Introduction: Enrollment in hematologic malignancy trials remains low, especially among older adults, racial and ethnic minoritized groups, and non-English speakers. Many eligible patients are missed due to limitations of manual screening. We implemented an artificial intelligence (AI)-powered Clinical Trial Patient Matching (CTPM) system to increase identification and enrollment for two interventional trials: a randomized phase 3 trial in lower-risk MDS and a phase 1b/2 trial in relapsed/refractory myeloma.

Methods: From August 2024 to July 2025, we implemented CTPM across 13 Yale Cancer Center sites. On average, the system reviewed ~74,000 patients with upcoming visits per week. Of these, ~550 were screened weekly for the myeloma trial and ~330 for the MDS trial. Weekly automated prescreening used trial-specific criteria and both structured and unstructured electronic medical record data. Identified patients were manually reviewed to confirm eligibility before outreach. We tracked identified, eligible, and enrolled patients. We compared accrual trends pre- and post-CTPM (MDS trial: August 2024; Myeloma trial: November 2024). IRB approval was obtained for independent evaluation of the CTPM, and approvals were also obtained to use the CTPM as a recruitment tool for both trials. Screening sensitivity was set high to avoid missed cases during this initial implementation in hematology trials.

Results: In the MDS trial, CTPM identified 216 patients. Of those, 151 were adults over 70, 15 were Black/African American, 15 were Hispanic/Latino, and 17 were non-English speakers (NES). After manual review, only 11 patients were deemed potentially eligible. The reasons for ineligibility included prior disallowed treatments (n=53); history of concurrent malignancy, transplant, or progression to AML (n=45); incomplete information (n=41); incorrect diagnosis (n=38); disease characteristics not meeting trial criteria (n=23); exclusion due to laboratory finding or comorbidities (n=17); and enrollment in another clinical trial (n=2).

Among the 11 eligible patients, 9 were over 70, 1 was Black/African American, and 2 were NES. Only one patient ultimately enrolled in the trial; this individual was White and over 70. Of the 10 non-enrollees, 7 declined participation (preference for standard of care or trial hesitancy) and 2 became ineligible due to clinical progression. Community sites contributed 6 potentially eligible patients but none enrolled.

In the myeloma trial, 147 patients were identified. Among them, 74 were over 70, 25 were Black/African American, 12 were Hispanic/Latino, and 5 were NES. After manual review, 10 patients were found potentially eligible. The reasons for ineligibility included absence of active progression (CR, VGPR, or PR; n=43), incomplete information (n=28), early initiation of next-line therapy (n=18), exclusion due to lab abnormalities or comorbidities (n=17), enrollment in or prior enrollment in this trial (n=4), and concurrent malignancy (n=4).

Among the 10 eligible patients, 8 were over 70, 1 was Black/African American, and 1 was NES. Five patients enrolled in the trial–3 were over 70, 1 was Black/African American, and all were English-speaking. Of the 5 non-enrollees, 3 started standard of care therapy by the time communication with the site occurred, 1 was deferred by the provider, and 1 progressed clinically. Community sites contributed 8 eligible patients, 3 of whom enrolled. Accrual improved with CTPM implementation from one patient every 2.6 months to one every 1.5 months. All patients enrolled during the study period were identified via CTPM.

Conclusions: The CTPM system increased identification of patients across all demographic groups, but substantial attrition occurred at both the eligibility and enrollment stages. The CTPM appears to increase enrollment into the myeloma trial, which targeted relapsed/refractory patients and was non-randomized, compared to the randomized nature and the lower risk status of the MDS trial. Higher enrollment in the myeloma trial may reflect preference for guaranteed therapy over randomization. These findings underscore the importance of combining automated prescreening with patient-centered trial design, decentralized trial access, and navigation services to improve enrollment in hematologic malignancies.

This content is only available as a PDF.
2025
Sign in via your Institution