View the talk at https://www.youtube.com/watch?v=IWu6XWzbPBs&feature=youtu.be
Cancer isn't a disease, it's a family of diseases, with a long-tail of variations. Targeted therapies are evolving to supersede chemotherapy, but conventional approaches tackle the most common diseases very slowly.
By contributing their data (disease signature, treatments, outcomes) to a crowdsourced database such as CancerCommons.org, patients can benefit from big-data techniques to identify novel potential treatments such as cocktails of off-label uses of existing drugs appropriate for their specific cases, and can contribute to our knowledge of what works and what doesn't.
6. 20th Century Medicine
Rank
Frequency • Antibiotics
• Vaccines
• Indigestion = alprazolam
• Headache = aceteminophen
• Very effective
for maladies with simple causes
that can be directly targeted
7. The Diseases Left Over
Don’t Respond to One-Size-Fits-All
Treatment
Fast-mutating viruses
Drug-resistant bacteria
Genetic diseases – like cancer
14. Marty Tenenbaum
• Diagnosed with metastatic melanoma
in 1998
• Beyond conventional treatments,
was dying…
• Cancervax was a failed trial
• Cancervax saved Marty’s life
21. Trials of N=1
• Marty and Lukas illustrate experiments that
patients and doctors are performing daily!
– Without access to the best data
– The learnings are seldom published
Especially the negative
26. Patient Privacy
• The threat of death is a powerful motivator
• Opportunity to contribute is a consolation prize
27. Competition
• Large organizations are disinclined to share
• Pace of publication is slow
• Low motivation to publish negative results
• Small organizations incented to share to compete
• Findings from the long tail can pay for failed drugs
28. Ethical Concerns
• Wealthy, motivated, connected patients
driving the learning?
• “Early adopters” in any field:
– Take outsized risk for outsized reward.
– Evolve the field for those who follow
• Alignment of incentives
29. Access to Drugs
• Red tape and rules
• Trials are very limited
• Access outside trials is hard
Rapid Learning Communities are set up to do just this:
Patients, doctors, researchers, and scientists all contribute and access data in a public knowledgebase.
Service providers, drug companies, and other institutions can use the data to offer treatments or enroll trials.
N of 1 trials are anecdotes.
But N of 1 trials pooled across many cases can reveal actionable data.
There are a number of challenges to overcome…
While patient privacy is a big deal, in most cases fear of death is a bigger driver than fear of exposure…
For a host of rational reasons, many organizations are reluctant to share their data,for example because sharing may preclude publishing a scientific paper, necessary to securing ongoing funding.
But smaller institutions have more to gain by sharing with each other in order to compete with the bigger.
And as the pool gets bigger, the next tier is incented to participate.
Finding novel treatments from the long-tail of experiments can lead to new ways to pay for otherwise failed drugs!
Lets touch on an ethical consideration:
These trials of N=1 are typically outside normal insurance coverage.
How should we think about wealthy, well-connected patients funding their own experiments, not available to all?
It drives learning, but also directly benefits only the select few.
Early adopters of any new technology always pay more, take bigger risks, and perhaps get outsized benefits.
Should we allow that for life-and-death treatment options?
The crowd-sourced learning platform has the opportunity to learn faster and cheaper than any other alternatives we have devised.
And the benefits of that learning will flow down to the wider community.
The key here is alignment of incentives: between patients, doctors, pharma, and insurers, so that learning is facilitated for everyone, and there are proper economic incentives to invest in new drugs and treatments.
Perhaps the biggest challenge we face is gaining access to the drugs for off-label novel uses in these experiments.
Access is blocked by red-tape and rules.
Trials are limited (for example, to those patients without complicating secondary conditions).
Trials are limiting (preclude participating in other treatments).
And access to experimental drugs outside controlled trials is hard.
This is not a scientific problem, but a social and political problem, which we can solve.
Building the knowledgebase is a first step.
Cancer Commons is a non-profit set up to create an Open Science Rapid Learning Community…
A public knowledgebase like I have described.
In the unfortunate case that you or a loved-on is diagnosed with cancer:
consider sharing your own data with Cancer Commons
Doctors using the platform may help craft a targeted therapy for your case.
And the data will certainly contribute to learning so that patients who follow will be better off.
Cancer is not a single disease, but a long tail of different diseases
We should apply the big data tools used by Google, the NSA, and Netflix
To crowd-source a massive knowledgebase of data from as many cancer patients as possible.
We can generate personalized treatment recommendations from patterns in the data
And each patient’s experience contributes to learning for those who follow.
Performing experiments on, and learning from every patient sounds horrible.
But it’s a much more efficient way to exploit every ounce of knowledge that can be learned from every positive or negative outcome.
We know no faster way to converge on effective treatments for as many patients as possible in as short a time as possible…
We are a long way from winning the war on cancer
But we have a new toolkit, in the form of Open Science,
and crowd-sourced learning
that can accelerate us in targeting not one disease, nor a dozen types, but thousands of subtypes with millions of possible interactions.
Done 20131115-1336