publication date: Sep. 27, 2019
Oncologists quickly react to label changes for immunotherapies, a study by Penn, Flatiron shows
Instructor, medical ethics and health policy,
University of Pennsylvania
Staff physician, Corporal Michael J. Crescenz VA Medical Center
Senior quantitative scientist,
In an unprecedented use of real-world data, researchers at the University of Pennsylvania and Flatiron Health have determined that oncologists are responding quickly to label restrictions announced by FDA.
Researchers found that, in treatment of advanced bladder cancer, there was an adjusted 50% reduction in the use of two immuno-oncology agents within six months after issuance of a safety alert.
The study’s findings, published in JAMA Sept. 24, come at a time when oncologists have to keep up with the rapid pace of approval of new therapies. This is especially important in the context of FDA’s accelerated approval program, which approves therapies based on metrics that are “reasonably likely to predict” patient benefit.
Using de-identified data from more than 280 oncology clinics across the United States, the study analyzed data on utilization of first-line immunotherapies and chemotherapy in patients with advanced bladder cancer between January 2016 and January 2019. The majority of the patients, 94%, were treated in community practices, with the remaining 6% receiving care at academic medical centers.
The study, Association Between FDA Label Restriction and Immunotherapy and Chemotherapy Use in Bladder Cancer, examined usage rates of PD-1 inhibitor pembrolizumab and PD-L1 inhibitor atezolizumab in advanced bladder cancer patients who are not eligible for standard cisplatin-based chemotherapy.
The two checkpoint inhibitors, manufactured by Merck and Roche, received accelerated approval in 2017 based on phase II studies. However, data from ongoing phase III studies showed patients with PD-L1-negative tumors had decreased survival when taking these drugs, compared to first-line chemotherapy.
This led FDA to issue a safety alert in May 2018, and subsequently restrict the label indications for patients with locally advanced or metastatic urothelial carcinoma. In August 2018, FDA updated the prescribing information for pembrolizumab and atezolizumab to require oncologists to determine PD-L1 levels in tumor tissue of these patients.
“I think in terms of the near 50% reduction in use of immunotherapy, it’s encouraging from the perspective that physicians rapidly respond to FDA guidance, which is based on real-time changes in the evidence,” Ravi Parikh, lead author of the study, and an instructor in medical ethics and health policy at the University of Pennsylvania, said to The Cancer Letter.
“So, in this case, the FDA warning was based on early reporting of two ongoing clinical trials that looked at the effectiveness of immunotherapy monotherapy as a first-line therapy for bladder cancer versus chemotherapy.”
While it wasn’t possible to conclusively determine whether all PD-L1-negative patients in the study cohort stopped receiving pembrolizumab and atezolizumab—because not all patients received biomarker testing—the results showed that most oncologists, especially in the community setting, are making clinical decisions that align with evidence-based announcements by FDA.
“The FDA is not mandated to collect information about how providers are using drugs,” Blythe Adamson, co-lead author of the study, and senior quantitative scientist at Flatiron Health, said to The Cancer Letter. “Before this study, they really haven’t been able to understand whether or not their guidances are being rapidly absorbed, understood, and changing clinical practice to improve patient outcomes.”
The study found that rates of PD-L1 testing more than doubled within the same six-month period, from 9.3% to 21.2% per 100 patients.
“That increase in testing is corresponding to the decrease in immunotherapy over time, with the mix of chemo and IO that we’re seeing now,” Adamson said. “We really hope it’s corresponding to the patients who are PD-L1-positive getting the immunotherapy, and the PD-L1-negative patients getting chemo.
“The decrease of 50% [in utilization of first-line immunotherapies for advanced bladder cancer] doesn’t mean there’s still 50% room to get better. Because the percentage that’s left, that doesn’t mean that those oncologists are not prescribing the best care. It means that those might be the PD-L1-positive.”
Parikh and Adamson spoke with Matthew Ong, a reporter with The Cancer Letter.
Is this the first-ever study measuring the compliance of oncologists with FDA label changes?
There have been some studies that look at physician prescribing patterns after randomized controlled trials that are intended to change practice. For example, there was a large trial done with cetuximab in colon cancer, looking at in the impact of randomized trial findings on this prescribing pattern.
But in terms of looking at the effect of an FDA label restriction like this on a drug that received accelerated approval without published evidence being put forth, this is the first study, to my knowledge, that shows that physicians respond to that warning even in the absence of published evidence.
The FDA is not mandated to collect information about how providers are using drugs. Before this study, they really haven’t been able to understand whether or not their guidances are being rapidly absorbed, understood, and changing clinical practice to improve patient outcomes.
Could you briefly describe the specific FDA label restriction that is the focus of your study?
Before the FDA alert, there were many different immunotherapy drugs available to treat bladder cancer: there was pembrolizumab, atezolizumab, nivolumab, avelumab, and durvalumab.
However, when the phase III trial results came out, they showed that patients with PD-L1-negative tumors who received immunotherapy had worse survival than just receiving standard platinum-based chemo. When these phase III trial results became public, an FDA alert was issued at the same time. The updated findings were announced at the big conferences, there were news articles, social media chatter, etc.
So, there was first a big wave of dissemination of information, which led up to the following month, when it became an official label change: patients with PD-L1-negative tumors were not to be treated with immunotherapy. When I first started designing the methods for this study, I used the exact date that the alert was announced; I had expected to see changes to start happening right after the alert. Working closely with Sean Khozin at FDA allowed me to understand the story and timeline from his perspective. And, in his mind, the alert was not the big deal, it was the label change that came a month later.
And so, it was really interesting to listen to his perspective, which was related more to the huge effort and work that it goes into making a label change happen, versus the novelty of the moment that an alert is communicated, and how it’s disseminated at medical meetings.
It was very interesting to kind of tie together how this knowledge was spread—from alert to label change. And we picked the date of June 1, right in the middle of when all of these events were happening together.
So, this would also be the first study to look into this matter in real-time, with real-world evidence?
Yes, exactly. In terms of even novel national real-world data sources like Flatiron, or ASCO CancerLinQ, or things like that that have come onto the market over the past two to three years, this is the first study, to my knowledge, that looked at the effect of a label restriction like this, using that data. And I think in some ways it’s a really promising avenue for real-world data sources that can look at the effects of these types of FDA policy or label changes in real time.
It’s important to have this affirmation and reassurance that the system is working as intended. Because, otherwise, one might think that we should be waiting longer or doing larger studies, collecting more evidence to ensure safety.
In this case, one subgroup of patients with a specific biomarker status wasn’t the specific purpose of the earlier trials. It wasn’t until additional information came that the FDA was able to reevaluate and provide recommendations on how best to treat these patients.
How did you arrive at the 50% number in this study? I mean, that’s a significant “response rate,” so to speak, in the reduction of pembrolizumab and atezolizumab in treatment regimens for bladder cancer within six months after FDA restricted the labels for these two immunotherapy agents.
That’s a good question. Essentially, on a monthly basis, we looked at the percent of patients that were starting first-line therapy, and then, with that being the denominator, we looked at the percentage that used pembrolizumab or atezolizumab and took that as the numerator.
And then we calculated the adjusted rates, and that’s how we arrived at some of the numbers that you see in the paper.
But one of the interesting methodologic techniques that we used is actually extrapolating on prior trends by calculating something called marginal effects, which are the difference between the observed rates minus the predicted counterfactual rates if no label restrictions had been in place.
The average marginal effect is another way of how we arrived at that number.
To estimate the marginal effect of the FDA alert, we used a causal inference method called interrupted time series regression. It takes advantage of time-varying covariates to isolate both the immediate shift in level of utilization and any change in slope, meaning the rate of change in use as increasing or decreasing, attributable to the FDA alert when controlling for measured confounders.
We found the effect on immunotherapy use was a decrease of 37 percentage points (%-pts), and chemotherapy use had a corresponding increase of 34%-pts because it was the substitution. At the same time, to provide this more personalized care the physicians needed to know PD-L1 status of patients, which is why we observed the doubling of the rate of PD-L1 testing after the alert.
So, is the adjusted 50% reduction rate an encouraging number?
It’s absolutely encouraging. It means patients are being purposefully matched to their best hope for treatment given the information known at that point in time. Before the FDA alert and label change, we saw the uptake of immunotherapy shoot up rapidly, as we would expect. There was a lot of hope for these patients.
But, because these drugs were really only able to offer benefit to patients with a specific biomarker status, that means that, after the alert, we would still want to see use of immunotherapy among everyone. Because there’s still a population of patients with a biomarker status where it is going to be helpful, and the best drug that they can take.
We wouldn’t want to see the immunotherapy rates plummet to zero or continue to rise at the same rate. In both those scenarios, there would be patients who may have achieved better outcomes with a different medicine.
Not only did we see changes in prescribing practices after the alert, but also a doubling of biomarker testing rates. This reflects this new consciousness that a specific group of patients could benefit from immunotherapy.
It is a small step towards capturing the benefits of personalized medicine.
I think in terms of the near 50% reduction in use of immunotherapy, it’s encouraging from the perspective that physicians rapidly respond to FDA guidance, which is based on real-time changes in the evidence.
So, in this case, the FDA warning was based on early reporting of two ongoing clinical trials that looked at the effectiveness of immunotherapy monotherapy as a first line therapy for bladder cancer versus chemotherapy.
Right, the phase III trials.
Exactly. Confirmatory phase III trials of the drugs that were approved in the phase II setting and received accelerated approval. And so, I think that the response is encouraging because it’s a marked drop in the usage of immunotherapy in a very short time period that is almost entirely explained by the FDA label change.
And so, from the perspective of whether these policies actually work and whether this process works for responding to safety concerns for drugs receiving accelerated approval, it is encouraging in that sense that physicians will actually respond.
Going into the study, were you expecting a higher or lower response rate? And also, what have previous studies shown or not shown?
So, there isn’t a huge evidence basis for this, because most of these studies aren’t able to study these effects in real time and so, retrospective studies of effect in prescribing patterns are sometimes confounded by the fact that you don’t know what’s the effect of the label change itself, and what’s the effect of , secular trends in practice patterns and introduction of novel therapies.
Ideally, a target trial would be possible if we knew the true biomarker status of every single patient, the treatment they were intended to receive, and the long-term health outcomes.
Then we would be able to tell over time, before June 1, 2018, no matter what your biomarker status was, at one point in time you had like an equal chance of getting immunotherapy or an immuno-oncology agent.
When more information was available and FDA communicated the alert, you would hope that if every single person’s biomarker status was known, that they would receive the medicine giving the best chance for the longest survival.
Right, which was why it was important to include looking at rates of PD-L1 testing and see it increase, within six months, from 9.3% to 21.2% per 100 patients.
Yes. In this study, we measured PD-L1 testing rates and prescribing trends across all patients. We did not break down prescribing by PD-L1 test result because the mix of patients who were being tested changed over time. Did everyone with a positive PD-L1 test get the drug that was best for them, with the information known at the time? It’s tricky to answer, because there are reasons that some people get tested and some people don’t get tested.
So, to me, that increase in testing is corresponding to the decrease in immunotherapy over time, with the mix of chemo and IO that we’re seeing now.
We hope it’s corresponding to the patients who are PD-L1-positive receiving immunotherapy, and the PD-L1-negative patients getting chemo. This means that we’re much closer to all of the patients receiving the drug that gives them the best hope for the longest survival.
It seems a study like this would have been difficult to do, if you had to retrospectively aggregate point of care information without well-curated real-world data.
Exactly. So, from the one study that I cited before, cetuximab in colon cancer, that looked at the effects of prescribing patterns after a large randomized controlled trial, we found nearly similar reduction in terms of magnitude over a multi-year period.
But coming into this study, because that published data wasn’t out there, I would have expected personally that, just based on my own practice and based on how much we were using immunotherapy and how much physicians and patients have actually bought into immunotherapy as a first line therapy for cisplatin-ineligible patients with bladder cancer, I would have expected that a label change like this that wasn’t accompanied by published data would have not resulted in as large a magnitude in the reduction.
Of course, you would expect a drop in some respects, because it is a change in the label, but not at this magnitude—and also, because the label was only changed for certain patients, patients with PD-L1-negative bladder cancer. So, the fact that we saw reductions in some way across the board for all patients with bladder cancer, was quite remarkable.
Now, I will mention that it’s tough to know whether doctors are reducing immunotherapy use for the right patients, the PD-L1-negative patients, because we don’t have access to reliable PD-L1 data in this data set. We hope to, but we don’t in this data set.
But we can assume that the majority of the reduction is being driven by patients who are likely appropriate, who are patients who are PD-L1 negative. All in all, I think that the results surprised us; even though we were expecting reduction, we weren’t expecting it as large as to this magnitude.
So, a pleasant surprise, really.
Yes, absolutely. Now, I think it’s one example and it’s the first example that has sort of studied a case like this in the accelerated approval process. So, it’s encouraging from the respect that this might be a model for responding to FDA safety concerns for drugs receiving accelerated approval.
In some ways, it helps to address some concerns about these processes, about what happens if you introduce drugs that don’t necessarily have the gold standard phase III evidence onto the market. So, I think it helps us assuage those concerns in some respects, but there still needs to be data for other types of drugs in this situation.
So, it looks like the Flatiron sample that was used included 280 oncology clinics. Were any academic cancer centers included in the study?
Yes. The study population included 6% of patients receiving care at academic medical centers.
We included both academic and community oncology centers. By virtue of the places that are within the Flatiron Health network, it’s predominantly skewed toward community oncology practices. So, that caveat has to be there.
That being said, community oncology is where most people are receiving these drugs and where most oncology care happens, and so we feel that this is a pretty good representation of what’s happening out there in the country. My a priori hypothesis is that even if we had sampled academic hospitals or academic campus centers, we wouldn’t have seen too many changes in this data.
I see. So, there may not be significant difference in the response between community oncology practices and academic cancer centers?
I don’t know if there would be. I mean, the magnitude of response in the community oncology practice is so large that I’m not quite sure if there’s much farther for academic practices to go, but it’s a question that definitely needs to be answered and probably needs larger follow-up with a data set that involves a lot more academic centers to help answer that question.
That’s a great point, because others have seen differential uptake of novel interventions at academic medical centers compared to community clinics. The 94% community patients in this population represents a random sample of advanced bladder cancer cases in a network of cancer care sites that fit the defined inclusion/exclusion criteria for the study.
The comparison of practice type wasn’t a question that we had pre-specified in this case. And I’m not sure, with only 6% of patients in this study receiving care at an academic medical centers, if we would have the sample size with sufficient power to be able to detect a difference.
Sometimes cancer care at academic medical centers is slightly different from community oncology clinics. It’s equally important to recognize the mix of patient population can be substantially different too.
Cancer patients at an academic medical center might be sicker, have different cancer types, have more advanced disease stage, or live closer in distance to the clinic. And all of those factors could be related to utilization and outcomes, making it challenging to isolate the impact of differences in practice patterns.
Right, and the biomarker testing rates might be different, and that would influence the proportions of various treatment regimens.
Yes. It can be a pretty tricky comparison.
What are the implications of your study for oncologists everywhere, and what does your study say about importance of uptake and response?
With regard to methodology, I think that emerging real-world data sources serve as a great data source to study real-time changes, and for different prescribing patterns, particularly in response to certain policy changes. And that applies for oncology, but it also would apply for any other real-world data source that has the type of resolution of data that ours had.
With regard to kind of uptake of practice, I think that basically what this shows is that, for drugs that receive accelerated approval that have safety concerns that arise in confirmatory phase III testing, the FDA has a playbook for responding to those concerns in a similar way that it did for atezolizumab and pembrolizumab in bladder cancer; and that it can use emerging data from phase III trials, to inform label changes to drugs and doctors will actually listen to that.
All in all, I think we have some reassurance that evidence-based practice translates relatively quickly in this particular setting.
The implication of this study is that oncologists can continue to be attuned to FDA guidances to offer the best treatment for every patient. The results from this study are encouraging and affirming, given the rapid growth of oncology therapies receiving accelerated approval.
Ideally, if we wait and see when the immunotherapy use levels off to a constant utilization rate, we hope it will correspond to the underlying fraction of the population that is PD-L1 positive. That would mean that patients are receiving the best medicine and hope for good outcomes.
Right, since there isn’t sufficient granularity in the data to describe the proportion of these patients that are actually PD-L1-negative, it’s hard to define the ideal response rate i.e. for immediate uptake all across the board. That said, since FDA doesn’t track clinical decision-making trends, as more data emerge, is there a need for formal accountability on the part of oncologists to keep up with FDA decisions?
Absolutely. With the FDA as the arbiter of emerging safety concerns that come about through clinical trials, I think oncologists are in some ways obligated to observe this data and to respond to it so that we’re not exposing patients to potentially unsafe drugs.
As oncologists, we can only base our decisions on the evidence that’s available. So, prior to the FDA releasing this warning, we had no indication that there were some of these concerns around immunotherapy monotherapy in first line treatment of bladder cancer, because that data wasn’t out there.
So, I think that oncologists certainly have to be vigilant and keep up with these FDA restrictions to ensure the safety of our patients. The onus is also on the FDA and for health systems and other guideline-producing bodies to make that evidence salient to providers, so that we know what we’re doing.
The main takeaway was affirmation of trust in our regulatory system. And I think oncologists are put in a really tough situation with so many new cancer drugs coming out.
I think that it is such a hard, hard thing to stay on top of. If we hadn’t seen these results, it would make one pause and think, “Do we need to reconsider how we’re doing this system?” Or, “Is it working?” Are our physicians able to stay on top of all the new drugs and all the alerts, so that patients are consistently receiving the best treatments, given the information that’s known at that time?
If we do learn that there indeed remains a significant proportion of patients that would not benefit from these drugs, is there anything that can be done to perhaps improve this already excellent response rate? I mean, yes, FDA doesn’t regulate the practice of medicine, but are there other ways of improving uptake and response to FDA decisions?
This is an opinion based on the study. A lot of people prior to this study may have argued that you need to publish the full trial before you release this information out to doctors, because who knows whether the safety concerns that arose from the FDA early review of the clinical trials actually pan out? So, there is an argument to be made from some that the FDA may have acted early.
But I think the counterargument, and the one that I believe, is that when these safety concerns arise and when patients are getting exposed to agents that potentially have significant safety concerns or overall survival detriment, that information should be released to physicians even prior to the publication of phase III evidence, because it’s our patients’ lives on the line.
We really need to have that information in our hands. And what our study shows is that, at least in this case, physicians will respond to it even if given that early information. I think that the blueprint of releasing clinical trial data earlier when safety of patients is an issue, is something that that could and perhaps should be followed.
Perhaps this is an opportunity for health technology companies to develop capabilities—e.g. within Flatiron’s OncoCloud—to inform providers and practices about the latest label changes and safety communications from FDA, as well as new evidence from clinical trials, so that timely treatment decisions are being made. I’m certain I’m not the first person to think of this.
I agree that is a promising and useful idea. These types of capabilities are a taste of the early promises of electronic health records—that one day at the point of care what has been learned from many patients will directly inform the next decision.
Demonstrations like this study highlight the progress in cutting down the time from real patient experiences to learning. With recent, curated data, we no longer have to wait years and years and years to collect, process, analyze, report, and then wait for knowledge to spread.
We have a lot of information available to us now, and the technology to process and learn from the data, allowing innovation faster than ever before.