Benchmark
TrialGPT
A benchmark dataset for evaluating large language models on clinical trial patient matching, containing matched samples of patient records and trial eligibility criteria.
Timeline (last 90 days)
Oct 27
The study drew 2,048 test patient-condition pairs from TrialGPT and from question-answering datasets built on MIMIC-III and MIMIC-IV.
Unverified50%
Oct 27
In batch mode, the macro F1 scores of JEV and Qwen on TrialGPT were 0.652 and 0.660, respectively.
Unverified50%