AMEND++ Benchmark Predicts Clinical Trial Eligibility Criteria Amendments
Researchers have introduced a novel task in natural language processing known as eligibility criteria amendment prediction, which aims to anticipate changes in the eligibility parameters of clinical trials. They developed AMEND++, a benchmark suite featuring two datasets: AMEND, which catalogs amendment histories and labels from public trials, and AMEND_LLM, a focused subset that highlights significant changes through a specialized language model denoising. Additionally, they unveiled Change-Aware Masked Language Modeling (CAMLM), a pretraining method utilizing historical modifications to improve learning about amendments. Their validation tests across various baselines confirm the method's capability to address common problems such as delays and rising costs in clinical trials.
Key facts
- Clinical trial amendments often cause delays, increased costs, and administrative burden.
- Eligibility criteria are the most commonly amended component.
- A new NLP task called eligibility criteria amendment prediction is introduced.
- The AMEND++ benchmark suite includes two datasets: AMEND and AMEND_LLM.
- AMEND captures eligibility-criteria version histories and amendment labels from public clinical trials.
- AMEND_LLM is a refined subset curated using an LLM-based denoising pipeline.
- Change-Aware Masked Language Modeling (CAMLM) is a revision-aware pretraining strategy.
- CAMLM leverages historical edits to learn amendment-sensitive representations.
Entities
Institutions
- arXiv