ACTIVATE: AI-driven Clinical Trial Information and Viability Assessment Tool for EHRs

PROJECT SUMMARY / ABSTRACT Clinical trials are critical to development of new cancer therapeutics. However, most adults with cancer do not participate in trials, and many trials struggle to reach their accrual goals. Modern cancer trials have complex eligibility criteria, often requiring that patients have tumors with specific biomarkers and clinical histories. The data necessary to confirm eligibility are often recorded only in unstructured form within electronic health records. This makes it difficult for cancer centers to assess the feasibility of a potential trial by estimating the number of patients at each center who may be eligible. It also contributes to the problem of low trial participation rates by creating barriers for oncologists trying to identify potential clinical trials for their patients, and for investigators to identify potential patients for their trials, in real time. This project, AI-driven Clinical Trial Information and Viability Assessment Tool for EHRs (ACTIVATE), aims to develop and deploy open-source AI tools to improve cancer clinical trial feasibility and recruitment. We will build on our existing MatchMiner tool, which matches patients to trials based on molecular criteria, extending it to include other core clinical variables, including cancer type, cancer stage/burden, treatment history, and key biomarkers. The project will use novel AI methods to extract these variables from longitudinal EHR text and clinical trial protocols, and it will enable cross-site sharing of patient phenotyping models using privacy-preserving techniques. We will create software frontends for cancer centers, trial investigators, and oncologists to use these AI tools for two main purposes: (a) feasibility assessment for sites considering a specific clinical trial via estimation of the number of eligible patients at the site; and (b) patient-trial matching, based on real-time identification of appropriate trial options for patients who need new treatments. The project has three specific aims: (1) Validate and deploy the Patient Recruitment Optimized Matching Pipeline Technology (PROMPT-AI) pipeline for efficiently extracting key clinical variables for trial eligibility from longitudinal EHR text and clinical trial documents; (2) build and test TrialForecast, a software package for cancer trial site feasibility assessment based on PROMPT-AI outputs; and (3) Deploy and evaluate TrialMatch, an open-source software tool to match patients to clinical trials using the PROMPT-AI pipeline. The project will evaluate the impact of the AI tools on trial accrual rates at Dana- Farber Cancer Institute (DFCI) and Mayo Cancer Center. By creating open-source trial matching tools that do not depend on commercial entities, our goal is to enhance the efficiency and effectiveness of clinical cancer research, ultimately accelerating the pipeline of new treatments for patients.