Elamur Research⌕
☰
home / papers / 2026 / Travel on the ICD Tree: Benchmarking Agentic Reasoning for ICD Coding from Chinese Electronic Medical Records

Travel on the ICD Tree: Benchmarking Agentic Reasoning for ICD Coding from Chinese Electronic Medical Records

Paper · 2026 · Xinjie Xu, Yongqi Fan, Shuang-shuang Chen et al.

Summary

Real-world ICD coding often requires more than matching a diagnosis string to a code. Coders must pull details such as subtype, body site, pathogen and complications out of full clinical records, then pick a leaf-level code under formal coding rules. Existing benchmarks such as MIMIC are in English and coarse-grained, so this setting has gone largely unevaluated. The authors built Code4Detail, a benchmark of real, de-identified Chinese EMRs annotated with fine-grained ICD-10 codes by medical records professionals. They also propose ToT-ICD, an agentic framework that frames coding as a walk down the ICD hierarchy. It combines similarity retrieval, path-guided navigation, dynamic backtracking and injected coding rules. Several open and closed LLMs were tested under retrieval-and-rerank (R&R) and agent workflows. R&R is cheap and stable on explicit cases, but its gains level off as models get larger. Agentic tree search reaches the best accuracy when paired with strong reasoning models, though at much higher token cost. The data and code are publicly released.

Key points

  • Code4Detail: 560 real Chinese clinical records spanning 434 frequently seen diseases drawn from 19 core ICD-10 chapters, with leaf-node annotation. Released at github.com/JOHNNY-fans/Travel-on-the-ICD-Tree.
  • Headline accuracy: ToT-ICD peaks at 43.75% with DeepSeek-V3.2, while R&R reaches 33.75% on Qwen3-8B.
  • R&R failure analysis: across 336 failures, 56.0% occurred because the correct code never appeared in the initial retrieval set. The best Top-k was 40 (40.00%); raising it to 50 dropped accuracy to 37.39%.
  • Agent error reductions: compared with the Base Agent, the Retrieval Agent lowers navigation errors from 9.8% to 8.6% and validation-conflict errors from 10.9% to 8.6%. Core disease selection remains the main failure mode.
  • Cost per case: R&R takes 2 steps and about 32K tokens, the Base Agent 14 steps and 323K tokens, and ToT-ICD 18 steps and 408K tokens.

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

🎓Fudan UniversityShanghai · China202 works · 22,509 citations🎓East China University of Science and TechnologyShanghai · China14 works · 87 citations

Also on the byline: Zhongshan Hospital.

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksBiochemistry, Genetics and Molecular Biology · 263 worksMolecular Biology · 260 worksHealth Professions · 75 worksHealth Information Management · 56 works

🧭 Topics 1Research subjects and languages the work is about.Reasoning · 1,685 works

Cite

Cite as Xinjie Xu, Yongqi Fan, Shuang-shuang Chen, Qi Ye, Weibin Guo and Xinxuan Hu. 2026. Travel on the ICD Tree: Benchmarking Agentic Reasoning for ICD Coding from Chinese Electronic Medical Records. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.191

@inproceedings{xu2026travel,
  title = {{Travel on the ICD Tree: Benchmarking Agentic Reasoning for ICD Coding from Chinese Electronic Medical Records}},
  author = {Xinjie Xu, Yongqi Fan, Shuang-shuang Chen, Qi Ye, Weibin Guo and Xinxuan Hu},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.191},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.191}
}

References

32 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Krishanu Das Baksi, Elijah Soba, John J Higgins, Ravi Saini, Jaden Wood, Jane Cook, Jack I Scott, Nirmala Pudota, Tim Weninger, Edward Bowen, and 1 others. 2025. Medcoder: A generative ai assistant for medical coding. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of th… find
  2. Leah S Larkey and W Bruce Croft. 1996. Combining classifiers in text categorization. In Proceedings of the 19th annual international ACM SIGIR conference on Research and development in information retrieval, pages 289–297. find
  3. Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, and 1 others. 2024a. Deepseek-v3 technical report. arXiv preprint arXiv:2412.19437. find
  4. YunZhi Chen, HuiJuan Lu, and LanJuan Li. 2017. Automatic icd-10 coding algorithm using an improved longest common subsequence based on semantic similarity. PloS one, 12(3):e0173410. find
  5. Jiaxiang Liu, Yuan Wang, Jiawei Du, Joey Tianyi Zhou, and Zuozhu Liu. 2024b. Medcot: Medical chain of thought via hierarchical expert. arXiv preprint arXiv:2412.13736. find
  6. Zhenghan Chen, Changzeng Fu, Ruoxue Wu, Ye Wang, Xunzhu Tang, and Xiaoxuan Liang. 2023. Lgfat-rgcn: Faster attention with heterogeneous rgcn for medical icd coding generation. In Proceedings of the 31st ACM International Conference on Multimedia, pages 5428–5435. find
  7. Leibo Liu, Oscar Perez-Concha, Anthony Nguyen, Vicki Bennett, and Louisa Jorm. 2023. Automated icd coding using extreme multi-label long text transformer-based models. Artificial Intelligence in Medicine, 144:102662. find
  8. Yongqi Fan, Hongli Sun, Kui Xue, Xiaofan Zhang, Shaoting Zhang, and Tong Ruan. 2025a. Medodyssey: A medical domain benchmark for long context evaluation up to 200k tokens. In Findings of the Association for Computational Linguistics: NAACL 2025, pages 32–56. find
  9. Yongqi Fan, Nan Wang, Kui Xue, Jingping Liu, and Tong Ruan. 2025b. Medeureka: A medical domain benchmark for multi-granularity and multi-data-type embedding-based retrieval. In Findings of the Association for Computational Linguistics: NAACL 2025, pages 2825–2851. find
  10. Yongqi Fan, Yansha Zhu, Kui Xue, Jingping Liu, and Tong Ruan. 2024. Rrnorm: A novel framework for chinese disease diagnoses normalization via llm-driven terminology component recognition and reconstruction. In Findings of the Association for Computational Linguistics: ACL 2024, pages 9162–9175. find
  11. Ruihui Hou, Shencheng Chen, Yongqi Fan, Guangya Yu, Lifeng Zhu, Jing Sun, Jingping Liu, and Tong Ruan. 2024. Msdiagnosis: A benchmark for evaluating large language models in multi-step clinical diagnosis. arXiv preprint arXiv:2408.10039. — [arXiv:2408.10039](https://arxiv.org/abs/2408.10039) find
  12. Chao-Wei Huang, Shang-Chi Tsai, and Yun-Nung Chen. 2022. Plm-icd: Automatic icd coding with pretrained language models. arXiv preprint arXiv:2207.05289. find
  13. Alistair EW Johnson, Lucas Bulgarelli, Lu Shen, Alvin Gayles, Ayad Shammout, Steven Horng, Tom J Pollard, Sicheng Hao, Benjamin Moody, Brian Gow, and 1 others. 2023. Mimic-iv, a freely accessible electronic health record dataset. Scientific data, 10(1):1.
  14. Alistair EW Johnson, Tom J Pollard, Lu Shen, Li-wei H Lehman, Mengling Feng, Mohammad Ghassemi, Benjamin Moody, Peter Szolovits, Leo Anthony Celi, and Roger G Mark. 2016. Mimic-iii, a freely accessible critical care database. Scientific data, 3(1):1–9.
  15. Zhanyu Liu, Shiyao Wang, Xingmei Wang, Rongzhou Zhang, Jiaxin Deng, Honghui Bao, Jinghao Zhang, Wuchao Li, Pengfei Zheng, Xiangyu Wu, and 1 others. 2025. Onerec-think: In-text reasoning for generative recommendation. arXiv preprint arXiv:2510.11639.
  16. Andreas Motzfeldt, Joakim Edin, Casper L Christensen, Christian Hardmeier, Lars Maaløe, and Anna Rogers. 2025. Code like humans: A multi-agent solution for medical coding. In Findings of the Association for Computational Linguistics: EMNLP 2025, pages 22612–22627. Association for Computational Lingu… find
  17. James Mullenbach, Sarah Wiegreffe, Jon Duke, Jimeng Sun, and Jacob Eisenstein. 2018. Explainable prediction of medical codes from clinical text. arXiv preprint arXiv:1802.05695.
  18. Adler Perotte, Rimma Pivovarov, Karthik Natarajan, Nicole Weiskopf, Frank Wood, and Noémie Elhadad. 2014. Diagnosis code assignment: models and evaluation metrics. Journal of the American Medical Informatics Association, 21(2):231–237. find
  19. Hugo Silva, Vítor Duque, Mário Macedo, and Mateus Mendes. 2024. Aiding icd-10 encoding of clinical health records using improved text cosine similarity and plm-icd. Algorithms, 17(4):144. find
  20. Yaoqian Sun, Lei Sang, Dan Wu, Shilin He, Yani Chen, Huilong Duan, Han Chen, and Xudong Lu. 2024. Enhanced icd-10 code assignment of clinical texts: A summarization-based approach. Artificial Intelligence in Medicine, 156:102967. find
  21. Gemini Team, Rohan Anil, Sebastian Borgeaud, Jean-Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew M Dai, Anja Hauth, Katie Millican, and 1 others. 2023. Gemini: a family of highly capable multimodal models. arXiv preprint arXiv:2312.11805.
  22. Tao Tu, Shekoofeh Azizi, Danny Driess, Mike Schaekermann, Mohamed Amin, Pi-Chuan Chang, Andrew Carroll, Charles Lau, Ryutaro Tanno, Ira Ktena, Basil Mustafa, Aakanksha Chowdhery, Yun Liu, Simon Kornblith, David J. Fleet, P. A. Mansfield, Sushant Prakash, Renee C Wong, Sunny Virmani, and 13 others. 2… find
  23. Xindi Wang, Robert Mercer, and Frank Rudzicz. 2024. Multi-stage retrieve and re-rank model for automatic medical coding recommendation. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Pa… find
  24. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Ed H. Chi, F. Xia, Quoc Le, and Denny Zhou. 2022. Chain of thought prompting elicits reasoning in large language models. ArXiv, abs/2201.11903.
  25. World Health Organization. 2004. International Statistical Classification of Diseases and related health problems: Alphabetical index. World Health Organization. find
  26. Yuzhou Wu, Zhigang Chen, Xin Yao, Xuechen Chen, Zeren Zhou, and Jinkai Xue. 2022. Jan: Joint attention networks for automatic icd coding. IEEE journal of biomedical and health informatics, 26(10):5235– 5246. find
  27. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, and 1 others. 2025. Qwen3 technical report. arXiv preprint arXiv:2505.09388.
  28. Qwen An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Guanting Dong, Haoran Wei, Huan Lin, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxin Yang, Jingren Zhou, Junyang Lin, and 25 others. 2024. Qwen2.5 technical report. ArXiv, a… find
  29. Zhichao Yang, Sunjae Kwon, Zonghai Yao, and Hong Yu. 2023. Multi-label few-shot icd coding as autoregressive generation with prompt. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 37, pages 5366–5374. find
  30. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2022. React: Synergizing reasoning and acting in language models. ArXiv, abs/2210.03629.
  31. Lingling Zhou, Cheng Cheng, Dong Ou, and Hao Huang. 2020. Construction of a semi-automatic icd-10 coding system. BMC medical informatics and decision making, 20(1):67. find
  32. Tong Zhou, Pengfei Cao, Yubo Chen, Kang Liu, Jun Zhao, Kun Niu, Weifeng Chong, and Shengping Liu. 2021. Automatic icd coding via interactive shared representation networks with self-distillation mechanism. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/xu-2026-travel-icd-tree-benchmarking-agentic/