Elamur Research⌕
☰
home / papers / 2026 / Compressed Representations of Patient Records: A Comparative Study of Template-Based and LLM-Based Methods for Clinical Data Summarization and Visualization

Compressed Representations of Patient Records: A Comparative Study of Template-Based and LLM-Based Methods for Clinical Data Summarization and Visualization

Paper · 2026 · Andreas Stöckl, Oliver Krauss and Sophie Bauernfeind

Summary

Electronic Health Records are often too long for clinicians to review quickly, especially when several patients must be compared side by side. This study compares ways of condensing a patient record into a short, standardized form that keeps the clinically essential facts. The authors built a synthetic cohort of 75 records using a custom generator modeled on Synthea, with three levels of patient complexity. They tested four strategies:

  • rule-based template extraction,
  • TF-IDF-style extractive key-phrase selection,
  • purely LLM-generated structured JSON summaries,
  • a hybrid that extracts structured fields algorithmically and uses an LLM only for a short comparative narrative.

The LLM and hybrid methods each ran with eight cloud and local models. Outputs were scored on information recall (diagnoses, medications, allergies, labs, vitals), compression ratio, field completeness and a composite quality score. Template extraction compressed records the most. The hybrid approach gave the best balance between brevity and clinical usefulness. The work also provides an open-source implementation with interactive HTML infographic dashboards for reviewing a whole cohort.

Key points

  • Dataset: 75 synthetic patients (29 low, 33 moderate, 13 high complexity), each with 1–7 ICD-10 diagnoses, 1–12 medications, labs, vitals and a clinical note of 1,500–2,500 characters.
  • Models for the LLM and hybrid methods (18 runs in total): GPT-4o, GPT-4o-mini, MedGemma 4B and 27B, Gemma-3-12B, Llama-3.1-8B, lfm2.5-thinking and nemotron-3-nano. A single prompt was used for all models, with no prompt engineering.
  • Diagnosis recall ranged from 0.637 to 1.000 across methods. Medication and allergy recall stayed above 0.880.
  • Template extraction reached the highest compression ratio (7.6×). The hybrid methods offered the best overall trade-off.
  • The composite quality score weights information preservation at 40%, field completeness at 25%, compression at 15% and presence of a clinical summary at 20%.

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

Also on the byline: University of Applied Sciences Upper Austria.

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksMedicine · 241 worksHealth Informatics · 184 worksHealth Professions · 69 worksHealth Information Management · 50 works

🧭 Topics 1Research subjects and languages the work is about.Summarization · 232 works

Cite

Cite as Andreas Stöckl, Oliver Krauss and Sophie Bauernfeind. 2026. Compressed Representations of Patient Records: A Comparative Study of Template-Based and LLM-Based Methods for Clinical Data Summarization and Visualization. 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 10.63317/4ph9cddo4vbn

@inproceedings{stockl2026compressed,
  title = {{Compressed Representations of Patient Records: A Comparative Study of Template-Based and LLM-Based Methods for Clinical Data Summarization and Visualization}},
  author = {Andreas Stöckl, Oliver Krauss and Sophie Bauernfeind},
  booktitle = {Proceedings of the Third Workshop on Patient-Oriented Language Processing (CL4Health) @ LREC 2026},
  year = {2026},
  doi = {10.63317/4ph9cddo4vbn},
  url = {https://doi.org/10.63317/4ph9cddo4vbn}
}

References

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

  1. Emily Alsentzer, John R Murphy, Willie Boag, Wei-Hung Weng, Di Jin, Tristan Naumann, and Matthew BA McDermott. 2019. Publicly available clinical bert embeddings. arXiv preprint arXiv:1904.03323.
  2. Daphne Barretto, Matthew Jin, and Bora Oztekin. 2026. Clinical text summarization with llm-based evaluation. Stanford CS224N Project. find
  3. Lydie Bednarczyk, Daniel Reichenpfader, et al. 2025. Scientific evidence for clinical text summarization using large language models: Scoping review. Journal of Medical Internet Research. find
  4. Ethical Considerations and Limitations The synthetic patient data was generated for research purposes only and does not represent real individuals. The study also has several limitations: Edward Choi, Zhen Xu, Yujia Li, Michael W Dusenberry, Gerardo Flores, Yuan Xue, and Andrew M Dai. 2020. Learning… find
  5. Hughes, Charles Lau, Justin Chen, Fereshteh Mahvar, Liron Yatziv, Tiffany Chen, Bram Sterling, Stefanie Anna Baby, Susanna Maria Baby, Jeremy Lai, Samuel Schmidgall, Lu Yang, Kejia Chen, Per Bjornsson, Shashir Reddy, Ryan Brush, Kenneth Philbrick, Mercy Asiedu, Ines Mezerreg, Howard Hu, Howard Yang,… find
  6. Saibo Geng, Hudson Cooper, Michal Moskal, Samuel Jenkins, Julian Berman, Nathan Ranchin, Robert West, Eric Horvitz, and Harsha Nori. 2025. JSONSchemabench: Evaluating constrained decoding with LLMs on efficiency, coverage and quality. In ES-FoMo III: 3rd Workshop on Efficient Systems for Foundation… find
  7. Jinhyuk Lee, Wonjin Yoon, Sungdong Kim, Donghyeon Kim, Sunkyu Kim, Chan Ho So, and Jaewoo Kang. 2020. Biobert: a pre-trained biomedical language representation model for biomedical text mining. Bioinformatics, 36(4):1234–1240.
  8. Diya Li, Yue Zhao, Zhifang Wang, Calvin Jung, and Zhe Zhang. 2024. Large language model-driven structured output: A comprehensive benchmark and spatial data generation framework. IS-PRS International Journal of Geo-Information, 13(11):405. find
  9. Yaxi Lu, Haolun Li, Xin Cong, Zhong Zhang, Yesai Wu, Yankai Lin, Zhiyuan Liu, Fangming Liu, and Maosong Sun. 2025. Learning to generate structured output with schema reinforcement learning. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025), pages 4… find
  10. OpenAI. 2023. Gpt-4 technical report. preprint arXiv:2303.08774. arXiv Rimma Pivovarov and Noemie Elhadad. 2015. Automated methods for the summarization of electronic health records. Journal of the American Medical Informatics Association, 22(5):938–947.
  11. Frank Schilder, Daniel Reichenpfader, et al. 2025. Scientific evidence for clinical text summarization using large language models: Scoping review. Journal of Medical Internet Research, 27(1):e68998. find
  12. Rosanne C Schoonbeek, Jessica D Workum, Stephanie C E Schuit, Anne H Hoekman, Taran-nom Mehri, Job N Doornberg, Tom P van der Laan, and Charlotte M H H T Bootsma-Robroeks. 2025. Quality and efficiency of integrating customised large language model-generated summaries versus physician-written summari… find
  13. Andrew Sellergren, Sahar Kazemzadeh, Tiam Jaroensri, Atilla Kiraly, Madeleine Traverse, Timo Kohlberger, Shawn Xu, Fayaz Jamil, Cían Benjamin Shickel, Patrick J Tighe, Azra Bihorac, and Parisa Rashidi. 2018. Deep ehr: A survey of recent advances in deep learning techniques for electronic health reco… find
  14. Yuqi Si, Jingcheng Du, Zhao Li, Xiaoqian Jiang, Timothy Miller, Fei Wang, W Jim Jiménez, and Lucila Ohno-Machado. 2021. Deep representation learning of patient data from electronic health records (ehr): A systematic review. Journal of Biomedical Informatics, 115:103671. find
  15. Karan Singhal, Shekoofeh Azizi, Tao Tu, Sara Mahdavi, Jason Wei, Hyung Won Chung, Nathan Scales, Ajay Tanwani, Heather Cole-Lewis, Stephen Pfohl, et al. 2023. Large language models encode clinical knowledge. Nature, 620(7972):172–180.
  16. Liyan Tang, Zhaoyi Sun, Betina Idnay, Jordan G Nestor, Ali Soroush, Noémie Elhadad, Chunhua Weng, and Yifan Peng. 2023. Evaluating large language models on medical evidence summarization. npj Digital Medicine, 6(1):158. find
  17. Dave Van Veen, Cara Van Uden, Louis Berry, and Yiliang and";"; and"; others Chen. 2024. Adapted large language models can outperform medical experts in clinical text summarization. Nature Medicine, 30:1134–1142.
  18. Jason Walonoski, Mark Kramer, Joseph Nichols, Andre Quina, Chris Moesel, Dylan Hall, Carlton Duffett, Kudakwashe Dube, Thomas Gallagher, and Scott McLachlan. 2018. Synthea: An approach, method, and software mechanism for generating synthetic patients and the synthetic electronic health care record.… find
  19. Michael Wornow, Yizhe Xu, Rahul Thapa, Birju Patel, Ethan Steinberg, Scott Fleming, Michael A Pfeffer, Jason Fries, and Nigam H Shah. 2023. The shaky foundations of large language models and foundation models for electronic health records. npj Digital Medicine, 6(1):135. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/stockl-2026-compressed-representations-patient-records-comparative/