SAN FRANCISCO — About a decade ago, David Fajgenbaum thought his life was over. He was a young, bright physician hoping to work in oncology in remembrance of his mother, who died of brain cancer a few years earlier. Fajgenbaum was having his last rites read to him, and his family braced for his death from Castleman disease, a rare inflammatory illness that impacts the lymph nodes and can severely damage other organs.
But, in a rare stroke of skill and luck, Fajgenbaum was able to repurpose a generic drug, sirolimus, and go into remission.
“After dying almost five times in three years, it’s been over nine years that I’ve been in remission on this drug,” Fajgenbaum said in a session Thursday here at STAT’s Breakthrough Summit.
Now, he said, he feels every day is “overtime, time I didn’t think I had.” He has devoted his work at the University of Pennsylvania Perelman School of Medicine and now at his nonprofit Every Cure to finding those glimmers of hope for other patients. When his story made news, he became a leader in the area of Castleman disease, and how repurposed drugs might save the lives of patients with orphan diseases — those with no known treatment. During the pandemic, his lab pivoted to find credible treatments for Covid. And by studying blood samples from Castleman patients, Fajgenbaum has been able to advise providers who’ve run out of options.
Luke Chen, a hematologist at Vancouver General Hospital and Dalhousie University, turned to that wisdom in fall 2020 when he diagnosed the first Castleman patient at his hospital. Chen, who studies inflammation-related conditions, thought Al, in his 40s, might have Covid or lymphoma. He had a fever, abdominal pain, and lacked the signature oversized lymph nodes of many Castleman patients. But a lymph node biopsy and blood tests revealed Al had the same Castleman subtype — called Tafro — that Fajgenbaum has, and that he wrote about in his book, “Chasing My Cure: A Doctor’s Race to Turn Hope into Action.”
Chen tried multiple treatments with Al, several of which worked for a while and then didn’t. “He’s had a very, very rocky course, and with multiple relapses. And so each time, David’s been there for me,” said Chen (who’s been asked to be a scientific adviser to Every Cure). In February, after an especially harrowing stretch — Al, too, was preparing for hospice care as recently as January — Fajgenbaum made one more recommendation based on his research: adalimumab, an immunosuppressant (sold as Humira, Amjevita, and others) used to treat Crohn’s disease and arthritis, but which had never been used for Castleman before.
It was a risky move to try something new when Al was so sick, but he started to feel better within a few weeks. His intense brain fog and fatigue began clearing up, and his bloodwork showed improvements in his inflammation and kidney function. He was able to go home, and to resume some activities he enjoyed before he got sick. While the treatment could still prove ineffective, it was able to bring Al back from the deepest low in his illness and give him more time to live, to be with his wife and young child. “I hope it’s bought him years,” Chen said. “But I’ll take the three months for right now.”
With Every Cure, Fajgenbaum wants to go beyond Castleman and take on the whole universe of 12,000 orphan diseases. Using a unique AI tool, built from a half-dozen other algorithms, he can scour “the world’s knowledge” to figure out potential matches between FDA-approved drugs and diseases without known treatments. Each match is given a score — some 36 million scores total — and the highest-scoring matches are where Fajgenbaum and team plan to spend their time and resources: proving generics can treat those conditions, via laboratory study and, down the road, clinical trials. They already have their eyes on a few targets. For instance, the drug bosutinib scored in the top 1% of all matches for the treatment of ALS.
And on March 29, Fajgenbaum’s birthday, he got a special gift: adalimumab, the treatment he recommended for Al, came back as a match for Castleman disease.
Every Cure has also now identified some dozen other repurposed treatments for drugs that were not intended for those diseases, he announced at the summit.
“It begs the question, how many drugs are sitting in your neighborhood CVS that could be a treatment for you or a loved one that we just don’t know yet?” he said.
Before his appearance, STAT spoke with Fajgenbaum about his work, Every Cure’s ambitious goals, and how he wants to flip the drug development process on its head to help rare disease patients. This interview has been edited for clarity and brevity.
There’s still a chance that these treatments could work for a bit, and then stop working or become less effective. How are you thinking about efficacy, and what comes next for these patients?
So in Castleman, the disease is really intense and aggressive. And it’s not self-limiting, which means that you need to treat it to get it under control. And so one question is, “Can you get it under control?” Which obviously it did. It saved his life. But to your point, the next question is, “How long is going to last for?” And that’s unclear. In my case, we repurposed a drug, sirolimus, to save my life. And it helped in the short term, but it also helped for now over nine years. I’m obviously very thankful for that. It’s impossible to know in Al’s case. The fact that it did work in the acute phase when he was really sick is a good omen that it’s likely to be helpful in keeping him in remission. But certainly, there’s no guarantees.
You’ve been working so far on a case-by-case basis, right?
In Castleman, we do these large proteomics projects or transcriptomics projects, genomic projects for large cohorts of patients. We really dig in to understand what’s happening in the disease. And then from there we ask the question, based on what we’re seeing, what FDA-approved drugs might have an impact on this disease. We utilize artificial intelligence to help with those predictions, particularly identifying subgroups that might respond to one drug or another. That was just to identify repurposed drugs within Castleman disease. And of course, then you can apply it in this case to someone like Al.
But of course, when it does work in someone like Al, then we get excited and say, “Well, how many other patients can it help?” And so then it’s: Should we then move forward to a clinical trial? Can we do more laboratory work? That’s one stream within Castleman disease. With Every Cure, we’re doing this at an all-disease, all-drug scale. The No. 1 drug predicted for Castleman with this new Every Cure algorithm is adalimumab, the same drug that we spent years getting to.
So it was not actually the algorithm that found Al’s treatment?
It was a proteomic approach that found this drug for Al. We treated him with it. And then the Every Cure algorithm also predicted it as No. 1.
Is the plan to make your algorithmic findings available to clinicians and researchers?
The really high-scoring hits, like adalimumab for Castleman or bosutinib for ALS, we’ll do further (and we are doing further) validation. The big question is: Can you show it works in a trial? And we would then raise the funds as a nonprofit to run the clinical trial and prove that it works. In parallel, we will also be making all of the scores publicly available. We’re not ready yet.
Right now, it’s just based on limited datasets. We want to integrate more data, we want to improve the algorithm even further, and then we’ll make the 36 million scores available. And with those scores, the hope is that researchers and disease organizations will pick up the top hits for their particular disease of interest, and will do further work to then hopefully move those into clinical trials, too.
There likely will also be people prescribing the drug, potentially in an off-label fashion, based on promising opportunities. And that already happens. People publish papers all the time about a drug looking promising and then it gets used off-label. But the goal here is to get away from anecdotal, off-label use and to a world where clinical trials are done to really substantiate these opportunities.
Looking ahead, this could create an interesting conundrum for the FDA, if there’s a lot of generics being repurposed for new conditions that aren’t on the label.
Totally. And we’ve started to have some discussions with them around it. It does create an interesting conundrum for them. You know, interestingly, over 20% of all prescriptions written today are off-label uses. So doctors are already prescribing things that are not on the FDA label, which the FDA recognizes, but there’s not really much they can do about it because their mandate is not to figure out all uses for their drugs. Their mandate is to say yes or no to drugs that are brought to them for certain diseases by the sponsor.
And so that’s where Every Cure really feels like we need to lean in, because there’s this gap in the system. You’ve got drugs that are clearly helping people for diseases that they weren’t intended for. And in some cases they’re even being prescribed for, and in other cases no one in the world knows about it yet. But there’s no one that’s responsible for lifting them up and making sure that the work’s done.
What steps are you taking to make sure your algorithm is working correctly — that this doesn’t become another cautionary tale of AI gone awry?
Number one is that AI learns off of what you trained it on. And in our case, we’re utilizing the world’s knowledge of curated datasets. They’re datasets that the NIH has already spent tens of millions of dollars to make sure that, “This drug actually works on this disease or really works on this target.” We’re not just sort of unleashing it.
Two, as we get these scores back, we immediately do validation of the scoring system. We say, “OK, among these 36 million scores, 9,000 of them are for drugs that are already approved for those diseases. So how did the drugs that are already approved for a disease perform in this scoring system? And how do they perform compared to the drugs that we know don’t work in diseases?” We actually know thousands of failed clinical trials where we know that drug does not work in that disease. So that’s helpful in evaluating the platform.
And then we can validate the really promising ones by actually looking at things in the lab, looking at things in clinical data to say, “Does this actually make sense?” And then we’re going to do a clinical trial, which is of course the gold standard for determining whether something works or doesn’t, before we then go out and say, “Let’s use this drug in an area that it wasn’t intended for.”
What are some of the other limitations of the Every Cure algorithm that you’re trying to address?
The first thing is getting more high-quality data. We want to work with other companies like Wolters Kluwer and Clarivate that have access to these large datasets. One is get more data into the system. Two, I’m really excited about also working directly with pharmaceutical companies to say, “Among your drugs that are generic, what are the other diseases you’ve thought about but never pursued? Or maybe you did pursue but never did a trial of because of commercial reasons?” We know the drug companies have to make tough decisions and decide against pursuing a disease or a given drug because it’s not going to be commercially viable. Right now, that information is locked within pharma companies. And if we can unlock that, that would be amazing for this. So we want to get access to private data, like Elsevier.
We’re a nonprofit organization, so we’re trying to do what drug companies do, and that’s to advance drugs down the pipeline. But we’re doing it with drugs that are already generic. They’re cheap, and there is no financial incentive. And so we can’t access these huge capital markets to move them forward. We need to utilize philanthropic dollars and hopefully government dollars, so we’re going to need to do them for as low a cost as we possibly can.
Can you share how much money you’ve raised so far?
In the bank, I think we raised somewhere around $600,000. We’re trying to raise $9.5 million. And the commitments get us somewhere in the middle of that — from where we are, where we need to be.
How is it to see Every Cure actually coming to life?
It’s a dream, but the last nine years has been a dream. I never thought that I would be alive. And it’s been a dream with sirolimus. But now it’s just this whole new dream where it’s being able to help so many patients with the drug I’m on, so many patients with other drugs that we’ve found in my lab. But obviously there’s so much more need out there. And so the idea that we can address the major unmet need of all these patients that are suffering with diseases that don’t have any treatments, and we can actually utilize the world’s knowledge to treat them regardless of the commercial incentives, it feels like a dream.
STAT’s coverage of chronic health issues is supported by a grant from Bloomberg Philanthropies. Our financial supporters are not involved in any decisions about our journalism.