Designing AI-Assisted Clinical Trial Matching
St. Jude's referral process was broken, parents overwhelmed, doctors guessing, staff buried in unqualified submissions. I redesigned it with AI-assisted tools that guided users, matched patients to trials intelligently, and connected eligibility directly to the referral form.
My Responsibilities
Impact of the design
73%
More
Clinician AI adoption
93%
Less
Abandonment rate
10mins
from 2 hrs
Time per referral
St. Jude Children’s Research Hospital is a leading center for treating childhood cancer and other pediatric diseases, with all care provided free to families. It also drives research to find cures worldwide. This summer, I worked with the Strategic Communication, Education & Outreach (SCEO) department on improving patient-facing digital platforms.
I interviewed Stakeholders, Business Analysts, Accessibility Expert and recruited Parents, and Healthcare Staff via Lyssna to understand current goals, pain points, and process gaps.
Why were the Users struggling?
Doctors and parents hit the same walls: eligibility criteria written in medical jargon, no filtering, no guidance. With nowhere to turn, they picked up the phone. The referral office was drowning in the same questions on repeat.
"I read the same page four times and still didn't know if she qualified. I just called and hoped someone could help"
Cross-referencing a patient's profile against 50+ open trials meant opening tabs, reading dense paragraphs, and hoping nothing was missed. One overlooked line meant starting the referral over.
"I missed a line about prior treatment history. We had to restart the whole referral."
Parents were submitting referrals without knowing if their child even qualified — flooding the referral staff with incomplete, unqualified submissions. There was no way to attach medical records. And every field was filled from scratch, even information already provided earlier.
"I submitted the form and waited weeks, only to find out my daughter didn't even qualify for any open trials."
Healthcare taught me that AI's job isn't to replace human judgment, it's to reduce cognitive load so humans decide better, faster. Surface the right information, flag uncertainty, keep the expert in control. That's the principle behind every AI authoring tool in pharma today.
I'd build a real-time physician dashboard to close the loop between submission and outcome, let the AI model learn from accepted and rejected referrals, and measure whether the eligibility checker actually reduced unqualified submissions, using that data to sharpen the confidence scoring model.