Manual Versus AI-Assisted Clinical Trial Screening Using Large-Language Models

Manual Versus AI-Assisted Clinical Trial Screening Using Large-Language Models

A prospective randomized controlled trial comparing manual review and AI screening for patient eligibility determination and enrollments. A structured query will identify potentially eligible patients from the Mass General Brigham Electronic Data Warehouse (EDW), who will then be randomized into either the manual review arm or the AI-assisted review arm.

Screening participants for clinical trials is a critical yet challenging process that requires significant time and resources. Traditionally, patient screening has been manual, relying on the diligence and judgment of study staff. However, manual screening is prone to human error and inefficiencies, contributing to high costs and prolonged trial durations.

Recent advancements in natural language processing (NLP) and large language models (LLMs), such as GPT-4, offer potential solutions to improve the accuracy, efficiency, and reliability of the screening process. Retrieval-Augmented Generation (RAG)-enabled systems, like RECTIFIER, have shown promise in enhancing clinical trial screening by automating the extraction and analysis of relevant data from electronic health records (EHRs).

In the investigators' previous study, RECTIFIER demonstrated high accuracy in screening patients for clinical trials, aligning closely with expert clinician reviews and outperforming manual study staff in several criteria. It underscored the potential for LLMs to transform clinical trial screening, making it more efficient and cost-effective while maintaining high standards of accuracy and reliability. However, before RECTIFIER is scaled to be used across many domains of clinical trials, it should be validated prospectively in the real-world setting to enroll patients.

In the Co-Operative Program for Implementation of Optimal Therapy in Heart Failure (COPILOT-HF) trial (NCT05734690), the investigators will identify potential participants through EHR queries followed by manual review, which provides an opportunity for RECTIFIER to improve the screening process. By leveraging RECTIFIER, this study aims to evaluate the effectiveness of automated AI screening compared to traditional manual methods for enrollments of patients into an ongoing clinical trial.

Inclusion Criteria:

• Documented diagnosis of heart failure (e.g., ICD-9 codes 428 ICD-10 codes I50 or Problem list in the electronic health record)
• Most recent left ventricular ejection fraction (LVEF) assessed within the past 24 months
• Seen Mass General Brigham provider within the last 24 months

Exclusion Criteria:

• LVEF <50% currently prescribed or intolerant to an evidence-based beta-blocker, ARNI, MRA, and SGLT2i at least 50% goal dose
• LVEF>50% currently prescribed or intolerant to SGLT2i
• Systolic blood pressure (SBP) <90 mmHg at last measure