You're diagnosed with cancer. Your life changes in an instant and you’re faced with big choices and no road map. Consider this scary statistic: Five-year survival rates for common cancers can vary by 50 percent depending on where a patient is treated. And this: You often can't get precise answers on which type of cancer responds to which type of treatment.
The uncertainties could drive anyone mad; and if you’re like Marty Tenenbaum, a cancer survivor, computer scientist and Internet entrepreneur who thrives on data, it can make you truly crazy. “Patients are dying because information is not evenly distributed – which is outrageous in the Internet age," Tenenbaum says. "Your treatment is based on your mail ZIP code, not the molecular ZIP code of your tumor.”
He cites the 50 percent variation number often as evidence that better information can save many lives. He recalls when he first learned of his cancer, “I went running around to six different doctors, each had a different treatment recommendation, but there was no data with which to make a rational decision on what would work best for me.”
Sixteen years later, Tenenbaum is now an advocate for what he calls "precision oncology 3.0" – using molecular profiling and sophisticated computational methods to reverse-engineer the putative networks that drive a given patient’s tumor, and attack these drivers with combinations of targeted therapies. He founded the nonprofit Cancer Commons to level the cancer playing field so that all patients get access to the same, top-rate data. "Awareness is not the problem today," he says. "We need science, data, so patients can approach their cancer in a systematic way."
Every patient experiences this: you face a life or death decision, which often must be made in days. You go out for second opinions and get conflicting recommendations. You're thrust into this strange world with no maps, no Zagat's, no nothing.
Cancer Commons, which exploits the “convergence of recent developments in genomics, big data informatics, social networks, and personalized medicine," aims to radically transform cancer research and treatment. Here's how it works. If you're a cancer patient, you share your data (anonymously) — what type of cancer you have, its molecular signature (if you've got that), what types of therapies and treatments you've tried and whether they worked or didn't.
What you get in return is highly targeted news and updates on developments that may be clinically relevant to you — including results from the latest medical conferences and researchers, tweets on the top takeaways from the annual personalized medicine meetings, and relevant patient blog postings. You also get access to a curated data base linking molecular subtypes of cancer, with recommended treatments and trials. That knowledge is continually updated based on scientific developments and actual patient outcomes.
When the Commons grows big enough, the thinking goes, there will be a large pool of useable data available for all. (Currently there are only a couple of thousand patients involved, with the focus on melanoma, lung and prostate cancer, but Tenenbaum says a big expansion is in the works.) "Once we get enough data, patients will be able to know, for the first time, what their peers are actually doing and how it’s working. If they then report back what they did, a virtuous learning cycle ensures, resulting in better and better data."
Put another way, he says Cancer Commons hopes to build "a consensus model of the various subtypes of cancer and how best to treat them with the latest targeted- and immuno-therapies, to learn from each patients' outcomes whether the experts got it right or not, and then to rapidly disseminate the results in time to help the next patient."
I caught up with Tenenbaum recently at MIT in Cambridge where he was giving a talk — provocatively titled, "How To Beat Cancer." In it, he argued that often, what are considered to be "incurable" cancer cases may, actually, "be beatable by exploiting biological features unique to each individual’s cancer." Like others, he suggests, "we're on the cusp of managing cancer as a chronic disease using new cocktails of targeted therapies much like treatment for HIV."
He agreed to answer a few more questions. Here, edited and condensed is some of our conversation:
RZ: You talk about a basic problem in cancer care that hinges on patient data. What is the problem?
MT: Every patient experiences this: you face a life or death decision, which often must be made in days. You go out for second opinions and get conflicting recommendations — each doctor knows what they know and they each know different things. You're thrust into this strange world with no maps, no Zagat's, no nothing. So no one could tell me: 'Which treatment is best for me?' [Part of the problem is that] no one shares data — neither the de-identified data from personal health records, nor the data that drug companies collect during clinical trials – not even the data from the control arms of trials, or from failed trials. The only ones with the incentive and urgency to share the data are cancer patients.
After your cancer recurred and you were enrolled in a clinical trial, you describe a kind of "aha" moment. Can you explain?
In 2003, I entered a cancer vaccine trial. Shortly after I went off the vaccine I had a recurrence. I opted for more surgery and went back on the vaccine, but after six months the vaccine was no longer available. The trial had been halted because, statistically, patients on the vaccine arm were not doing better than those on the control arm. However, the vaccine appeared to help some people – and I was fortunate to be among them, having experienced a particularly strong immune response. The vaccine company had no interest in trying to understand why a few patients, like me, benefitted. This is a big shortcoming with clinical trials based on population statistics...to do science, you really need to figure out why it worked in one person and why it didn't in another person. Many good drugs have been rejected by failing to do this level of analysis.
How is Cancer Commons unique? There are other certainly other data-sharing, disease specific, patient-driven advocacy groups out there, Patients Like Me, for instance.
We’re patient focused and science based; Our mission is to aggregate and analyze data, to provide patients with the best information — up-to-the-moment, personalized, and actionable to help them make informed decisions...like a Lonely Planet guide to cancer.
Patients have the legal right to their data — the HIPAA law just changed this year and it makes it much easier for patients to get their data in digital form. But beyond that we want to build this consensus knowledge base — what are the molecular subtypes of this cancer and how should each subtype be treated.
Typically, tumors are analyzed with a genomic or panomic panel — you have data, then you have treatments recommended by experts based on trials. You want patients and their doctors to be able to consult this knowledge base, determine their subtype, determine their options or have a different option. The point is, do whatever it is you want, but tell us what you did and how it worked so this becomes a virtuous learning cycle. This way we can continually test the hypotheses of experts and continually refine them. Cancer is not generic. Patients in the same group who were thought to have the same disease respond differently. For instance, the current melanoma model has about 30 actionable subtypes [a few years ago we knew about 3] and this comes from widespread availability of molecular testing.
[An aside: Exhibit A when it comes to the potential of this molecularly personalized diagnostic testing and treatment is the high-profile case of Lukas Wartman, a young doctor diagnosed with Acute Lymphoblastic Leukemia, a cancer of the blood that is highly treatable in children, but often fatal in adults. Doctors discovered that in Wartman's case, a gene called FLT3 was being expressed at a much higher level than normal. So, using a drug-gene interaction database, doctors at the Genome Institute at Washington University "found a drug, Sutent, normally used in kidney cancer that targets a “hyperactive” FLT3 gene." Wartman's cancer went into remission.]
Why do you compare the current state of cancer care to the early days of AIDS?
Genetically, every cancer appears to be unique, and like AIDS, requires a custom cocktail of three or more drugs to treat it, and prevent it from evolving into a resistant form. With thousands of subtypes and tens of thousands of therapy combinations , the current clinical trials system, which was designed to test drugs as monotherapies on homogeneous populations, is unsustainable. There simply aren’t enough patients to populate a randomized trial for each rational drug combination.
For this reason, we’re designing Cancer Commons to support rapid proof of concept studies in small numbers of patients — or even individuals – by connecting them directly with researchers interested in their subtype of cancer.
Suppose you have a patient who's dying, who has just donated some cells to research. Let’s say a researcher gets lucky, figures out what’s driving those cells, and comes up with a rational combination therapy to arrest their development. Today, the researcher would typically first test the proposed therapy on cells in vitro, and if successful, would then implant some of these cells in an immune deficient mouse and try to treat the mouse. It's not a perfect model, but let's say the drugs are killing the cancer and not harming the mouse? The next step would be to propose a clinical trial, which could take a decade to plan, get approved, fund, run and publish.
Now consider this alternative: I have a combination treatment that looks very promising on the cells of a patient who is dying — how would you like to try your luck on this treatment that seems to be working on your cells? For a patient who is dying and out of options, this process provides an ethical way to get an early proof of concept in a human (versus a mouse), with the potential to knock 10 years off the current process and potentially save the life of the patient.
But for this ‘rapid translation’ approach to work, the researcher would need to have access to the drugs in the cocktail, and that’s currently problematic. Cancer drugs must first be approved as monotherapies before they can be tested in combination. However, cancer drugs often fail as monotherapies for a simple reason: most cancers, when exposed to a single drug, selectively evolve into a resistant form.
AIDS researchers faced a similar conundrum: the early protease inhibitors where not great drugs, and were not being approved. AIDS activists said they're not that good but they're all we have, and pressured the FDA to approve them as long as they were safe and demonstrated biological activity. That gave the researchers a toolkit of drugs which they used to develop effective combination therapies. By focusing on patients, Cancer Commons is mobilizing an army, which will provide the political muscle to effect similar policies for cancer drugs.
How do you envision cancer care and treatment in 15 years or so?
I think there will be a very comprehensive analysis of the molecular alterations driving the tumor and the possible treatments. There will be software for picking optimal therapies and quantitative models of cellular pathways that will enable those treatments to be tested rapidly in ‘silicon avatars’ before being administered to patients. Data will be captured from every clinical encounter and used to improve the software and models. Most of what’s needed to do this, at least to get started on the learning curve, is available today – just not at a single location.
Cancer Commons is therefore assembling an ecosystem that will not only do data and knowledge sharing but also resource sharing across the community. We're trying to connect patients with information and then with the drugs they need to act on that information and then connect them with the specialist doctors and testing services and connect the companies with each other… We’re also working to overcome the formidable sociologic, economic, and legal barriers to collaboration by aligning everyone’s incentives with those of the patient.
For example, when there’s sufficiently strong scientific rationale that a custom drug combination could benefit a patient, pharmaceutical companies should provide enough drug for the first course of treatment for free. If it works insurers would reimburse, and the drug companies involved would gain evidence that could lead to a valuable new indication. If it doesn’t work, the companies would not be reimbursed but they would gain valuable knowledge about their drugs, which could be shared with the community so that other patients and physicians do not waste time replicating a failed experiment.
The biggest cancer centers and drug companies may initially be reluctant to participate, but the smaller ones will realize that they must collaborate to be competitive. The vast majority of patients are seen in the smaller community cancer centers — less than 10 percent are seen in the big-name cancer centers. So the bigger centers will eventually come around.
At MIT, you made a plea for collaborators. How's that effort evolving? And what do you need most?
I'd like MIT students to work on some of the deep, interesting problems raised by all this: the mathematics of reverse engineering tumors, and aggregating incremental evidence across “N-of-1” studies in a rigorous way. Media Lab students are already working on some of the important ancillary problems, such as using avatars to take case histories from patients and creating data visualizations so at a glance you could see how your individual treatment and outcome data compared with that of other patients like you. As noted above, there are also challenging socio-economic and legal issues in trying to get drugs approved and access to experimental drugs for testing in humans.
[A recent announcement by drug giant Johnson & Johnson that it will make its clinical trial data available to scientists around the world is "quite progressive," Tenenbaum says.]
How is your health these days?
I’m currently healthy, but melanoma is tricky, and a little more than 15 years out, I continue to watch carefully for any signs of a recurrence. Over the years I've had about 40 scans — so I’m anxious about cumulative radiation damage. But three weeks ago, I became the first patient in the U.S. to be scanned on a new PET/MRI machine developed by GE. This scan has much lower radiation exposure than current PET/CT scanners, and will likely transform the standard of care for the growing ranks of long term cancer survivors.
Researchers are also looking for biomarkers that can detect incidence or recurrence of a cancer years before a scan. If we can detect cancer at its earliest stages with an annual blood test, and nip it in the bud, that would revolutionize how cancer is managed. Going forward, we will be expanding the Cancer Commons community to support collaborative biomarker discovery and validation.