TrialTriage, a Semiautonomous Prescreening Workflow for Resolving Ambiguity in Phase I Oncology Trial Eligibility:...

Area of Science:

• Oncology
• Clinical Trials
• Artificial Intelligence

Background:

• Enrollment in phase I oncology trials is hindered by slow patient identification and evaluation.
• Current systems struggle with missing or uncertain eligibility data, leading to manual reviews and delays.

Purpose of the Study:

• To evaluate TrialTriage, a semiautonomous system designed to resolve eligibility ambiguities in phase I oncology trial prescreening.
• To improve the speed and accuracy of determining patient eligibility for clinical trials.

Main Methods:

• TrialTriage uses large language models for data extraction and a rule engine for classification.
• Ambiguous cases trigger automated investigator email queries for clarification.
• The system was tested on 90 synthetic patient cases generated by three different AI models.

Main Results:

• TrialTriage achieved 100% concordance with ground truth across 90 synthetic cases.
• The system processed cases rapidly, with a mean time of 2.3 minutes per 30 cases.
• Investigator email queries successfully reclassified 4 out of 6 ambiguous cases in a subset test.

Conclusions:

• TrialTriage demonstrates a feasible semiautonomous workflow for resolving eligibility ambiguities.
• Integrating automated email queries into the workflow reduces reliance on manual review.
• Further validation with real-world electronic health record data is recommended.