Elamur Research⌕
☰
home / papers / 2026 / Neural at ArchEHR-QA 2026: One Method Fits All: Unified Prompt Optimization for Clinical QA over EHRs

Neural at ArchEHR-QA 2026: One Method Fits All: Unified Prompt Optimization for Clinical QA over EHRs

Paper · 2026 · Abrar Majeedi, Viswanatha Reddy Gajjala, Sai Prasanna Teja Reddy Bogireddy et al.

Summary

The paper describes Neural1.5, a system for the ArchEHR-QA 2026 shared task at CL4Health@LREC 2026. The task asks systems to answer patient questions from electronic health record (EHR) notes and to tie every claim to specific note sentences. It has four parts: rewriting a patient's question as a short clinician-style query, picking the evidence sentences, writing a grounded answer, and linking answer sentences to supporting evidence.

The authors avoid fine-tuning. Each subtask becomes a separate DSPy program, and the MIPROv2 optimizer tunes its instructions and few-shot examples against an objective specific to that stage. Reliability comes from two extra mechanisms:

  • repeated stochastic runs combined by majority vote or consolidation;
  • for alignment, a three-step chain of initial linking, self-reflection and chain-of-verification.

Among teams that entered all four subtasks, Neural1.5 placed second overall. It took first place on evidence identification. The authors conclude that tuning prompts separately for each stage is a cheaper practical alternative to fine-tuning for multi-part clinical QA, where labelled data is scarce.

Key points

  • Placements were 4th, 1st, 4th and 7th on Subtasks 1–4, giving a mean rank of 4.00. That was second among teams that completed every subtask.
  • Evidence identification (Subtask 2) reached a Strict Micro F1 of 63.7. Labels came from a majority vote over 5 runs at temperature 0.8.
  • Question interpretation was optimized with a weighted metric: semantic alignment judged by an LLM (60%), preservation of key terms (25%) and question structure (15%). Outputs were capped at 15 words.
  • For answer generation, 5 candidates were sampled at temperature 0.9. A consolidation prompt then kept only claims shared across candidates, within a 75-word limit.
  • The alignment pipeline was run several times. A link was kept only if it won a majority of votes and its average confidence was at least 0.9.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓University of Wisconsin–MadisonMadison · United States54 works · 49,089 citations

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksMedicine · 275 worksHealth Informatics · 214 works

🧭 Topics 1Research subjects and languages the work is about.Question answering · 487 works

Cite

Cite as Abrar Majeedi, Viswanatha Reddy Gajjala, Sai Prasanna Teja Reddy Bogireddy and Siddhant Rai. 2026. Neural at ArchEHR-QA 2026: One Method Fits All: Unified Prompt Optimization for Clinical QA over EHRs. Proceedings of the Third Workshop on Patient-Oriented Language Processing (CL4Health) @ LREC 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/52yxj7aduxxg

@inproceedings{majeedi2026neural,
  title = {{Neural at ArchEHR-QA 2026: One Method Fits All: Unified Prompt Optimization for Clinical QA over EHRs}},
  author = {Abrar Majeedi, Viswanatha Reddy Gajjala, Sai Prasanna Teja Reddy Bogireddy and Siddhant Rai},
  booktitle = {Proceedings of the Third Workshop on Patient-Oriented Language Processing (CL4Health) @ LREC 2026},
  year = {2026},
  doi = {10.63317/52yxj7aduxxg},
  url = {https://doi.org/10.63317/52yxj7aduxxg}
}

References

14 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Brian G Arndt, Mark A Micek, Adam Rule, Christina M Shafer, Jeffrey J Baltus, and Christine A Sinsky. 2024. More tethered to the EHR: EHR workload trends among academic primary care physicians, 2019–2023. Annals of Family Medicine, 22(1):12–18. find
  2. Sai Prasanna Teja Reddy Bogireddy, Abrar Majeedi, Viswanath Gajjala, Zhuoyan Xu, Siddhant Rai, and Vaishnav Potlapalli. 2025. Neural at archehr-qa 2025: Agentic prompt optimization for evidence-grounded clinical question answering. In Proceedings of the 24th Workshop on Biomedical Language Processin… find
  3. Nader Karayanni, Aya Awwad, Chein-Lien Hsiao, and Surish P Shanmugam. 2024. Keeping experts in the loop: Expert-guided optimization for 1 https://github.com/ bogireddytejareddy/ArchEHR-QA-Neural clinical data classification using large language models. arXiv preprint arXiv:2412.02173. — [arXiv:2412.02173](https://arxiv.org/abs/2412.02173) ↗
  4. Omar Khattab, Arnav Singhvi, Paridhi Maheshwari, Zhiyuan Zhang, Keshav Santhanam, Sri Vardhamanan, Saiful Haq, Ashutosh Sharma, Thomas T. Joshi, Hanna Moazam, Heather Miller, Matei Zaharia, and Christopher Potts. 2024. Dspy: Compiling declarative language model calls into self-improving pipelines. I…
  5. Chin-Yew Lin. 2004. Rouge: A package for automatic evaluation of summaries. In Text summarization branches out, pages 74–81.
  6. Krista Opsahl-Ong, Michael J Ryan, Josh Purtell, David Broman, Christopher Potts, Matei Zaharia, and Omar Khattab. 2024. Optimizing instructions and demonstrations for multi-stage language model programs. arXiv preprint arXiv:2406.11695.
  7. Anusri Pampari, Preethi Raghavan, Jennifer Liang, and Jian Peng. 2018. emrqa: A large corpus for question answering on electronic medical records. arXiv preprint arXiv:1809.00732.
  8. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. Bleu: a method for automatic evaluation of machine translation. In Proceedings of the 40th Annual Meeting on Association for Computational Linguistics, ACL ’02, page 311–318, USA. Association for Computational Linguistics.
  9. Karan Singhal, Tao Tu, Juraj Gottweis, Rory Sayres, Ellery Wulczyn, Mohamed Amin, Le Hou, Kevin Clark, Stephen R Pfohl, Heather Cole-Lewis, et al. 2025. Toward expert-level medical question answering with large language models. Nature Medicine, pages 1–8. find
  10. Sarvesh Soni and Dina Demner-Fushman. 2026a. A dataset for addressing patient’s information needs related to clinical course of hospitalization. Scientific Data. find
  11. Sarvesh Soni and Dina Demner-Fushman. 2026b. Overview of the archehr-qa 2026 shared task on grounded question answering from electronic health records. In Proceedings of the Third Workshop on Patient-Oriented Language Processing (CL4Health), Palma, Mallorca (Spain). ELRA. find
  12. Sarvesh Soni, Soumya Gayen, and Dina Demner-Fushman. 2025. Overview of the archehr-qa 2025 shared task on grounded question answering from electronic health records. In Proceedings of the 24th Workshop on Biomedical Language Processing, pages 396–405, Vienna, Austria. Association for Computational L… find
  13. Xinyuan Wang, Chenxi Li, Zhen Wang, Fan Bai, Haotian Luo, Jiayou Zhang, Nebojsa Jojic, Eric P Xing, and Zhiting Hu. 2023. Promptagent: Strategic planning with language models enables expert-level prompt optimization. arXiv preprint arXiv:2310.16427.
  14. A.1. Subtask 1: Question Interpretation Prompt Template: Question Interpretation find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/majeedi-2026-neural-archehr-qa-2026-one/