A multi-modal EHR-based approach to screen patients for oncology clinical trials.

Abstract

411 Background: Inefficiencies in patient (pt) screening may impair enrollment in oncology trials and could potentially be mitigated by a centralized trial matching infrastructure. The objective of this study was to implement a novel clinical trial screening service using electronic health record (EHR)-based structured variables, machine learning (ML) and human abstraction. Methods: To construct the model, we defined a set of inclusion and exclusion (IE) criteria unique to each trial and routinely documented in EHRs. Patients that met IE criteria via structured and ML-extracted clinical data were selected for review by trained specialists via a technology-enabled abstraction platform. Real-time notifications for eligible pts were issued via EHR alerts to research teams at participating study sites. Results: Between January and May 2024, over 80,000 pts were screened prior to a clinic visit for two trials at three sites. Using structured IE criteria alone, 98.7% and 99.6% of pts were excluded. An additional 31.9% of pts were excluded when employing ML modeling. A total of 15 (3.0%) and 5 (4.0%) pts were deemed eligible when using structured, ML, and abstraction-based criteria (Table). Seventeen percent of surfaced pts were consented within 10 days of implementation. Median time for abstractor confirmation of pt eligibility was 13.3 hours (interquartile range: 2.8, 27.4) with abstraction of highest priority pts taking a median of 3.0 and 9.9 hours, respectively. Conclusions: Our results support the implementation of a centralized, technology-enabled pt screening service for effective trial eligibility assessments across large patient populations. The value of using structured, ML- and abstraction-based data models, improved accuracy by more than 95% compared to structured data alone. The impact of multi-modal centralized screening is being evaluated across additional clinical trials in a nationwide community oncology research network. Pts, No. (% pts excluded). Trial 1 Trial 2 1. Pts with upcoming visit 37,865 44,973 2. Pts meeting structured criteria (including: age, stage, prior IV therapy) 497 (98.7) 182 (99.6) 3. Pts meeting ML model-based criteria (including: prior oral therapy) - 124 (31.9) 4. Pts meeting steps 1-3 and abstraction criteria (including: cancer diagnosis, biomarker status, prior malignancies) 15 (97.0) 5 (96.0)