Transforming Metastatic Breast Cancer Care With AI and Real-World Evidence

In this interview, Adam Brufsky, MD, PhD, discusses the evolving treatment landscape for hormone receptor-positive/HER2-negative metastatic breast cancer, the impact of CDK4/6 inhibitors, the role of genomic assays and artificial intelligence in therapy selection, and ongoing challenges such as brain metastases and treatment sequencing.

Please state your name, title, and any relevant experience you would like to share.

Adam Brufsky, MD, PhD: My name is Adam Brufsky, MD, PhD. I am a professor of medicine at the University of Pittsburgh, where I have been for almost 30 years. For about 20 to 25 of those years, I was codirector of the Comprehensive Breast Cancer Center.

I'm currently codirector of the Cancer Therapeutics Program at UPMC Hillman Cancer Center in Pittsburgh, Pennsylvania. I have published numerous articles, abstracts, and over 350 papers over the years, and I have conducted many clinical trials focused on the treatment of all forms of breast cancer.

How have treatment paradigms evolved for hormone receptor-positive/HER2-negative metastatic breast cancer in recent years?

Dr Brufsky: When you really think about breast cancer, it accounts for 30% of all cancers. The two biggest things that have happened in the breast cancer business are the introduction of anti-HER2 agents with trastuzumab, starting around 1999, then adjuvant therapy in 2005. The other big one has been CDK4/6 inhibitors, which really directly impact ER-positive metastatic breast cancer.

Palbociclib was approved almost 10 years ago, on February 5th, 2015. Since then, we have seen a complete paradigm shift in the way we treat patients. We've compared it to chemotherapy in the first line with ribociclib in the RIGHT Choice trial. It looks like CDK4/6 inhibitors and endocrine therapy are better than chemotherapy in the first-line setting. We have progression-free survival (PFS) data close to 3 years. We have overall survival (OS) data that's over 5 years, probably more in some of the more recent studies. That has totally changed the paradigm.

We've shifted from a disease that most people would die from in 3 years to now, I have people in my practice who are 10 to 15 years out now with metastatic ER-positive breast cancer. On top of that, we have started to use genomic assays, particularly circulating tumor DNA, to help us figure out where to go when someone progresses on CDK4/6 upfront.

The recent developments include the phosphatidylinositol 3-kinase (PI3K) inhibitors, the first one being alpelisib, which turned out to be fairly toxic for some patients. We now have capivacertib, an AKT inhibitor that we can use in patients with PI3K mutations. The postMONARCH trial showed that if someone is on palbociclib and an aromatase inhibitor, we switch them to fulvestrant and abemaciclib. They get, in some cases, up to a year or more of PFS advantage.

There are now PI3K inhibitors approved in the first-line setting, so people with a high risk of recurrence and a PI3K mutation can get  palbociclib, fulvestrant, and inavolisib upfront. On top of all of that, we now have antibody-drug conjugates. We have a wealth of options, and even more are on the horizon. We have such a wealth of things that are available, the question becomes in what order do we use them? Who do we use them in? Who will benefit from what? We're going to have to try to figure that out either using big data sets and artificial intelligence (AI) or through some sort of genomic assay.

People are developing genomic assays for CDK4/6 resistance and for predicting benefit from various targeted agents. That's our real challenge now—we can get people out to 7 years, but how do you get them there? What's the most cost-effective approach? What's the least toxic approach? Are there things we can do even beyond that, especially for patients with long lifespans? For a 70-year-old woman it's great to get her out to 85, but we have a 50 year old woman as well, and we'd love to her to 85 too. That's the challenge now, but we have a wealth of things we haven't had before, which is really good.

With the expanding array of therapeutic options, how do you approach balancing treatment efficacy with quality of life considerations in sequencing therapies for patients with metastatic breast cancer?

Dr Brufsky: I usually divide the patients. Everybody will get a CDK4/6 upfront—that is pretty much the standard of care. Which CDK4/6 you use is a personal choice, but I think that they all have a decent survival advantage.

Within that bucket, there will be people who progress very quickly, within 6 months. Those people probably need some sort of chemotherapy or an antibody-drug conjugate—probably either datopotamab, which was recently approved for this, sacituzumab, or, if they're HER2-low, trastuzumab deruxtecan.

Then we have the patients who are really far out—probably about half of them—who make it 3, 4, 5, or 6 years. We perform next-generation sequencing either on a recurrent tumor or, more likely, in the blood. If they have an ESR1 mutation, all those patients will get a single oral selective estrogen receptor degrader (SERD), such as elacestrant, if they can. If they're ER null or wild type and PI3K wild type, then you need to talk about, maybe, fulvestrant everolimus, or sometimes fulvestrant alone, depending on the patient.

The intermediate bucket is really where it gets interesting. Everybody will get a next-generation sequencing at that point. It will depend, again, if they're ER null, I think a lot of people will still get fulvestrant and abemaciclib because of the postMONARCH data. They may get imlunestrant and, if it's approved for your patients who have ESR1 wild-type, they'll get elacestrant and  abemaciclib, if they haven't had  abemaciclib before.

I still think patients with ESR1 mutations will probably get  fulvestrant and everolimus for the time being—until we have oral SERDs and combination therapies. If they have a PI3K wild-type mutation, for the time being, it's going to be fulvestrant and probably capivasertib. The intermediate bucket is where all the targeted agents are going to go.

That's how I'm evaluating how I treat my patients with ER-positive metastatic breast cancer that progressed in the first line. It is guided largely by the length of therapy, the bulk of disease, what they've had in the past, what their next-generation sequencing shows, and what they tolerate, their comorbidities. You put it all together—its not cut and dry—but I personally like the buckets approach.

Recent trials have evaluated trastuzumab deruxtecan in HER2-low metastatic breast cancer. What are your thoughts on incorporating HER2-low classification into clinical practice, and how might it affect treatment selection?

Dr Brufsky: HER2-low is defined as 1+ or 2+. All DESTINY-Breast06 did was say there had to be some HER2 staining. I'm still a believer. There are people now who say, "HER2 0, it doesn't matter; you don't need to do HER2 testing at all." I don't believe that because, in the preclinical models, some HER2 expression was required to derive benefit from trastuzumab deruxtecan. Maybe I'm a holdout, but I still believe HER2 staining is necessary.

What DESTINY-Breast06 did was show that 1+ means faint in 10% of the cells.  DESTINY-Breast06 took a lot of the subjectivity of what constitutes 10% or not away from the pathologist. It demonstrated that if there is any staining whatsoever, we can give trastuzumab deruxtecan. I think that's great. The issue is where to use it.

Women and, occasionally men, who have metastatic breast cancer, generally don't like IV chemotherapy and coming in every 3 weeks, the fatigue, and the nausea. Many patients prefer to try other options first—perhaps capecitabine if they have only minimal progression, or occasionally intravenous paclitaxel if they have not recently received it—before proceeding to trastuzumab deruxtecan. That's where it fits in.

Then we're going to have a number of patients who will likely have rapidly progressing disease in the adjuvant setting. Those with high-risk disease so they probably are on a CDK4/6 inhibitor in the adjuvant setting. If those people progress really fast with a lot of bulk disease, many of those patients will receive trastuzumab deruxtecan. That's where we're incorporating it. Again, it comes down to clinical judgment—where to use it and at what stage of disease.

Everybody likes the idea of long-term oral therapy if we can get away with it. We want to try that in the first-line setting because it allows patients to maintain their daily lives—they don't have to come in every 3 weeks or worry about if they start coughing or get short of breath. For now, the vast majority of us say, "Let's just try to get away with oral therapy and CDK4/6 for the time being, if we can."

How do you foresee artificial intelligence and machine learning impacting the management of metastatic breast cancer, particularly in predicting treatment responses and personalizing therapy?

Dr Brufsky: I'll tell you, it's going to be wild. We have big databases that are ripe for large learning models. One example is Flatiron—a database that everybody uses. It contains 4 million patient records in it, with probably 10 or 15 years of follow-up for some individuals.

Previously, when we queried Flatiron, we would have to call them up, and they would have to do a manual search for what we wanted. Now, they use AI, and we get the data back in 2 weeks. You have to validate your AI, obviously, on manually curated databases so you can be sure the AI isn't hallucinating, but I think we're going to answer a lot of complex clinical questions that we just can't answer through a clinical trial.

For example, sequencing—what do we give after CDK4/6 inhibitors? What's the best way to do this? Who are the best candidates for each treatment? I've done it through my clinical buckets, relying on my own judgment after 30 years of experience, but AI may be able to refine this further. AI could potentially identify which patients need an antibody-drug conjugate, which need a targeted agent, and which could be managed with a single agent or an oral SERD. I’m very encouraged by this.

The only issues with AI are primarily legal, regulatory, and related to hospital trade secrets—it’s tough to share data. We would love to have a Surveillance, Epidemiology, and End Results (SEER) database where they share a lot more granular data. That's probably where we're going—toward the development of a large, publicly accessible database that could rival Flatiron. Flatiron is obviously a proprietary company, and you have to pay for it. I think at some point in the future, there will likely be a standard established and a requirement that when you stage cases, you've got to input a lot more data. That's just one idea of many that people have floated.

Ultimately, the large learning model you use is only as good as the data you put into it. We have to be very careful, especially when we use real-world data. You still have to use care, and it's going to take a lot of human effort—at least for the first couple of years of this—to put it together. Hopefully, we'll train AI to handle this internally. Maybe, at the University of Pittsburgh, where I work, we will develop our own AI system to analyze our institutional databases, format the data appropriately, and send it to a national warehouse.

I'm on my inpatient service right now, and during rounds, we had a clinical question. We simply asked AI. We had our phones open, someone said, "I'm taking this supplement." We said, "Well, what is it?" We simply asked AI and in 10 seconds it told us what it was for, what the data for and against it was, and I think that's the future of medicine.

Despite advancements, what do you identify as the most pressing unmet needs in metastatic breast cancer research, and what areas should future studies prioritize?

Dr Brufsky: Why do people die from metastatic breast cancer? It comes down to 2 main reasons: liver failure and brain metastases. Interestingly, the immunology of the liver is a little bit different than the rest of the body, so maybe there's something there that we could exploit—who knows?

Related to that, the interaction between the host and the tumor—not just the tumor microenvironment but also the macroenvironment of the host—is something we are just beginning to explore. It's very hard to explore because it's very complicated. We're all different; we each have an individual immune system, so it's hard to get insights that span everybody. You can take a genetically homogeneous mouse, conduct an experiment, and maybe the experiment looked great, but when we try to apply it to humans—who are all genetically and immunologically distinct, even from our parents—it's a little hard to do that. However, with advances in AI, data analysis, doing experiments, I think it's going to crack. They've already started to do it; they're looking at common mutations now and there's a common mutation that actually correlates with recurrence in metastatic breast cancer. People are starting to think about the host.

The second major issue is brain metastases. What we're starting to find, even in HER2-positive disease, is that brain metastases happen at least 60% to 80% of the time, and eventually it's how patients die. We treat the brain metastases, but eventually, we run out of options. We need to find things that get into the brain, which we're starting to do, that prevent patients from needing whole-brain radiation. Once patients receive whole-brain radiation, at some point they start to feel the side effects of the radiation, if they live long enough. You want to avoid that. That's the big frontier.

When you look at the history of cancer treatment, one of the big early successes that people had—nearly 50 years ago—was in childhood leukemia. They tested multiagent chemotherapy in the National Institutes of Health (NIH) and National Cancer Institute (NCI) in the 70s, and it happened to work. However, it only worked about 60% of the time. The other 40% of kids would have disease in sanctuary sites, including the brain and the testes. When they figured it out, the cure rate went up to 90%.

The same kind of thing is happening in breast cancer. We are making significant progress in controlling disease below the neck, particularly in ER-positive breast cancer. I've been talking to my colleagues around the country, and what we're starting to see is people get leptomeningeal disease in ER-positive breast cancer. Understanding this process and finding ways to prevent or delay it will be a major advancement in the field.

The cure of metastatic disease will look like managing a chronic disease, such as heart disease and diabetes. Patients will live with metastatic cancer, but we will control it, hopefully, with medication. Everyone has to die at some point, but we hope everybody dies in their 80s or 90s after a long and productive life. Those are the big-picture challenges. I'm at the other end of my career, having done this for 30 years, but I think that the next generation may start to see some of this stuff actually come to fruition—which is really cool.