Automating Chart Review Using an Artificial Intelligence−Enabled System for Assessing Transthyretin Amyloid Cardiomyopathy Trial Eligibility

Due to the complexity of contemporary EMRs, clinical research remains heavily reliant on time-intensive and costly manual chart reviews done by medically knowledgeable personnel. (1) Prior automated chart review efforts have demonstrated the effectiveness of large language-model (LLM)-driven screening for trial criteria; however, generalizability of these studies has been limited due to 1) their reliance on simulated patient notes lacking the complexity of real-world electronic medical record (EMR) data that exists across hundreds of notes and dozens of note types for an individual patient, 2) their use of either structured data or unstructured data criterion assessments without synthesizing both data types to make singular conclusions, and 3) the utilization of binary criterion answers without auditable textual justifications to support the assessments, making verification more cumbersome.