---
title: "HiTZ-IXA at ArchEHR-QA 2026: Evidence Alignment Through"
description: "The paper describes the HiTZ-IXA team's entry in the ArchEHR-QA 2026 shared task. 22 references."
authors: ["Xabier Irastortza-Urbieta", "Maite Oronoz", "Alicia Pérez"]
year: 2026
venue: "Proceedings of the Third Workshop on Patient-Oriented Language Processing (CL4Health) @ LREC 2026"
source: "https://aclanthology.org/2026.cl4health-1.40/"
doi: "10.63317/3zuhjyhh6vy9"
licence: "CC BY-NC 4.0 (ELRA, LREC proceedings)"
licence_url: "https://creativecommons.org/licenses/by-nc/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/urbieta-2026-hitz-ixa-archehr-qa-2026/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/urbieta-2026-hitz-ixa-archehr-qa-2026/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/32/) / HiTZ-IXA at ArchEHR-QA 2026: Evidence Alignment Through Self-Consistency and Prompt Curation in Memory-Constrained Environments

# HiTZ-IXA at ArchEHR-QA 2026: Evidence Alignment Through Self-Consistency and Prompt Curation in Memory-Constrained Environments

**Authors:** [Xabier Irastortza-Urbieta](https://md.research.elamur.ai/free/en/authors/xabier-irastortza-urbieta/), Maite Oronoz and [Alicia Pérez](https://md.research.elamur.ai/free/en/authors/alicia-perez/) · **Year:** 2026 · **Venue:** [Proceedings of the Third Workshop on Patient-Oriented Language Processing (CL4Health) @ LREC 2026](https://md.research.elamur.ai/free/en/venues/cl4health/) ([Workshop on Patient-Oriented Language Processing @ LREC 2026](https://md.research.elamur.ai/free/en/venues/cl4health/2026/)) · **Publisher:** [European Language Resources Association (ELRA)](https://md.research.elamur.ai/free/en/publishers/european-language-resources-association-elra/) · **Licence:** [CC BY-NC 4.0 (ELRA, LREC proceedings)](https://creativecommons.org/licenses/by-nc/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.cl4health-1.40/) · [PDF](https://aclanthology.org/2026.cl4health-1.40.pdf) · [DOI 10.63317/3zuhjyhh6vy9](https://doi.org/10.63317/3zuhjyhh6vy9) · [OpenAlex](https://openalex.org/W7166532533)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
The paper describes the HiTZ-IXA team's entry in the ArchEHR-QA 2026 shared task. In this task, systems must ground answers to patients' health questions in specific sentences of their Electronic Health Records (EHRs). The team gave limited attention to Subtask 2 (Evidence Identification) and focused on Subtask 4 (Evidence Alignment), which links each answer sentence to the EHR sentences that support it.

A deliberate design constraint was to use only open-source models of at most 8B parameters, run on local GPUs. The models were Aloe, Qwen3, Llama Instruct (8B each) and MedGemma (4B). Using the organizers' 167 clinical cases, the authors compared several approaches:
- cosine-similarity baselines with BERT and MedBERT embeddings,
- zero-shot LLM agents with several curated prompt variants,
- output post-processing, self-consistency, and combinations of LLMs.

They conclude that careful prompt design combined with robust post-processing is what makes small-model pipelines stable. Self-consistency added substantial accuracy gains on top of that. This matters for settings without access to large models or paid APIs, because it shows that inexpensive techniques can make compact LLMs effective at evidence grounding.

## Key points
- **Data:** 167 cases with patient narratives, clinician-interpreted questions, EHRs and sentence-level grounding.
  - Cases 1–20 serve as development data.
  - Cases 121–167 are the Subtask 2 test set; cases 21–167 are the Subtask 4 test set.
  - Ranking uses micro F1.
- **Embedding baseline for Subtask 4** (dev set, cosine threshold T = 0.60):
  - Random baseline: F1 11.52.
  - BERT: F1 15.77.
  - MedBERT: F1 40.83, showing the strong effect of training on clinical-domain text.
- **Subtask 2 with Aloe:**
  - Splitting EHRs into windows of 15 sentences gave a dev F1 of 49.40.
  - A three-agent "essential vs. non-relevant" scheme (Ess-Nrev) raised dev F1 from 48.16 to 53.25.
  - On the test set the gain was marginal: 45.29 → 46.01.
- **Prompt variants tested for Subtask 4:** base template (First), no question (NoQu), Chain of Thought (CoT), inverted order (Invr), and no instructions block (NoIns).
  - The final Subtask 4 scores fall in a section omitted from the supplied text, so they are not reported here.
- **Code and prompts:** released at github.com/XabierIU/HiTZ-IXA-ArchEHR-QA.

*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 the Basque Country](https://md.research.elamur.ai/free/en/universities/university-of-the-basque-country/) — University, Leioa, Spain · 57 works · 16,520 citations

📚 Shelves (6): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works · [🩺 Medicine](https://md.research.elamur.ai/free/en/shelves/medicine/) · 241 works · [🩺 Health Informatics](https://md.research.elamur.ai/free/en/shelves/medicine/health-informatics/) · 184 works

🧭 Topics (1): [❓ Question answering](https://md.research.elamur.ai/free/en/topics/question-answering/) · 441 works

## Cite

Cite as: Xabier Irastortza-Urbieta, Maite Oronoz and Alicia Pérez. 2026. HiTZ-IXA at ArchEHR-QA 2026: Evidence Alignment Through Self-Consistency and Prompt Curation in Memory-Constrained Environments. 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.

```bibtex
@inproceedings{irastortzaurbieta2026hitz,
  title = {{HiTZ-IXA at ArchEHR-QA 2026: Evidence Alignment Through Self-Consistency and Prompt Curation in Memory-Constrained Environments}},
  author = {Xabier Irastortza-Urbieta, Maite Oronoz and Alicia Pérez},
  booktitle = {Proceedings of the Third Workshop on Patient-Oriented Language Processing (CL4Health) @ LREC 2026},
  year = {2026},
  doi = {10.63317/3zuhjyhh6vy9},
  url = {https://doi.org/10.63317/3zuhjyhh6vy9}
}
```

- [← previous HistoryBankQA: Multilingual Temporal Question Answering on Historical Events](https://md.research.elamur.ai/free/en/papers/mandal-2026-historybankqa-multilingual-temporal-question-answering/)
- [next → HOPE: Hybrid Optimized Parallel Encoding with Supervised and Unsupervised Semantic Fusion for Depression Symptom Detection](https://md.research.elamur.ai/free/en/papers/mai-2026-hope-hybrid-optimized-parallel-encoding/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/32/)

## References

22 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Abdul Mohaimen Al Radi, Xu Cao, Fanyang Yu, Yuyuan Liu, Fengbei Liu, Chong Wang, Yuanhong Chen, Jintai Chen, Hu Wang, Yanda Meng, et al. 2025. Agentic large-language-model systems in medicine: A systematic review and taxonomy. Authorea Preprints. [find](https://scholar.google.com/scholar?q=Abdul+Mohaimen+Al+Radi%2C+Xu+Cao%2C+Fanyang+Yu%2C+Yuyuan+Liu%2C+Fengbei+Liu%2C+Chong+Wang%2C+Yuanhong+Chen%2C+Jintai+Chen%2C+Hu+Wang%2C+Yanda+Meng%2C+et+al%2E+2025%2E+Agentic+large%2Dlanguage%2Dmodel+systems+in+medicine%3A+A+syste)
2. Workshop on Biomedical Language Processing (Shared Tasks), pages 1–10, Vienna, Austria. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Workshop+on+Biomedical+Language+Processing+%28Shared+Tasks%29%2C+pages+1%E2%80%9310%2C+Vienna%2C+Austria%2E+Association+for+Computational+Linguistics%2E)
3. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. [BERT: Pre-training of deep bidirectional transformers for language understanding](https://md.research.elamur.ai/free/en/papers/devlin-2019-bert-pre-training-deep-bidirectional/). In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies,…
4. Dario Garcia-Gasulla, Jordi Bayarri-Planas, Ashwin Kumar Gururajan, Enrique Lopez-Cuena, Adrian Tormos, Daniel Hinjos, Pablo Bernabeu-Perez, Anna Arias-Duart, Pablo Agustin Martin-Torres, Marta Gonzalez-Mallo, et al. 2025. The aloe family recipe for open and specialized healthcare llms. [find](https://scholar.google.com/scholar?q=Dario+Garcia%2DGasulla%2C+Jordi+Bayarri%2DPlanas%2C+Ashwin+Kumar+Gururajan%2C+Enrique+Lopez%2DCuena%2C+Adrian+Tormos%2C+Daniel+Hinjos%2C+Pablo+Bernabeu%2DPerez%2C+Anna+Arias%2DDuart%2C+Pablo+Agustin+Martin%2DTorres%2C+Marta+Gonzal)
5. Hyeon Hwang, Hyeongsoon Hwang, Jongmyung Jung, Jaehoon Yun, Minju Song, Yein Park, Dain Kim, Taewhoo Lee, Jiwoong Sohn, Chanwoong Yoon, Sihyeon Park, Jiwoo Lee, Heechul Yang, and Jaewoo Kang. 2025. DMIS lab at ArchEHR-QA 2025: Evidence-grounded answer generation for EHR-based QA via a multi-agent fr… [find](https://scholar.google.com/scholar?q=Hyeon+Hwang%2C+Hyeongsoon+Hwang%2C+Jongmyung+Jung%2C+Jaehoon+Yun%2C+Minju+Song%2C+Yein+Park%2C+Dain+Kim%2C+Taewhoo+Lee%2C+Jiwoong+Sohn%2C+Chanwoong+Yoon%2C+Sihyeon+Park%2C+Jiwoo+Lee%2C+Heechul+Yang%2C+and+Jaewoo+Kang%2E+2025%2E+DM)
6. Jayetri Bardhan, Kirk Roberts, and Daisy Zhe Wang. 2024. Question answering for electronic health records: Scoping review of datasets and models. J Med Internet Res, 26:e53636. [find](https://scholar.google.com/scholar?q=Jayetri+Bardhan%2C+Kirk+Roberts%2C+and+Daisy+Zhe+Wang%2E+2024%2E+Question+answering+for+electronic+health+records%3A+Scoping+review+of+datasets+and+models%2E+J+Med+Internet+Res%2C+26%3Ae53636%2E)
7. Yubin Kim, Chanwoo Park, Hyewon Jeong, Yik Siu Chan, Xuhai Xu, Daniel McDuff, Hyeonhoon Lee, Marzyeh Ghassemi, Cynthia Breazeal, and Hae Won Park. 2024. Mdagents: An adaptive collaboration of llms for medical decision-making. In Advances in Neural Information Processing Systems, volume 37, pages 794… [find](https://scholar.google.com/scholar?q=Yubin+Kim%2C+Chanwoo+Park%2C+Hyewon+Jeong%2C+Yik+Siu+Chan%2C+Xuhai+Xu%2C+Daniel+McDuff%2C+Hyeonhoon+Lee%2C+Marzyeh+Ghassemi%2C+Cynthia+Breazeal%2C+and+Hae+Won+Park%2E+2024%2E+Mdagents%3A+An+adaptive+collaboration+of+llms+for)
8. Sai Prasanna Teja Reddy Bogireddy, Abrar Ma-jeedi, 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 Processi… [find](https://scholar.google.com/scholar?q=Sai+Prasanna+Teja+Reddy+Bogireddy%2C+Abrar+Ma%2Djeedi%2C+Viswanath+Gajjala%2C+Zhuoyan+Xu%2C+Siddhant+Rai%2C+and+Vaishnav+Potlapalli%2E+2025%2E+Neural+at+ArchEHR%2DQA+2025%3A+Agentic+prompt+optimization+for+evidence%2Dgroun)
9. Tuan Dung Le, Thanh Duong, Shohreh Haddadan, Behzad Jazayeri, Brandon Manley, and Thanh Thieu. 2025. LAILab at ArchEHR-QA 2025: Test-time scaling for evidence selection in grounded question answering from electronic health records. In Proceedings of the 24th Workshop on Biomedical Language Processin… [find](https://scholar.google.com/scholar?q=Tuan+Dung+Le%2C+Thanh+Duong%2C+Shohreh+Haddadan%2C+Behzad+Jazayeri%2C+Brandon+Manley%2C+and+Thanh+Thieu%2E+2025%2E+LAILab+at+ArchEHR%2DQA+2025%3A+Test%2Dtime+scaling+for+evidence+selection+in+grounded+question+answering+)
10. Adrian Cuadron Cortes, Aimar Sagasti, Mai-tane Urruela, Iker De La Iglesia, Ane García Domingo-aldama, Aitziber Atutxa Salazar, Josu Goikoetxea, and Ander Barrena. 2025. ArgHiTZ at ArchEHR-QA 2025: A two-step divide and conquer approach to patient question answering for top factuality. In Proceeding… [find](https://scholar.google.com/scholar?q=Adrian+Cuadron+Cortes%2C+Aimar+Sagasti%2C+Mai%2Dtane+Urruela%2C+Iker+De+La+Iglesia%2C+Ane+Garc%C3%ADa+Domingo%2Daldama%2C+Aitziber+Atutxa+Salazar%2C+Josu+Goikoetxea%2C+and+Ander+Barrena%2E+2025%2E+ArgHiTZ+at+ArchEHR%2DQA+2025%3A+A+)
11. Timo Schick, Jane Dwivedi-Yu, Roberto Dessi, Roberta Raileanu, Maria Lomeli, Eric Hambro, Luke Zettlemoyer, Nicola Cancedda, and Thomas Scialom. 2023. [Toolformer: Language models can teach themselves to use tools](https://md.research.elamur.ai/free/en/papers/schick-2023-toolformer-language-models-can-teach/). In Advances in Neural Information Processing Systems, volume 36, pages 68539–68551. Cu…
12. Andrew Sellergren, Sahar Kazemzadeh, Tiam Jaroensri, Atilla Kiraly, Madeleine Traverse, Timo Kohlberger, Shawn Xu, Fayaz Jamil, Cían Hughes, Charles Lau, et al. 2025. Medgemma technical report. arXiv preprint arXiv:2507.05201. [find](https://scholar.google.com/scholar?q=Andrew+Sellergren%2C+Sahar+Kazemzadeh%2C+Tiam+Jaroensri%2C+Atilla+Kiraly%2C+Madeleine+Traverse%2C+Timo+Kohlberger%2C+Shawn+Xu%2C+Fayaz+Jamil%2C+C%C3%ADan+Hughes%2C+Charles+Lau%2C+et+al%2E+2025%2E+Medgemma+technical+report%2E+arXiv+)
13. Karan Singhal, Tao Tu, Juraj Gottweis, Rory Sayres, Ellery Wulczyn, Mohamed Amin, Le Hou, Kevin Clark, Stephen R. Pfohl, Heather Cole-Lewis, Darlene Neal, Qazi Mamunur Rashid, Mike Schaekermann, Amy Wang, Dev Dash, Jonathan H. Chen, Nigam H. Shah, Sami Lachgar, Philip Andrew Mansfield, Sushant Praka… [find](https://scholar.google.com/scholar?q=Karan+Singhal%2C+Tao+Tu%2C+Juraj+Gottweis%2C+Rory+Sayres%2C+Ellery+Wulczyn%2C+Mohamed+Amin%2C+Le+Hou%2C+Kevin+Clark%2C+Stephen+R%2E+Pfohl%2C+Heather+Cole%2DLewis%2C+Darlene+Neal%2C+Qazi+Mamunur+Rashid%2C+Mike+Schaekermann%2C+Amy+W)
14. Sarvesh Soni and Dina Demner-Fushman. 2026a. A dataset for addressing patient’s information needs related to clinical course of hospitalization. Scientific Data. [find](https://scholar.google.com/scholar?q=Sarvesh+Soni+and+Dina+Demner%2DFushman%2E+2026a%2E+A+dataset+for+addressing+patient%E2%80%99s+information+needs+related+to+clinical+course+of+hospitalization%2E+Scientific+Data%2E)
15. 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](https://scholar.google.com/scholar?q=Sarvesh+Soni+and+Dina+Demner%2DFushman%2E+2026b%2E+Overview+of+the+ArchEHR%2DQA+2026+Shared+Task+on+Grounded+Question+Answering+from+Electronic+Health+Records%2E+In+Proceedings+of+the+Third+Workshop+on+Patient%2D)
16. Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. 2022. [Self-consistency improves chain of thought reasoning in language models](https://md.research.elamur.ai/free/en/papers/wang-2022-self-consistency-improves-chain-thought/). arXiv preprint arXiv:2203.11171.
17. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed H. Chi, Quoc V. Le, and Denny Zhou. 2022. [Chain-of-thought prompting elicits reasoning in large language models](https://md.research.elamur.ai/free/en/papers/wei-2022-chain-thought-prompting-elicits-reasoning/). In Proceedings of the 36th International Conference on Neural Information Processing Systems, NIPS ’22, R…
18. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, and Chenxu Lv et al. 2025a. [Qwen3 technical report](https://md.research.elamur.ai/free/en/papers/yang-2025-qwen3-technical-report/). arXiv preprint arXiv:2505.09388.
19. Yifan Yang, Qiao Jin, Qingqing Zhu, Zhizheng Wang, Francisco Erramuspe Álvarez, Nicholas Wan, Benjamin Hou, and Zhiyong Lu. 2025b. Beyond multiple-choice accuracy: Real-world challenges of implementing large language models in healthcare. Annual Review of Biomedical Data Science, 8(Volume 8, 2025):3… [find](https://scholar.google.com/scholar?q=Yifan+Yang%2C+Qiao+Jin%2C+Qingqing+Zhu%2C+Zhizheng+Wang%2C+Francisco+Erramuspe+%C3%81lvarez%2C+Nicholas+Wan%2C+Benjamin+Hou%2C+and+Zhiyong+Lu%2E+2025b%2E+Beyond+multiple%2Dchoice+accuracy%3A+Real%2Dworld+challenges+of+implementin)
20. Kai Zhang, Yangyang Kang, Fubang Zhao, and Xiaozhong Liu. 2024. LLM-based medical assistant personalization with short- and long-term memory coordination. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies… [find](https://scholar.google.com/scholar?q=Kai+Zhang%2C+Yangyang+Kang%2C+Fubang+Zhao%2C+and+Xiaozhong+Liu%2E+2024%2E+LLM%2Dbased+medical+assistant+personalization+with+short%2D+and+long%2Dterm+memory+coordination%2E+In+Proceedings+of+the+2024+Conference+of+the+)
21. 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, Viena, Austria. Association for Computational Li… [find](https://scholar.google.com/scholar?q=Sarvesh+Soni%2C+Soumya+Gayen%2C+and+Dina+Demner%2DFushman%2E+2025%2E+Overview+of+the+ArchEHR%2DQA+2025+shared+task+on+grounded+question+answering+from+electronic+health+records%2E+In+Proceedings+of+the+24th+Worksho)
22. Charangan Vasantharajan, Kyaw Zin Tun, Ho Thi-Nga, Sparsh Jain, Tong Rong, and Chng Eng Siong. 2022. Medbert: A pre-trained language model for biomedical named entity recognition. In 2022 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC), pages 1482… [find](https://scholar.google.com/scholar?q=Charangan+Vasantharajan%2C+Kyaw+Zin+Tun%2C+Ho+Thi%2DNga%2C+Sparsh+Jain%2C+Tong+Rong%2C+and+Chng+Eng+Siong%2E+2022%2E+Medbert%3A+A+pre%2Dtrained+language+model+for+biomedical+named+entity+recognition%2E+In+2022+Asia%2DPacific)

*Updated 1 November 2026*