Effective medicine has always relied on clear and verifiable diagnoses. Of course, for patients, the wait and uncertainty of diagnostics can be especially trying.
Dr. Connie Lehman is among the scientists and practitioners trying to change that. And she’s doing it in myriad ways—from leveraging both new technologies like artificial intelligence (AI) to conducting old-fashioned operations management fixes—to drastically reduce wait times and detect cancers earlier.
With more than 250 peer-reviewed publications to her name, Dr. Lehman has led meticulous studies of advanced imaging tools to identify breast cancer at its earliest stages—when it’s all but guaranteed to be cured. Dr. Lehman, a BCRF investigator since 2019, is a professor of radiology at Harvard Medical School, and chief of Breast Imaging and co-director of the Avon Comprehensive Breast Evaluation Center at Massachusetts General Hospital.
Chris Riback: Dr. Lehman, thanks for joining. I appreciate your time.
Dr. Connie Lehman: Oh, thanks for having me. I’m glad to be here.
Chris Riback: So, let’s start with the area, the word, the procedure, that everyone knows about, mammography. Is mammography still the gold standard? What are the benefits of mammography and other imaging techniques? And what do you wish were better about the process?
Dr. Connie Lehman: You know, it’s so interesting, because almost every new grant that I review or read, in the area of early breast cancer detection, they start by talking about the challenges of mammography, and we all know those. There’s human variation, some people are really great at reading mammograms, others can miss cancers that are on the mammogram, there’s a lot of human variation. We have challenges with access, any women will tell you that she doesn’t look forward to having her mammogram, the compression, women have concerns about the radiation. So it is a test that’s fraught with challenges, but it is absolutely the best method we have to detect breast cancer early when it can be cured, so we’re working very hard in all kinds of domains [like] liquid biopsy. Wouldn’t it be great if there was a blood test that could let you know whether or not you needed more intervention or whether or not you were harboring a small, tiny pre-clinical breast cancer? So, a lot of fantastic research, but at this time, without any question, screening mammography is the best tool we have to detect cancer when it can be cured.
Chris Riback: One of our recent conversations in the series was on liquid biopsy and yes, that would be fantastic. Not wouldn’t it be, won’t it be fantastic.
Dr. Connie Lehman: Exactly, it will be.
Chris Riback: It will be. As we’re waiting for that, what is an immediate-read mammography, and how in the world did you and the team reduce the median time to report finalization from 61 minutes to just 4 minutes?
Dr. Connie Lehman: So, I love that you asked the question, with what my team and I did, because it took a team effort. It was really our entire community of our administrators, technologists, our fellows in training, our faculty saying, “We need to do screening mammography differently for our patients.” What really stimulated us to start thinking differently was the pandemic, and many people have talked about the challenges of the pandemic. But that silver lining, where we all found we could be more nimble than we realized, and we could tackle some of our challenges, they were actually there before the pandemic, but that the pandemic sort of opened our eyes to the problems. And one problem was the differential experience that our patients had in that first screening that would show an abnormality, to the final diagnosis of cancer and becoming engaged in treatment. And those differences we saw across our patients were really worrisome to us, and we feared that they were getting even more exacerbated with the pandemic.
So, we decided to do something about that. We said, what if we could treat every single patient that came through our doors for a screening mammogram as a VIP. Let’s make sure they have their result while they wait, if they need more imaging, whether for a simple technical repeat or for actually an evaluation of a lesion, let’s just do it while they’re there. Let’s change our workflow, let’s work together as a team to figure out how, rather than having the patient go home and have that screening mammogram sit on our pack station, where we store our images, waiting for a radiologist to review it. What if we flipped it all around, made it totally patient-centered, and said, “We’re going to read these real-time, immediately, while the patient’s still there.” So, for that percentage of patients that need more, more imaging, more workup, we do it right then rather than have them come back, days, weeks, unfortunately for some of our patients, even months later.
And so that’s what we did, and it was really fantastic, not just for our patients, for individually, for them to be so grateful. But also we completely eliminated the disparities we had seen before this program, across differential races and ethnicities in our patient population.
Chris Riback: Why is that? What happened there?
Dr. Connie Lehman: Well, we found out that before we implemented our immediate read screen, the amount of time for patients that we had, our patients of color, whether they were African American self-reporting as Black, Asian, Hispanic. The amount of time between having an abnormal screen and having that confirmation of a breast cancer diagnosis was significantly longer than for our white patients, and that really concerned us. And we have lots of hypotheses on why that might be, but we just knew that it was there. A lot of groups have said, well, we need to educate these groups better, they need to understand the importance of coming back in. But how amazing when we just changed our workflow that we got rid of the problem. So, it became something where we thought, Oh that’s something where you can actually change the system, rather than expect the individual to change their behavior.
Chris Riback: As an outsider, listening to this, among the things that struck me as so remarkable. So we all think about, what can be the breakthroughs in improving breast cancer diagnosis care, all of the things that you were talking about, and we all think about. I think about the extraordinary innovation that is going to take millions of dollars and years. And we do need all of those, liquid biopsy, I mean, there’s an example. And yet still while doing that, there are the things that can be done thinking about workflow. It’s almost like a Harvard Business School case that you just described. Like, how can we improve the internal operations of our manufacturing plant, and on some level, that’s what you did, creating a tangible difference in care, without tens of millions of dollars and five to 10 to 20 years of research.
Dr. Connie Lehman: You know, I’m so glad that you’re highlighting that. In fact, we should do a little shout-out and kudos to the Harvard Business School. Because I was fortunate to take a year-long program for healthcare leaders at the Harvard Business School, and it was exactly what you said. It was like these case studies where someone would say, “Well, wait a minute, can’t we do this better, can’t we just change and question the way we’ve always done things and do it a little bit differently.” And, certainly, in other industries out of healthcare, we have innumerable examples of people just thinking differently and really having that, customer-centric approach. And we’re bringing a lot of those processes and those paradigms and those approaches, into our healthcare system. It’s been really exciting to see how quickly a group can actually change their thinking and change the healthcare delivery paradigm or model.
Chris Riback: Yes, the mindset. But what happened on the workflow change that we were just talking about in the immediate read mammography? What happened to the accuracy as you cut the time to final reporting?
Dr. Connie Lehman: Well, this is a great question because many radiologists who I had talked with said, we’re a little bit worried that if we’re doing immediate reads, we might get a little stressed, we might be tempted to race through looking at the mammograms. We might get distracted, someone needs us to go do a biopsy or to look at a diagnostic mammogram. And when we read our screening mammograms, we are not distracted, we’re doing it in batches and we don’t have anyone else bothering us. So we set up our immediate read using those same critical elements of the best reading conditions. So, our radiologists that are reading the screens in real-time, they’re in a room where they close the door. All they’re doing is reading the screens.
We also have these monitors up. So, all of us that are in the clinic that day can see the unread screens, and if the number starts to creep up, [say if] I’m doing diagnostics, I can jump in and read a few of the screens real-time. So, it was not just saying, “Well, we’re just going to start reading these as fast as we can.” It was actually building a full system to support all the good things about, quiet, protected, batch reading, but timed so that it was happening while the patient was still there.
Chris Riback: Interesting. Part of me is wondering, did changing the operations, and cutting down the time dedicated, or available to that read, or focusing the time, I guess I should say, on the read. Did that change the mindset of the people reading it? Maybe you were even more focused, because they knew that they were trying to do it in a tighter period? Anyhow, for all the Harvard Business School professors out there listening, I think we’ve got their next case study and, get ready Dr. Lehman, now you’re going to get to be the star of an HBS case study. Unless they’ve done it already. Have they done it already and I’m late to the game?
Dr. Connie Lehman: No, no, not yet. Yes, we definitely should. And I do like it. We’d studied very carefully, and published our findings and our results, showing that our accuracy was equivalent. That we weren’t calling back more patients, we weren’t seeing fewer cancers. But that we have the same performance level, and obviously that’s critical for this type of program to be successful. So we didn’t have that downtime, where the images were just sitting in our storage system with nobody looking at them. That’s where we really adjusted the timing.
Chris Riback: Thank you for clarifying. Let’s turn to AI. What role does AI play in breast cancer detection? How does it work in terms of identifying personalized risk?
Dr. Connie Lehman: So, I am obviously incredibly excited about this revolution in healthcare with artificial intelligence, the possibilities are limitless. We have had such an exciting two decades of what I refer to as -omics. Genomics, proteomics, radiomics. We have so much data, so much information, but all the information was outpacing our human ability to analyze, to process, and for me, as a radiologist, certainly for my human eyes and my human brain to be able to take in. And thank goodness along came the unbelievably fast computers, and the entire revolution of artificial intelligence and deep learning, and we’re leveraging those tools to have the highest impact as possible on our patients. And it’s going to be in every domain, from risk assessment, to intervention, prevention, early detection, diagnosis, treatment, returning to surveillance. But it’s going to be up to us humans to use them well, and use them with a real attention to rigorous science, to quality, to equity, to all of those things that we’re trying to carry forward in this revolution, in this new domain.
I think one of the things that’s challenging whenever there’s something as exciting as AI, is everyone can a little bit get ahead of themselves. So we have some claims out there that, computers are reading mammograms better than radiologists. This is still early, it’s mainly retrospective studies, reader studies. We all learn from the story of CAT and mammography, that reader studies don’t always translate over into actual clinical practice. So we’re not going to repeat those sort of sins of the past, and we’re going to do this in a really smart way.
The specific area that’s so exciting for me with AI, is risk prediction. So we’ve always been using the mammogram to try to find a cancer. But as a mammographer reading mammograms all the time, I always notice the things you can tell about a patient from her mammogram like, oh, this is a woman that has actually gained a lot of weight since her last mammogram, or wow, this woman lost weight since her last mammogram. We can see that the woman has had a prior needle biopsy, or an excisional biopsy, or even cancer treatment. We can see that a woman has started to go through menopause. Maybe she’s gone on chemo prevention, or maybe she started hormone replacement. Maybe she’s lactating, all of these different factors and features, which we know influence the breast tissue, and impact the risk of future breast cancer. We can see that on the mammogram, but all I could really do is observe it.
But now with deep learning, we’re taking that data out of the digital mammogram, and we’re using it to predict a woman’s future risk of cancer. And that has been incredibly exciting for us to start to explore, it’s something that the Breast Cancer Research Foundation has been equally enthusiastic about. Not only allowing us to do investigations in new domains, novel applications of AI that others aren’t doing, but also create those environments for partnerships. So we’re not just doing this with, amazing computer scientist at MIT and fantastic breast imagers at Mass General Hospital. But also medical oncologists, surgical oncologists, epidemiologists from all the different groups and teams that the Breast Cancer Research Foundation brings together.
Chris Riback: Yes, cross-disciplinary work. These conversations that I’ve been so privileged to have with people like you, so often that’s what comes up.
Tell me, my understanding is that the study that you are working on, the one that you were just talking about. I don’t know if you are still in year two of three, if you are now in year three of three. You’re taking, if I understand correctly, a collection of digital mammograms from the participants of the Nurses’ Health Study II, which I think is data that I’ve talked about previously in other conversations. And this pilot study, you just talked about, that it’s predictive, and you talked about the model. How did your colleagues, your team create the model? How do you anticipate what the model should show? You talked a little bit about some of those factors. And what are you finding? I know it’s early days, but the predictions that the model is developing, how does it compare with actual results?
Dr. Connie Lehman: Exactly. So well, first your question about what year we’re in. We are in the early phase of our year three. And we’ve been able thankfully to make an incredible amount of progress, even despite the pandemic.
So the model was trained on a large population of mammograms that we had within the Mass General system. And then we just, as you said, wanted to evaluate and test it in other data sets. So, an external validation, so Rulla Tamimi and others that were heavily engaged in the Nurses’ Health Study, worked closely with us on this Breast Cancer Research Foundation project, and we learned so much. One thing we learned is some of these older databases of mammograms aren’t what we refer to as AI-ready. It took a lot of work for those mammograms to be saved and stored, but they weren’t always saved and stored in the exact ways that we need to be able to test and train our models, etc.
In research, everything’s not always a win the way we might think, some of the areas where we find it didn’t work the way we thought, it leads us to greater understanding.
I was lucky to have an early, early mentor when I was getting my PhD at Yale. And I would do the ‘Gosh, darn,” when the research didn’t turn out the way I wanted. It was like, this is what’s fun about science. Like it’s, the data are friendly, they’re going to guide you. They won’t always show you what you thought you were going to get or what you wanted, but they’re going to guide you in the right direction. And so Rulla and I just sort of picked ourselves up, brushed ourselves off and said, “Okay, now we’ve learned and know where can we go for the next phase,” and I think that’s going to help a lot of people.
Dr. Connie Lehman: So, we pivoted, and we had partnerships with seven hospitals around the world, and we did our external validation, and we also had a real eye for making sure that we were going to be able to see that this model worked across various races and ethnic groups. And the reason that was so important is we found that our traditional risk models perform extremely poorly outside of European Caucasian women. So for example, Emory was one of our validation partners and almost half of their women undergoing screening mammography self-identify as African Americans. So we were really pleased and excited to see that the model validated very well, had very robust performance at these seven hospitals around the world from Brazil to Asia, in the US, Europe. So we published on that and then we continued to work in other domains, and now we’ve built a very robust infrastructure so that every mammogram at Mass General is processed through the AI model.
And we’re starting to evaluate that now more. Now we’re set up more to do prospective evaluation, which is going to be critical because the bulk of AI work today in this domain has been retrospective studies, and we really need to shift towards prospective evaluation. And then the most exciting, for me, part of this third year is we’re bringing in the other areas of information on patients, biological factors, other tests, very rich databases that we have on patients. Now, we had expected to do this in the Nurses’ Health Study, and we’re still looking at different ways that we can bring that information in. But we also have a very rich source of databases within the Mass General Brigham system, and so we’re pivoting and moving forward into that next level of, what if you had both the biological information about the patient, as well as the imaging information and data.
Chris Riback: And just so that I’m understanding you correctly, that additional data will help evolve and inform the AI model that you will then apply. So you’ll be able to bring in not only that initial set of data, but also this biological data.
Dr. Connie Lehman: Yes. What we’re hopeful is that all of these sources of data are going to be additive. So we can move further in being even more targeted and more precise for each individual woman. Traditional approaches with risk is, you can say, “Well, this whole group of women has a higher risk of breast cancer than this other group of women.” But getting down to being able to tell an individual woman more precisely what her own individual risk is, has proven really challenging.
Chris Riback: Yes.
Dr. Connie Lehman: When we looked at our AI values in this Breast Cancer Research Foundation project, we had the AI values of all of our patients that had undergone screening MRI. And there was a significant group that had known genetic mutations for breast cancer. We were surprised, and I always like it in research when we find something that surprises us because it’s one of those, eureka, a-ha, kind of exciting moments.
Maybe it’s because those women with genetic mutations are on chemo prevention and maybe that chemo prevention, which is reducing their risk, is making their breasts on the mammogram look like, their neighbor that doesn’t have a genetic mutation is bringing that risk down, and we can actually see it with AI on the mammogram. But see it with AI in quotes because I’m not seeing it. So we’re excited about that. It also suggests that there might be multiple sources of information that can help guide us, in more precise prediction, that the genetic information is distinct, does not totally overlap with the imaging radiomics AI-assisted information.
Chris Riback: Yes. I’m curious as well for you personally. I assume that you were trained as a radiologist. You are now getting neck-deep into computer science and artificial intelligence. Is that something that you always were interested in? How has that transition been for you?
Dr. Connie Lehman: Well, when I was younger, I was just interested in everything. In fact, I thought I was never going to be able to decide what I wanted to do because I just wanted to do everything. I was certainly always interested in biology and the sciences and human behavior and the brain. I decided to pursue a PhD in psychology at Yale. It was my first mentor that said, “Ah, you have to combine this with medical training. That’s really what you want to do. I can just tell.”
So, I combined my MD and PhD training at Yale, and then when I was going into radiology, my friends were a little bit surprised because they thought, Oh for sure you’ll be doing the neurosciences or something. They knew I was really passionate about women’s health as well. But for me, radiology’s been just the absolute perfect sandbox to do my work in, because it provides us so many opportunities to have a very high impact on human health and all the domains that are interesting to me. How the brain works, how we change patients’ cognitions and behaviors, and how people make decisions, whether it’s a radiologist, making a decision that a mammogram is normal or abnormal, or a patient making a decision on whether she wants to undergo breast MRI or not. Whether she wants to know more about her risk or not, or if she’s going to come back for a screening mammogram.
So I don’t know, I just keep finding these incredible opportunities. And to me also, my career has always been about relationships. So it was when, Regina Barzilay at MIT had just completed her own treatment and said, “Do you want to work on a project together? I’d really like to make a difference in this domain.” That started this really intense exploration.
Chris Riback: Yes. Well, that would be a hard person to ever say no to. And I assume, I found myself wondering with all of those different interests and potential occupations, French scholar, was that ever on the table?
Dr. Connie Lehman: Do you know, having French as a language never seemed like it was going to bring any added value to any part of my work. Until I was working on a project in Uganda, with some absolutely fantastic colleagues at Makerere University, there in Kampala, and I was walking through the hospital and there was a man who was clearly confused and no one could understand him. And I felt so sorry for him, and the closer I got, then all of a sudden, I thought, Oh, he’s speaking French. And I was able to translate from French to English, with my Uganda colleagues, who then helped the man find out where he was going to go. And there’ve been two other experiences where all of a sudden French was helpful, but many a time I had wished in the US, I’d just learned Spanish way back. But anyway, and then of course travel and fun. French is always going to be a good language to have.
Chris Riback: It’s always good for that. Just to close out the conversation, you mentioned it earlier, but the role that BCRF has been able to play. I know that you’re affiliated with a range of organizations, and you’ve mentioned some of them, and working with MIT and different institutions as well. But how would you characterize what role has BCRF played in your research?
Dr. Connie Lehman: The Breast Cancer Research Foundation has been so critical to the work that I’m doing. And I will tell you, one of the things that really resonates with me is the culture that they set for their community. It is such a culture of inclusiveness, of being excited about partnerships, about being excited about creativity. I’m never worried that when, for example, Rulla and I realized the Nurses’ Health Study mammogram data didn’t work the way that we thought it was going to work, like, “Oh, they’re going to be so upset.” We pivoted, we started to explore other areas. It allowed us an opportunity to do new regions and new areas that we hadn’t even anticipated.
And so that culture of work together, partner together, ask the challenging questions. There’s going to be successes, there’s going to be setbacks, but we’re all in this together. Our patients have been through layer after layer of challenges that we didn’t wish on anyone, and yet you have this organization that the entire time was like, we’re right there with you, and it goes a long way.
Chris Riback: So actually, as we come to the close of the conversation, are there connections? I mean, we talked earlier about the immediate read. We’ve been talking about the AI work, does it all tie together in some way?
Dr. Connie Lehman: You know, it really does, and I’m so glad you asked that. Because it’s easy for us to state our mission statement, that we want to provide equitable access to high-quality healthcare for the full diversity of patients we serve. So we say that, but then we start to think about, well, where do we fall short of that? Certainly when we reopened after we had to shut down during the pandemic, when we found that of our six screening centers, the screening centers that we’re serving our more vulnerable patients were slower to open on Saturdays when more of our patients needed access. And they were slower to open to full capacity, than those screening centers that happened to be in neighborhoods that served our less-vulnerable patients, we knew we had to change. And by studying that, we pivoted quickly, so that early on, we saw those changes and then we corrected by the end of the year. But it really took a lot of effort and a lot of attention that was part of the immediate read screen.
Let’s get more people in and get them fully taken care of. So we don’t have this inequity and who is able to talk their way into, like, “Oh, can you have the doctor read it while I’m here,” and another patient might be less comfortable in asking that, or have a harder time coming back to the hospital. It was a really challenging time. So the whole immediate read screens, we also had our same-day biopsy program looking for equitable access across all of our centers for weekends, for evening hours. That was a big part of the work, but also in the AI, it was so exciting for me to see how the AI risk model worked, and it worked better than our traditional risk models. It was chilling for me to see in my own patients that I feel responsible for taking care of, how inequitable the traditional risk models were.
We found and will be presenting in Chicago at the Radiological Society of North America [annual meeting], that our Caucasian patients are two-and-a-half to three times more likely to be given access to risk prevention, and risk reduction based on the traditional risk models. Even though we don’t see a difference in the rates of cancers in those populations, and that’s really chilling. And there was a beautiful article written in the New England Journal of Medicine, not only tackling these inequities with race across risk models in breast cancer but across everything from renal disease, cardiac disease. So, we have a real opportunity, and I think you may have heard some people say, “Well, we’re worried that these AI tools are going to have even a greater divide in the haves and the have nots, or even greater inequities.” So in healthcare we’ve got to double down and really make sure from the beginning to the end, we’re training the models in the right way. We’re testing them across the full diversity of our patients, and we’re bringing the AI tools into delivering on that mission we have of equitable healthcare for all of the patients that we serve.
So, we see that it all comes together. We’re humans and we know what our mission is, and we are grateful to have so many different tools to make things better, not just for some, but for all of our patients.
Chris Riback: Yes, it sure does, it starts with the patients, as you just mentioned, and then the folks like you. Dr. Lehman, thank you. Thank you for the conversation. Thank you for the work that you do.
Dr. Connie Lehman: Thank you so much. It’s been a pleasure.
Please remember BCRF in your will planning. Learn More
Breast Cancer Research Foundation28 West 44th Street, Suite 609, New York, NY 10036
General Office: 646-497-2600 | Toll Free: email@example.com | BCRF is a 501 (c)(3) | EIN: 13-3727250