Read on Twitter
What happens when everyone is empowered to participate in data driven research?
We're very close to this happening:
- Access to health records increasing
- Sensors proliferating
- New tech enables us to learn from data without sacrificing privacy
We're very close to this happening:
- Access to health records increasing
- Sensors proliferating
- New tech enables us to learn from data without sacrificing privacy
Billions of $ have been spent building out electronic health records & new startups, like @1up_health & @humanAPI, provide APIs to facilitate easy data sharing.
With this layer created devs can focus on higher order applications, like analytics or novel ways of engaging people.
With this layer created devs can focus on higher order applications, like analytics or novel ways of engaging people.
On top of this the ONC recently finalized new regs requiring that patients can access all of their health data at no cost, electronically. Fines will be levied against organizations that don't comply.
Net result = people will have easy access to their health data
Net result = people will have easy access to their health data
Or perhaps they already have some of it: more people than ever have wearables, smart homes, or other things quantifying their lives. Often times these folks are already taking the initiative to track themselves and their habits to gain more insights into their lives.
New tools like federated learning, multiparty computation, and others let us learn from our data without sacrificing privacy. No need to fork over data to a 3rd party or for deidentification, just let an algo train on your phone. https://twitter.com/bertcmiller/status/1261735806789697536
With information at our fingertips and armed with privacy preserving technologies more people will be willing and able to contribute their data for research purposes.
I think there are two ways to think about this.
I think there are two ways to think about this.
The first is in terms of scale: the largest observational studies ever could be amassed this way.
Early federated learning networks show was is possible. This network training an algo to identify brain tumors has 29 organizations participating. https://www.businesswire.com/news/home/20200511005132/en/%C2%A0Intel-Works-University-Pennsylvania-Privacy-Preserving-AI-Identify
Early federated learning networks show was is possible. This network training an algo to identify brain tumors has 29 organizations participating. https://www.businesswire.com/news/home/20200511005132/en/%C2%A0Intel-Works-University-Pennsylvania-Privacy-Preserving-AI-Identify
I think @apple's heart studies are also instructive. When participating in research is as easy as downloading an app a lot of people participate.
As traditional clinical trials struggle to enroll 100s this @apple study enrolled 400,000 people! https://twitter.com/bertcmiller/status/1234300406748123136
As traditional clinical trials struggle to enroll 100s this @apple study enrolled 400,000 people! https://twitter.com/bertcmiller/status/1234300406748123136
The second is in terms of democratization: any sophisticated and motivated person can use these technologies to learn from data.
I expect to see research, a la DIY bio, being done outside of the purview of traditional institutions.
I expect to see research, a la DIY bio, being done outside of the purview of traditional institutions.
There are lots of existing communities oriented around learning and sharing info, e.g. fitness, nutrition, or chronic disease communities. Could easily see these communities leveraging new tech to research something in their domains.
Some communities will work with traditional institutions, and some will go their own way.
Those that go their own way will inevitably cause tension as the authority of ivory towers is challenged by upstarts with new agendas, evidence, and methods.
Those that go their own way will inevitably cause tension as the authority of ivory towers is challenged by upstarts with new agendas, evidence, and methods.
Of course the limitation of the research that I've described here is that these will be observational studies limited to drawing inferences from data.
But that alone is immensely useful, and could be enough to generate hypotheses to be tested in randomized control trials.
But that alone is immensely useful, and could be enough to generate hypotheses to be tested in randomized control trials.
There's also an issue of reproducibility. Privacy preserving technologies, by definition, mean that the underlying data behind some derived insight aren't shared. That means a 3rd party can't replicate your analysis.
Anyway, I've had these thoughts in a Notion doc for awhile and figured I'd tweet them out to see what Twitter thought. I'm working on a blog about this that I hope to finish soon-ish.
