AI&DHI Innovation Spotlight: Adaptation, validation, and implementation of an AI prostate cancer clinical trial cohort identification tool
This edition of the AI&DHI Innovation Spotlight features work led by Dr. Kristian Stensland, Assistant Professor of Urology and Learning Health Sciences at Michigan Medicine.
Dr. Stensland and his team are using AI&DHI’s MiCohort Screener tool to identify patients who may be eligible for prostate cancer clinical trials at U-M. The team is also applying methods to test how well this system works in practice.
Clinical trials are essential for the development of new cancer treatments; however, many struggle to recruit enough participants, with 1 in 5 failing due to low enrollment. A key barrier to clinical trial enrollment is the time- and labor-intensive process of determining which patients meet a trial’s eligibility criteria. Currently, research coordinators must spend weeks manually reviewing patient charts, resulting in delayed access to promising clinical research.
Supported by a 2-year, $60,000 grant from the National Cancer Institute (NCI) Specialized Programs of Research Excellence (SPORE) Career Enhancement Program at U-M, Dr. Stensland’s team aims to address this challenge by examining the effectiveness of large language models (LLMs) in streamlining the process of identifying participants for prostate cancer trials using clinical data from U-M Health.
The team’s study will involve the refinement and testing of MiCohort Screener, an AI-assisted tool developed by AI&DHI that automatically screens patient charts against a study’s inclusion and exclusion criteria. MiCohort Screener can analyze hundreds of patient charts against multi-criterion protocols in a single run, cutting the time needed for review down from weeks to minutes.
“AI&DHI has been an outstanding partner throughout this work. Their continued support will help us better understand how AI can responsibly enhance clinical trial recruitment and, ultimately, expand patients’ access to the latest in cancer research.”
Additionally, since MiCohort Screener uses UM GPT, the university’s own institutionally hosted, HIPAA-approved LLM, it offers the team a way of navigating around the challenges associated with adopting externally developed AI tools in clinical settings, including cost, limited real-world validation and concerns around data privacy and security.
“There have been many recent advances in LLMs that have created exciting opportunities for streamlining tasks that once required extensive manual effort,” said Dr. Stensland. “However, these approaches are mostly expensive, unproven in actual clinical settings and are owned by entities outside of our institution. By developing, implementing, and evaluating our own AI/LLM solution to patient identification, we will have more control over its deployment, be able to better understand its value and ensure that the resources required to run it are worth it.”
AI&DHI provided Dr. Stensland and his team with early beta access to MiCohort Screener, giving the researchers the opportunity to rigorously evaluate and refine the technology in a real clinical environment.
“AI&DHI has been an outstanding partner throughout this work,” said Dr. Stensland. “Their continued support will help us better understand how AI can responsibly enhance clinical trial recruitment and, ultimately, expand patients’ access to the latest in cancer research.”
About AI & Digital Health Innovation
AI & Digital Health Innovation (formerly Precision Health at U-M) is dedicated to empowering researchers at the University Michigan to change the future of digital healthcare. They work with multi-disciplinary teams of health providers, basic scientists, engineers, and administrators to tackle the most difficult research problems and help rapidly bring ideas to the bedside. For more information visit aidhi.umich.edu.