The Advanced Research Projects Agency for Health (ARPA-H) in September awarded Duality Technologies a contract worth up to $6 million to develop a framework for enabling healthcare organizations to share highly sensitive patient data. If successful, the project will enable smaller healthcare organizations to securely access sensitive health data to conduct research into rare diseases, including those that have a disparate impact on racial minorities.
The phrase “rare disease” is a bit of a misnomer. While some diseases statistically are very rare, the fact is that roughly 20% of the country’s population is affected by a rare disease at some point in their lifetime. And while there is active research into rare diseases, the bulk of it is aimed at people with northwestern European backgrounds and genetics, says Kurt Rohloff, the CTO and co-founder of Duality Technologies.
“There’s much less understanding of the genetics and genetic makeup and mutation correlations between mutations and cancer or other kinds of diseases outside of the classic focus of northern and western European heritage individuals,” Rohloff says. “We have a bit of an institutional bias in the world.”
Very large healthcare organizations, such as the Broad Institute, Mass General, and Intermountain Health have a large amount of valuable data themselves to conduct medical research on things like rare diseases. However, much of the data they have is skewed toward population centers with a European genetic heritage, Rohloff says.
The good news is that if those large healthcare organizations want a data set from a certain city, they have the legal resources to write data use agreement that provides the necessary privacy protections.
“There’s nothing untoward about it. They have administrative policies about how they handle the data when they take it in to keep it private and secure. All best practices. They do it right,” Rohloff tells BigDATAwire.
“The challenge is, as you go to the smaller organizations, the mid-market health centers, research centers, mid-tier university research centers, they don’t necessarily have infinite resources for legal budgets,” he continues. “They don’t have infinite IRB [institutional review board] kind of activities. They need basically ways of accelerating access to data without necessarily having lawyer time.”
That’s the goal of the new ARPA-H project that it has started. Dubbed SQUEEZES, the project will use Duality’s fully homomorphic encryption (FHE) technology to enable rural and native healthcare organizations in the United States to pool together their healthcare data and analyze it, but without enabling each other to read it.
The healthcare organizations will still need to get consent from individuals before using their data for research into rare diseases. But since the data remains encrypted the entire time, the amount of legal work required to obtain the necessary consent is reduced, Rohloff says.
“All these various [organizations]… have their own data,” says Rohloff, who has worked broadly in the DARPA community with Duality’s homomorphic encryption technology. “An organization would encrypt their data locally, using a local encryption key…and upload it to a server, which might be at a cancer research center. And multiple rural or tribal health agencies might do this, each encrypting with their own key.”
Once all the encrypted data is centralized, it can be analyzed and used to build machine learning models within Duality’s FHE environment.
“This might be, for example, covariate-based models or just simple correlation-type models to identify what kind of mutations are indicative of certain kinds of cancers,” says Rohloff, who has a PhD in electrical engineering and computer science (EECS) from the University of Michigan. “You get higher quality models, more accurate results, and potentially see things you wouldn’t normally see.”
It’s a form of federated learning with FHE mixed in, he says.
“All this is being done encrypted. You run the analytics, get out an encrypted results,” Rohloff says. “We could send the encrypted result back to each of the health agencies that contributed data. Each can run basically an approval process with their local key…to basically grant access to the analytic party, so that if all of the contributors of data of encrypted data agree or give consent of access … then the analytic party then eventually is able to get the result.”
Building these types of systems is not easy, Rohloff says. While FHE has gotten a bad rap in some circles due to poor performance, those are mostly due to poor implementations.
“It does take a bit of a fine touch and a bit of experience to design workloads that run very efficiently on top of the privacy tech,” says Rohloff, who won a DARPA Young Faculty Award while working at the New Jersey Institute of Technology. “There’s a performance penalty for doing homomorphic encryption poorly.”
Duality has already built this type of system before, including in partnership with the Dana-Farber Cancer Institute, the Tel Aviv Medical Center, and others, Rohloff says. The system it’s developing as part of the ARPA-H project is designed to be a reference implementation of an open architecture for FHE that can be deployed more widely.
Whether Duality can become the Red Hat of FHE remains to be seen. The company is an undisputed leader in homomorphic encryption, which has been proven to work. As the company builds its commercial base, it’s happy to do some good works along the way.
“A big part of our mission is enabling secure collaboration on sensitive data,” Rohloff says. “Whether it’s helping organizations share in a privacy protected, regulated manner, financial transaction data to go after financial crime, cut down fraud, stop money laundering, or counter terror financing–or if it’s on the civil public health side of helping cancer research centers to share data to develop better treatments for rare diseases and help historically underrepresented and underserved communities like tribal health centers and rural health centers–this is a big part of what we do: Enabling secure collaborations for the public good overall.”
Related Items:
Will New Government Guidelines Spur Adoption of Privacy-Preserving Tech?
In Homomorphic Encryption We (Must) Trust
Three Privacy Enhancing Techniques That Can Bolster the COVID-19 Response
The post ARPA-H Taps Duality’s Homomorphic Encryption for Rare Disease Research appeared first on BigDATAwire.
0 Commentaires