Fig. 1: The overview of TrialMind pipeline.
From: Accelerating clinical evidence synthesis with large language models
a It has four main steps: literature search, literature screening, data extraction, and evidence synthesis. b (1) Utilizing input PICO as the review objective, TrialMind generates search terms to identify studies from literature databases. (2) TrialMind suggests the inclusion and exclusion criteria for users to check and edit. Then, TrialMind scans all the identified studies and assesses each individual’s eligibility for each criterion. Last, the user decides on the final strategy to aggregate the criterion-level assessments into an overall eligibility score to rank the eligibility of all identified studies. (3) Given the defined data field, TrialMind processes the content of the involved studies and produces the structured outputs for each data field, grounded on the source indices for user inspection. (4) TrialMind takes the extracted data to create the standardized trial outcomes and works with users to aggregate all involved trial outcomes with a meta-analysis to create new clinical evidence.
