you are looking at value. Could you tell us what some of the low and high value services or treatments are?

Ankur Pandya: Yeah. In fact, let me try something. Let me give you a analogy for something that's not related to health. But we can all think about. And then try to cross over and do a health example to hopefully make it stick. So the choice of what smartphone we choose to buy or have can kind of also be viewed as this value, bang-for-buck scenario or like a cost-quality tradeoff so to speak.

So imagine you walk into the phone store. And the cheapest, lowest-quality phone you can get is a flip phone. It works. But it is very low cost. But maybe you want to upgrade. And on the other end of the store, there's the version 12 smartphone, the newest thing that talks to you and does all this stuff, very high quality. High effectiveness, you could say, but also really high cost. I mean, the phone costs like $2,000 or something. The flip phone is only $50.

So if those are your only two options, you could see there's a delta effectiveness. There's an incremental gain in effectiveness. And there's incremental costs. But those probably aren't the only two options. There's probably also the version 10 and the version 8 and the version 4. And you could kind of construct this-- we would call it the efficient frontier. But you get this curve. And it would have this-- oh, gosh. We call it concave shape. But it's this curved shape where the slope is getting flatter and flatter on the-- I guess it would be effectiveness versus cost.

And you can imagine that maybe the upgrade from the version 10 to the version 12 would give you a little bit improvement in effectiveness but will cost $2,000 or something. And it's like, is that bang for buck worth it? Maybe I choose, based on my budget and preferences, to have the version 10 or the version 4. Or if I don't care about phones that much and I don't have a lot of money to spend, I go with the flip phone. On the opposite side, I'm very rich. I don't care. I just want the most expensive thing. I'm going to go with the version 12.

So we make these choices all the time. And so that's the hopefully very accessible scenario. Let me cross over now into health. There's a patient who comes in with-- maybe call them intermediate risk for heart disease. They might have some symptoms. And we want to know if there's some treatable ischemia. I'm going to give you a secret. I'm not a cardiologist. I work with them.

Oby Ukadike: [laughing]

Ankur Pandya: I am a PhD. I work with them. So if I get some of the technical terms a little off--

Oby Ukadike: It's OK.

Ankur Pandya: Yeah. You can edit that.

Oby Ukadike: Luckily, none of us are cardiologists either.

Ankur Pandya: OK. So this is safe space for this example. But, OK, so this patient comes in. They might have treatable ischemia. We don't know. So there are different diagnosis strategies we could use. The most expensive and intensive is the cath. We go in. It's the most expensive. But we know for sure whether or not it's in there. That's kind of like the version 12 phone. The flip phone option is just kind of like wait and see. Let's just see if they develop symptoms. We'll just do lowest cost but perhaps lowest effectiveness because we don't know what's going on.

And then there are some intermediate options in between, like cardiac MR, cardiac CT, SPECT, these are all, as I understand it, different tests that give us some information about what's going on. But they're not as intensive and costly as that cath procedure. So you can construct this kind of curve. And then you get these tradeoffs between effectiveness-- how much health are we getting, and in this case, it's derived from the information we get about what's going on in the patient's heart-- versus the costs.

And if we didn't care about costs, if we just wanted to look at effectiveness, we'd cath everyone and get the most information. But I'm working with some imagers at the Brigham who specialize in cardiac MR. And, basically, you could get a lot of that information for a lot less cost using MR instead of cath. And so the value idea here is maybe that's where the best bang for buck is. If we start thinking about that cost dimension, it could be that the highest value option is not necessarily the most effective.

Oby Ukadike: How can physicians and hospitals become more open to adjusting the services they recommend to become more cost effective?

Ankur Pandya: It's a great question. In my view, cost-effectiveness analysis is most useful when thinking about societal-level decision making, so broad health policies. And that then feeds into how, I think, health policy responses can be shaped to address low-value care or incentivize high-value care in this framework. So, for example, let's say there is a-- oh, gosh.

It's a tired example. But let's say there's a new drug. I hate picking on drugs because everyone does it. But it's probably the easiest to think about. So there's this new drug, comes out. And it is effective. That's good, compared to the current options. But it is very high priced. And if you do the cost-effectiveness math, the cost-effectiveness ratio is very high. Meaning, its bang-for-buck is poor. I would call it a low-value service in my research. So what can we do about this an expensive but effective drug?

I think there are two options. First, we can try to lower the price. We being, let's say, health care payers trying to lower the price such that the cost-effectiveness combination now goes from low value to high value because we've helped reduce some of the cost to the system by negotiating for a lower price. And there are various tools that payers can use to try to do this. I mean, some is just straight negotiation. Others are maybe incentives.

They could put the drug in like a-- threaten to put the drug in a tier with very high copays. So then the manufacturer wouldn't want that because less people would use it. And so there are ways to negotiate the price down to convert a low-value service to a high-value service just by the price. And previous research has shown that, why does the US spend so much on health care? There's this article called-- it's The Price is Stupid. Like, it's our prices that are so high.

So then my question is, well, which prices? Is it all prices? Or are some prices worse than others? And so using cost-effectiveness analysis to define value, you can say, hey, it's these prices that we should really worry about and try to lower, especially because they're for health improving services. We would like to use these. They would help patients. It's just that they cost too much. So that's one mechanism. And then the other is, well, if we can't change the price, then these are just more services to flag and say, hey, maybe we shouldn't do these so much. Kind of like we do for no-value care.

So like Choosing Wisely campaign, other groups and movements towards the less-is-more movement-- if you've heard about that-- they're trying to say, we don't have to do everything in health care. There's some stuff we shouldn't do. And if we are willing to, we can include low-value services using this cost-effectiveness definition in that space and say, these are low value.

And then a clinical guideline or maybe a set of quality measures used by hospitals or payers or physicians can then include these low-value services as part of their measures for quality or appropriateness. And these are ways to essentially nudge physicians, hospitals, and maybe even patients towards high-value care and away from low-value care.

[music playing]

Speaker 2: Hi, ThinkResearch listeners. We'd like to take a moment to tell you about one of our upcoming online courses-- Industry Funding, Application Process, and Establishing Industry Relationships, running February 15 through April 26, highlights the differences between investigator-led research and industry-sponsored research initiatives and covers skill building around networking, mentorship, and communication. Expert speakers will share tips on how to engage with industry and succeed in obtaining research funding.

This course is open to those both in and outside of Harvard. To learn more, visit the link in the episode description. Registration ends February 1. Thank you and enjoy the rest of today's episode.

[music playing]

Oby Ukadike: Dr. Lin joins us in 2020 to talk about the power of baking and of great mentorship.

[music playing]

Oby Ukadike: In your opinion, why is it so important to have mentorship in a clinical and translational research career? It sounds very much like it's a critical component of it. And I just wanted to know your perspective on why it's so important.

Samuel Lin: Absolutely. I think, especially for a lot of the students that have come through the program, they may not necessarily have thought about a career in research. And you're already busy as it is as a medical student. So to have them have this experience, I think, is really important to show them the why research is important and of course try to encourage them to go on to do that. And we've seen that. I think we've had a number of different students that have come through our program.

Mohamed Mohamed, Manuel Enrique, [inaudible], Abraham Isaac, they were all part of the [inaudible] program. They were all medical students. And they've gone on to do great things. Obviously, they all continue in their line of medicine. Mohamed Mohamed, for example, is an ER doc now in Stony Brook. But we do see them start doing a lot of great research experience. Like Walker Kenan was a student in the [inaudible] program, went to UCSF medical school and then eventually to Yale now for his-- he's a resident in psychiatry.

And all through that process, you can see how he has done research in various groups. Enrique was at Albert Einstein. And he had a special elective. And so we actually ended up, in 2017, coming back-- he was initially in program 2015-- to do some additional research with us. And it's great to encourage that kind of interaction and allow them to continue doing research. And we always encourage that. We have another, [inaudible], more recently in 2017, where she continued to work with us and ended up getting a publication out of it, a first authorship in ice hockey concussion.

And so it's really great to be able to continue these relationships and hopefully encourage them to do research and see why it's important in their own clinical career to be able to translate a lot of this kind of knowledge into clinical practice. obviously, they're the best people to do that kind of work.

[music playing]

Oby Ukadike: Finally, in the summer of 2021, Dr. Tambe sat down to have a conversation with us about bringing together different groups of professionals to educate homeless youth in LA about HIV prevention.

[music playing]

Oby Ukadike: So before coming to Harvard, you were at the University of Southern California. While you were there, you partnered with a group in Los Angeles that focused on HIV prevention among youth experiencing homelessness. Tell us about that study and what the goal was.

Milind Tambe: This work that you're referencing was jointly done with the USC School of Social Work. So I was in computer science. And my colleague collaborator friend, Dr. Eric Rice, was in the School of Social Work. And so together, the goal was, on the one hand, to show social impact, to actually achieve social impact in the community of youth experiencing homelessness for HIV prevention.

And on the other hand, advanced AI research because AI had not paid attention to these situations which come about in achieving this kind of social impact through exploiting the social network of these youth. And so the goal of the study was to improve knowledge of HIV prevention behavior and cause those changes. There are 6,000 youth who sleep on the streets of Los Angeles every night.

The rates of HIV among these youth are 10 times the rate of normal host populations. And so dropping centers, other organizations conduct peer-led campaigns in order to spread information about HIV, to reduce HIV risk behaviors. In a sense, you're trying to recruit key peer leaders because you cannot, obviously, talk to all 6,000 youth. Educate them about HIV prevention, expect these peer leaders to talk to their friends. And their friends should talk to their friends and information to spread in a social network.

Now, this is face-to-face conversation. This is not over some other electronic social media or something like that. The question then became, could we do something better in selecting the key peer leaders compared to how it was traditionally done using AI techniques? And the result was indeed that we were able to show that our AI algorithms for selecting peer leaders were far more effective in causing spread of HIV risk information and changes in HIV risk behaviors compared to traditional approaches.

Oby Ukadike: I'm so curious to the AI techniques you used and what that means as far as-- we're talking about social networks and basically determining who within that larger network is the person who can disseminate the information, right? Who's kind of the influencer? And so I'm very interested to know more about, what were the techniques? How did you how did you get to that point?

Milind Tambe: Awesome question. So the basic idea here is we want to select those influencers, those key peer leaders in the network who would obviously be able to spread the information in a way that reaches all of the different subcommunities. So this is a network where there are many different subcommunities. Maybe youth play basketball together, youth who hang out on Venice Beach together and so forth.

The initial approach that was adopted in earlier work was to select the most popular youth, the nodes that, in a social network, would have the highest degree. Selecting them makes sense. They are the most popular. However, they are all concentrated, if you will, at the center of the network. Meaning, they all know each other. They are very popular. But you don't, then, get to reach more of these subcommunities. What we want to do instead is to select maybe one or two nodes in the center.

But maybe the other nodes are more strategically placed. Maybe nodes that connect different communities are more effective in spreading information. What this AI algorithm is doing is not looking at demographic information. It's not looking at any other information other than purely the strategic placement of the nodes in the network. So if you imagine a network of something like, say, 300 youth. And within that, we want to select, let's say, 30 youth who are the peer leaders.

We want to get that combination of 30 out of 300 that are strategically placed in order to spread this information. So choosing a combination of 30 out of 300, if you think about it-- 300, choose 30-- that's a massive number of combinations for us to think through. And the AI algorithm is essentially sifting through all these combinations to figure out what's the most effective way of putting together a coalition of peer leaders that would be most effective in spreading this information.

So they need to be placed just the right amount away from each other because, if you just choose neighbors, that's not as effective. They need to be able to reach different subcommunities. So the algorithm's sort of simulating through all of these different possibilities and then ultimately coming up with the right choice of peer leaders. And, often, it is a surprising choice.

It often ends up with youth who, at first glance, may not be the most popular youth, may appear to be on the edges of the network. But they just happen to be the right youth for spreading this information. And that's what was clearly seen in the result of the experiment. I should say, though, that the network itself is not given to us. So it's not as though we know the social network ahead of time.

And so in the first pilot studies we did, our social work colleagues, students of Professor Rice, for example, would painfully collect this data by querying all of the youth in the study and doing field observations and so forth. And that in itself became a very complex operation. So the next approach then was, could we sample a small fraction of the network? And the key is to figure out what nodes to sample.

So in a sense, we don't know the network. Can we figure out what the network might be by sampling intelligently through a few nodes? And then figuring out, based on just that small sample that we have, who are the key peer leaders? The interesting part of the algorithm was the sampling and then selecting the right peer leaders.

Oby Ukadike: I'm actually very interested and fascinated by everything that you're talking about. One of the things you just talked about was the social network and how people kind of painfully collected the information to give you the information about the social networks. And I know one of the things you had to do in the study was map the social networks of the youth experiencing homelessness. How did you do that part?

Milind Tambe: This is a very interesting question because, often in computer science, the kind of work that was done on social networks for influence maximization was often driven by, let's say, something like viral marketing, things of that nature, where the idea was that you are given a social network ahead of time. And then you go ahead, given that the social network is given, a very clever algorithm for selecting the right influencers. The problem in our case is indeed that the network is not given.

And, as I said, one approach would be to painfully collect all of that data. But that is not scalable. That's not going to work if you want to use this technique across many different locations. And so the next thing was to sample the network and to sample-- essentially, you ask a particular youth, who are your top five friends or something some simple query that gives us information about their edges, the connections that they have. The key is to figure out who to sample, which youth to ask questions.

And that's where the cleverness of our algorithm comes in. I should mention that the study was led by my great PhD student Bryan Wilder who has been the lead in this work. And so he came up with this clever algorithm for sampling the right kind of use. So you sample, let's say, 15% of the youth-- that's it-- by asking them who their friends are. And then just having that much information is sufficient to then select the right peer leaders and then proceed from there.

What we learned here, as I've learned in my other work in AI for social impact, is that going to the real world to solve these problems that are of social impact often opens up new research challenges in computer science that we had not thought about. And this was one example because, traditionally, the work that is done assumes that the social network is given here. The social network is not given. And so it leads to some very interesting, new research challenges.

So in some sense, we get wonderful problems to work on in advancing AI. But these are being driven by socially impactful applications.

[music playing]

Oby Ukadike: Thank you so much for joining us for this compilation episode. And again, Happy New Year. We're glad to be back. Until next time.

[music playing]