Research map / Large language models / Agents, code and math
EHRStruct: A Comprehensive Benchmark Framework for Evaluating Large Language Models on Structured Electronic Health Record Tasks
Xiao Yang, Xuejiao Zhao, Zhiqi Shen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use general LLMs instead of medical LLMs | EHRStruct zero-shot · EHRStruct (Synthea) | accuracy / AUC | Improved |
| add EHRMaster (plan, concept alignment, adaptive code execution) to Gemini 2.5 | Gemini 2.5 · EHRStruct D-R3 (Synthea) | accuracy: 85 -> 97 | Improved |
| add EHRMaster | Gemini 2.5 · EHRStruct K-R3 (Synthea) | AUC: 58.4 -> 69.2 | Improved |
| add EHRMaster | Gemini 2.5 · EHRStruct K-R1 (Synthea) | AUC: 66.3 -> 59.3 | Hurt |
| add EHRMaster | Gemini 2.0 · EHRStruct D-R4 (Synthea) | accuracy: 90 -> 80 | Hurt |
| add EHRMaster | Gemini 1.5 · EHRStruct D-R1 (Synthea) | accuracy: 80 -> 96 | Improved |
| few-shot prompting (1-shot, 3-shot) instead of zero-shot (ablation) | Gemini 1.5/2.0/2.5 · EHRStruct Knowledge-Driven tasks | AUC | Improved |
| use 5-shot instead of 1-shot or 3-shot (ablation) | Gemini series · EHRStruct Knowledge-Driven tasks | AUC | Hurt |
| present table as natural language description (ablation) | Gemini and GPT series · EHRStruct Data-Driven reasoning | accuracy | Improved |
| present table as graph-structured representation (ablation) | LLMs · EHRStruct Data-Driven understanding | accuracy | Improved |
| change input format for Knowledge-Driven tasks (ablation) | LLMs · EHRStruct Knowledge-Driven | AUC | No clear effect |
| LoRA multi-task fine-tuning instead of single-task fine-tuning (ablation) | Qwen-7B · EHRStruct (all categories) | accuracy / AUC | Improved |
Closest papers by meaning
- EHRNoteQA: An LLM Benchmark for Real-World Clinical Practice Using Discharge Summaries (NeurIPS 2024 · similarity 0.90)
- Context Clues: Evaluating Long Context Models for Clinical Prediction Tasks on EHR Data (ICLR 2025 · similarity 0.89)
- SM3-Text-to-Query: Synthetic Multi-Model Medical Text-to-Query Benchmark (NeurIPS 2024 · similarity 0.89)
- MedAlign: A Clinician-Generated Dataset for Instruction Following with Electronic Medical Records (AAAI 2024 · similarity 0.89)
- MedJourney: Benchmark and Evaluation of Large Language Models over Patient Clinical Journey (NeurIPS 2024 · similarity 0.88)
- Instruction Tuning Large Language Models to Understand Electronic Health Records (NeurIPS 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
