MedConclusion: A Benchmark for Biomedical Conclusion Generation from Structured Abstracts
This provides a reusable benchmark for studying evidence-to-conclusion reasoning in biomedical domains, but it is incremental as it builds on existing data and evaluation methods.
The authors tackled the lack of resources for testing LLMs' ability to infer scientific conclusions from structured biomedical evidence by introducing MedConclusion, a dataset of 5.7M PubMed structured abstracts, and found that conclusion writing is distinct from summary writing with strong models clustering under current metrics.
Large language models (LLMs) are widely explored for reasoning-intensive research tasks, yet resources for testing whether they can infer scientific conclusions from structured biomedical evidence remain limited. We introduce $\textbf{MedConclusion}$, a large-scale dataset of $\textbf{5.7M}$ PubMed structured abstracts for biomedical conclusion generation. Each instance pairs the non-conclusion sections of an abstract with the original author-written conclusion, providing naturally occurring supervision for evidence-to-conclusion reasoning. MedConclusion also includes journal-level metadata such as biomedical category and SJR, enabling subgroup analysis across biomedical domains. As an initial study, we evaluate diverse LLMs under conclusion and summary prompting settings and score outputs with both reference-based metrics and LLM-as-a-judge. We find that conclusion writing is behaviorally distinct from summary writing, strong models remain closely clustered under current automatic metrics, and judge identity can substantially shift absolute scores. MedConclusion provides a reusable data resource for studying scientific evidence-to-conclusion reasoning. Our code and data are available at: https://github.com/Harvard-AI-and-Robotics-Lab/MedConclusion.