“The only way to get better health care is to get better data and the only way to get better data is to incentivize its release.” - Marzyeh Ghassemi, Assistant Professor, MIT
Machine learning models hold great promise in radically improving patient care and outcomes. They can make healthcare more cost-effective, unbiased, personalized, and equitable. But in order for the machine learning models to be the most helpful, they must be trained on large amounts of clean, accurate, and complete data.
Given the privacy and technology restrictions common in the healthcare system, large amounts of useful data are not easily and cheaply accessible. Researchers are left to piece together training data from public, disparate, or unreliable sources.
Solution:
A shared repository of healthcare data where personal identifiable information is encrypted and researchers can train models with controlled access.
- Zero-knowledge proofs can be used to prove that a model was used to make predictions without leaking any information about the model itself.
- It may be possible to also use zero-knowledge proofs to train models without ever accessing the actual data.
- Incentives can be built in the form of digital assets to promote the sharing of data to this shared repository while still maintaining compliance with privacy and security regulations.
- Have a global repository that public and private institutions can contribute to in order to create the most comprehensive and current store of healthcare data from around the world.
Incredible potential for lasting impact is currently locked away because of the silos of healthcare data; blockchains are uniquely suited to unleash a new wave of healthcare research and innovation.