"Novel model deidentifies data, employs a federated learning model to retain data control."
Kudos @jhalamka, Cris Ross
cc: @bertcmiller @GYNOBioethicist
A few more thoughts about the Mayo announcement.
Read more in the @HealthLeaders article by Mandy Roth:
"Mayo Clinic Designs Novel Data Privacy Model; Announces Venture to Accelerate Drug Discovery"
bit.ly/370GcnG
First...
The Mayo announcement is a BFD among BFDs.
Let me repeat that.
The Mayo announcement is a BFD among BFDs.
Caveat: not easy to wrap your head around this.
@bertcmiller & @GYNOBioethicist wrote about the technological possibilities in @THCBstaff
bit.ly/2QmDvXR
They posed a provocative question:
"What if there was a way for you to allow others to learn from your data without compromising your privacy?"
The Mayo project promises to do exactly that.
Here are a few more details about the Mayo project (from the @HealthLeaders article):
* partnership with @_nference
* Reidentification is prohibited
* Mayo data cannot be combined with data from other sources
* From the article: "As the healthcare industry struggles to find a BALANCE between the desire for innovation and the need to protect patient data"
The word "balance" implies that there is an inherent tradeoff between innovation & data protection.
IMO, ain't necessarily so.
I'd rephrase:
As the healthcare industry struggles to OPTIMIZE the desire for innovation and the need to protect patient data.
"OPTIMIZE" suggests a need to maximize BOTH innovation & data protection and that there isn't necessarily a tradeoff
* Use only deidentified patient data
* Use independent experts to certify that the data has been fully deidentified.
* All parties are prohibited from pooling Mayo Clinic data with other data sources, such as consumer financial or geolocation data.
* Eliminate the possibility of reidentification by employing a federated learning model. While others can query the data set, Mayo Clinic retains full possession of it and only sends participants the results.
re: "federated" tech, see bit.ly/2QmDvXR
more info on "federated" concept: bit.ly/2ZS08WI
* Mayo Clinic's role in the process will be to piece together disease phenotypes and determine how to treat subpopulations of patients differently
Recall when @WSJ revealed details of Ascension deal w/ Google -- that patient ID would be shared:
* Outcries from privacy advocates
* Lashings from legislators & regulators
* Fed investigation by OCR
I'd like to be a mouse in the corner when the Ascension CEO asks the CIO:
"Why didn't we do our deal with Google like Mayo did?"
Let's think a bit further about the Mayo/ @_nerence partnership (federated learning model) & value creation/capture dynamics.
Strikes me that big tech co.s are unlikely to lead a march toward federated learning models.
Why?
Current big tech business model is one of surveillance. Gather as much data as you can, analyze, profile. Rinse, repeat.
Federated learning changes power dynamic between health system & big tech co.s
For a health system like Mayo that can negotiate a deal to keep patient data private, they have much greater opportunity BOTH to create & capture value from data.
This doesn't appear to be lost on Mayo and @MayoInvents.
Remains to be see how many health systems have the foresight, brand, resources, hutzpah like Mayo to negotiate a federated learning model (or blockchain, or ??) with big tech co.s.
Stay tuned.
Strikes me that it would be smart for ALL health systems immediately to shift to insisting on federated learning models in their deals with big tech co.s:
* Opportunity for health systems to create/capture the value from data
* The more health systems that did this, the less influence for big tech co.s
Expect pushback from big tech co.s that want unfettered access to & use of data.
More details on Mayo federated learning model | @FierceBiotech
The project aims to digitize 25 million pathology slides within the next 2 1/2 years, which would provide the single largest source of labeled medical data in the world
-- per @_nference co-founder @in4ence
“Platform business models have been a force of disruption in many sectors"
-- @jhalamka
Notable that @MayoClinic
focus is shifting to biz models, not just platform techologies
I'm thinkin' that Joe-the-Plumber hasn't yet figured out how transformative this @MayoClinic / @MayoInvents
federated learning model could be.
John Halamka ( @jhalamka ), President of Mayo Clinic Platform, posted his own take on their federated learning model:
"Cloud-hosted, de-identified, federated learning in which the tools are brought to the data instead of sending data to the tools"
geekdoctor.blogspot.com/2020/01/a-new-…
Thinking out loud re: further possible implications of @MayoClinic platform partnership w/ @_nference.
Q. What biz model(s) could maximize potential for network effects, especially DATA network effects?
What are "data network effects"?
"When a product’s value increases with more data, and when additional usage of that product yields data..."
bit.ly/3aDWlBJ
Consider alternative federated learning models:
a) AI/ML algorithms can be run on one provider's data -- likely a few million patients
b) AI/ML algorithms can be run on multiple providers' data -- possibly HUNDREDS OF MILLIONS of patients over time
c) ??
Data network effects will grow exponentially based on number of patients.
Could be especially important where n of patients is very low, e.g. rare diseases (HT @bertcmiller )
My hypothesis (speculation) is that win/win biz model is (b).
Why?
From deals announced so far that I've seen, big tech co. (e.g. @Google) partnering w/ providers has been prohibited from combining one provider's data w/ another.
I infer due to:
* fear of PR backlash
* lack of trust of big tech co.
* lack of using a FL model?
* other?
BUT...if the "network" was viewed as controlled by Mayo -- using FL tech -- IMO there's potential for higher level of trust and potential for combining data ACROSS providers.
A win/win biz & clinical model?
Speculative for now but potential is there.
Also worth thinking about biz model implications from Mayo partner POV, i.e, @_nference.
Do they have an opportunity to be a hub/platform across networks of patient data?
The co. secured a $60 M earlier this month, including @MayoInvents
We don't know the terms of the investment or the terms of the contract between @_nference & @MayoClinic,
so it's hard to speculate.
...but worth keeping an eye on this.
Detailed dive into @_nference & @MayoClinic partnership
by @nversel bit.ly/37Ln5OH
"What Nference & Mayo are beginning to do together is to understand the human genotype-phenotype relationship at a scale here that is likely unprecedented"
-- @in4ence, CEO nference
More broadly, the @MayoClinic
Platform increasingly looks COMPETITIVE to the Google Health Platform:
Google Health’s Platform Play: Is Google Creating the Android for Doctors? | @Applico
bit.ly/37MKUpq