Can LLM Agents Generate Real-World Evidence? Evaluating Observational Studies in Medical Databases

TL;DR

This paper evaluates the ability of large language model agents to generate real-world observational evidence in medical databases, revealing significant limitations and variability in performance across tasks and scaffolds.

Contribution

Introduces RWE-bench, a novel benchmark for assessing LLM agents in executing and constructing evidence bundles from real-world medical data, with comprehensive evaluation and error analysis.

Findings

Best agent achieves 39.9% task success rate

Agent scaffold choice significantly impacts performance

Automated error localization aids in identifying failure modes

Abstract

Observational studies can yield clinically actionable evidence at scale, but executing them on real-world databases is open-ended and requires coherent decisions across cohort construction, analysis, and reporting. Prior evaluations of LLM agents emphasize isolated steps or single answers, missing the integrity and internal structure of the resulting evidence bundle. To address this gap, we introduce RWE-bench, a benchmark grounded in MIMIC-IV and derived from peer-reviewed observational studies. Each task provides the corresponding study protocol as the reference standard, requiring agents to execute experiments in a real database and iteratively generate tree-structured evidence bundles. We evaluate six LLMs (three open-source, three closed-source) under three agent scaffolds using both question-level correctness and end-to-end task metrics. Across 162 tasks, task success is low: the best agent reaches 39.9%, and the best open-source model reaches 30.4%. Agent scaffolds also matter substantially, causing over 30% variation in performance metrics. Furthermore, we implement an automated cohort evaluation method to rapidly localize errors and identify agent failure modes. Overall, the results highlight persistent limitations in agents' ability to produce end-to-end evidence bundles, and efficient validation remains an important direction for future work. Code and data are available at https://github.com/somewordstoolate/RWE-bench.