Understanding unrealized trial enrollments following patient-to-trial matching with large language models

Clinical trials are essential for advancing cancer care, but identifying eligible patients in surgical clinics can be challenging due to the manual and time-consuming enrollment process. Artificial intelligence tools, such as large language models, have the potential to automate aspects of clinical trial matching. This study identified reasons why patients did not enroll in a clinical trial after receiving a recommendation from OncoLLM-MCW, a large language model-based platform for matching patients to clinical trials.

This retrospective study included patients seen by surgeons participating in a prospective pilot using a large language model-based platform within gastrointestinal surgical oncology clinics between July and December 2024. OncoLLM-MCW, a fine-tuned large language model trained on institutional clinical data and oncology guidelines, was used to identify eligible clinical trial matches. Each week, new patients were processed, and matches were shared with provider teams for review. The primary outcome was unrealized trial enrollment, defined as a match that did not result in enrollment. Secondary outcomes included reasons for non-enrollment.

Using OncoLLM-MCW, 514 patients were evaluated, resulting in 34 trial matches across 32 patients. Of these, 9 matches (26.5%) resulted in enrollment, whereas 25 (73.5%) did not. Among the 25 non-enrolling matches, 9 (36%) patients were ineligible, 5 (20%) declined participation, 4 (16%) were not enrolled due to provider discretion, and 7 (28%) had no documented reason.

OncoLLM-MCW automated clinical trial matching in surgical clinics, screening more than 500 patients. Identifying and addressing reasons for unrealized enrollments may optimize accrual and advance cancer care.