p{AI}tient

Inspiration

Healthcare is a very traditional and rigid industry, where it is hard to replace decade long traditions - I found about healthcare de-identification while working with a healthcare provider looking to share their data. I found the process to be quite lengthy and inefficient, especially since it was done manually. With the ongoing presence of AI in the medical field, I had the idea of fully automating the process of de-identification as a means to replace the manual honest broker system of de-identification in the status quo. I am particularly passionate about this idea because it eliminates any presence of human bias, potential harm to patient safety, and speeds up the research process.

What it does

p{AI}tient is a fully automated de-identification tool that takes a patient file, removes any personal identifiers according to HIPAA's safe harbor and returns the same data with retracted personal identifier values (replaced with a "xxxxx"). This process currently occurs on a web portal open to all users.

How we built it

We used react and node.js along with javascript, html, css to build a web app that takes in a patient form PDF file and uses the openAI API to remove key identifiers in the patient form and outputs both the updated form and the original form to the frontend.

Challenges we ran into

It was very hard to transition into patient privacy data so we had to do a lot of research into HIPAA's legal framework - we also were tasked with creating a product that had a simple ease of integration and did not require a full on platform onboarding (something that other products have offered and researchers have steered away from consequently)

Accomplishments that we're proud of

We are proud of creating a fully functional web app with our most important features that accurately removes identifiers from a patient form PDF.

What's next for p{AI}tient

As we continue working on this product, we want to increase file intake, partner with large academic institutions and hospitals, and buy synthetic data to build our own model.