---
title: "Unlocking Public Catalogues — paper by Lenz et al. (2025)"
description: "Lenz et al. test whether instruction-tuning on freely available German catalogues closes this gap. 31 references."
authors: ["Stefan Lenz", "Lakisha Ortiz Rosario", "Georg Vollmar", "Arsenij Ustjanzew", "Fatma Alickovic", "Thomas Kindler", "Torsten Panholzer"]
year: 2025
venue: "arXiv:2510.13624"
source: "https://arxiv.org/abs/2510.13624"
doi: "10.48550/arxiv.2510.13624"
arxiv: "2510.13624"
licence: "arXiv — non-exclusive license to distribute"
licence_url: "http://arxiv.org/licenses/nonexclusive-distrib/1.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/lenz-2025-unlocking-public-catalogues-instruction-tuning/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/lenz-2025-unlocking-public-catalogues-instruction-tuning/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2025](https://md.research.elamur.ai/free/en/papers/2025/page/16/) / Unlocking Public Catalogues: Instruction-Tuning LLMs for ICD Coding of German Tumor Diagnoses

# Unlocking Public Catalogues: Instruction-Tuning LLMs for ICD Coding of German Tumor Diagnoses

**Authors:** Stefan Lenz, Lakisha Ortiz Rosario, Georg Vollmar, Arsenij Ustjanzew, Fatma Alickovic, Thomas Kindler and Torsten Panholzer · **Year:** 2025 · **Venue:** arXiv:2510.13624 · **Licence:** [arXiv — non-exclusive license to distribute](http://arxiv.org/licenses/nonexclusive-distrib/1.0/)

**Links:** [Publisher's page](https://arxiv.org/abs/2510.13624) · [PDF](https://arxiv.org/pdf/2510.13624) · [DOI 10.48550/arxiv.2510.13624](https://doi.org/10.48550/arxiv.2510.13624) · [arXiv 2510.13624](https://arxiv.org/abs/2510.13624) · [OpenAlex](https://openalex.org/W4415276861)

**Citations:** 1 citations (OpenAlex, 2026-10-14)

## Summary
German cancer documentation must code each tumor diagnosis in both ICD-10-GM and ICD-O-3. Small open-weight LLMs suit this work because they can run locally and protect patient privacy, but they code German diagnoses poorly. Lenz et al. test whether instruction-tuning on freely available German catalogues closes this gap. They converted Alpha-ID (the ICD-10-GM index), ICD-O-3 and the OPS procedure classification into more than half a million short German question-answer pairs. These cover code lookup and yes/no recognition of tumor diagnoses, with non-tumor diagnoses and procedures used as negatives. Eight Qwen, Llama and Mistral models (7B to 70B) were fine-tuned. They were then tested on real, deduplicated diagnosis strings from the Mainz tumor documentation system. Two reviewers checked a sample to estimate how often the short texts actually contain enough information to derive the gold code. Fine-tuning raised exact ICD-10 accuracy from near zero to roughly half of cases, close to that estimated ceiling. It also eliminated malformed outputs and narrowed the gap between small and large models. The paper shows that public catalogues are an underused source of training data.

## Key points
- **Training data (ICDOPS-QA-2024).** The dataset has 518,116 QA pairs built from:
  - Alpha-ID 2024: 13,600 tumor descriptions mapped to 809 codes.
  - ICD-O-3: 400 topography codes and 1,145 morphology codes.
  - OPS 2024: procedures used as negatives.
  - Each question type is phrased in 4–10 equivalent wordings. The dataset and the best checkpoints are released on Hugging Face.
- **Test set.** It contains 2,024 unique diagnosis texts from 2023–2024. A 100-case sample was double-annotated and put the achievable ICD-10 accuracy at 60–79% for exact codes and 81–94% at the three-character level (95% CI).
- **ICD-10 results.**
  - Exact accuracy rose from 1.4–24% to 41–58%.
  - Partial accuracy rose from 31–74% to 73–83%.
  - Malformed codes fell to 0%.
  - Tumor-diagnosis recognition reached 99%.
- **ICD-O topography results.** Topography stayed harder, at 22–40% exact and 56–67% partial after tuning.
- **Model size and reasoning.**
  - Larger models scored better both before and after tuning, but the gap shrank after tuning.
  - Qwen3's reasoning mode generally scored lower than fine-tuning and was more than 100× slower.

## Interesting ideas
- Turning a reference catalogue into many paraphrased question templates gives cheap, broad supervision without any annotated corpus, and the paraphrases teach robustness to prompt wording.
- Pulling negatives from a neighbouring but conceptually different catalogue (procedures vs. diagnoses) teaches the model where a concept's boundary lies, not just what lies inside it.
- Before judging a model, the authors had two annotators estimate how often the input actually determines the gold label, so scores can be read against a realistic ceiling.
- Training on deliberately short, code-only answers drove malformed outputs to zero and made outputs fast to generate and trivial to parse.
- The authors suggest a cost-aware hybrid, with a lightweight model for easy cases and an LLM only for hard ones, especially since slow reasoning modes did not pay off here.

## For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

- **I1** (learn): VerbNet classes, WordNet senses and FrameNet frames could be rendered as varied QA pairs to train or probe the LLM teacher tiers on Mova's own label inventory.
- **I2** (world): ontology typing could be trained with near-miss negatives, e.g. events vs. states or artefacts vs. organisms, so entity classification learns real distinctions.
- **I3** (answer): for fable QA and school tests, mark whether each answer is derivable from the text itself, so world-knowledge gaps are separated from reader errors.
- **I4** (learn): LLM teacher prompts could require terse, schema-constrained labels (a deprel, a frame name) so silver data needs no fragile output parsing.
- **I5** (tree): let the perceptron parser handle sentences where witness parsers agree, and escalate only disputed ones to gparse or an LLM teacher.

*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

- [Johannes Gutenberg University Mainz](https://md.research.elamur.ai/free/en/universities/johannes-gutenberg-university-mainz/) — University, Mainz, Germany · 6 works · 665 citations

Also on the byline: [University Medical Center](https://ror.org/036pt7h44) · [University Medical Center of the Johannes Gutenberg University Mainz](https://ror.org/00q1fsf04).

📚 Shelves (6): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🧬 Biochemistry, Genetics and Molecular Biology](https://md.research.elamur.ai/free/en/shelves/biochemistry-genetics-and-molecular-biology/) · 263 works · [🧬 Molecular Biology](https://md.research.elamur.ai/free/en/shelves/biochemistry-genetics-and-molecular-biology/molecular-biology/) · 260 works · [🏥 Health Professions](https://md.research.elamur.ai/free/en/shelves/health-professions/) · 75 works · [🏥 Health Information Management](https://md.research.elamur.ai/free/en/shelves/health-professions/health-information-management/) · 56 works

🧭 Topics (1): [🧠 Language models & pretraining](https://md.research.elamur.ai/free/en/topics/language-models/) · 3,894 works

## Cite

Cite as: Stefan Lenz, Lakisha Ortiz Rosario, Georg Vollmar, Arsenij Ustjanzew, Fatma Alickovic, Thomas Kindler and Torsten Panholzer. 2025. Unlocking Public Catalogues: Instruction-Tuning LLMs for ICD Coding of German Tumor Diagnoses. arXiv:2510.13624.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@misc{lenz2025unlocking,
  title = {{Unlocking Public Catalogues: Instruction-Tuning LLMs for ICD Coding of German Tumor Diagnoses}},
  author = {Stefan Lenz, Lakisha Ortiz Rosario, Georg Vollmar, Arsenij Ustjanzew, Fatma Alickovic, Thomas Kindler and Torsten Panholzer},
  year = {2025},
  doi = {10.48550/arxiv.2510.13624},
  eprint = {2510.13624},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2510.13624}
}
```

- [← previous Universal Dependencies for Amahuaca](https://md.research.elamur.ai/free/en/papers/angulo-2025-universal-dependencies-amahuaca/)
- [next → Unsupervised Classification of English Words Based on Phonological Information: Discovery of Germanic and Latinate Clusters](https://md.research.elamur.ai/free/en/papers/morita-2025-unsupervised-classification-english-words-based/)
- [↑ back to papers of 2025](https://md.research.elamur.ai/free/en/papers/2025/page/16/)

## References

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

1. Eguia H, Sánchez-Bocanegra CL, Vinciarelli F, Alvarez-Lopez F, Saigí-Rubió F. Clinical Decision Support and Natural Language Processing in Medicine: Systematic Literature Review. J Med Internet Res. 2024 Sept 30;26:e55315. [find](https://scholar.google.com/scholar?q=Eguia+H%2C+S%C3%A1nchez%2DBocanegra+CL%2C+Vinciarelli+F%2C+Alvarez%2DLopez+F%2C+Saig%C3%AD%2DRubi%C3%B3+F%2E+Clinical+Decision+Support+and+Natural+Language+Processing+in+Medicine%3A+Systematic+Literature+Review%2E+J+Med+Internet+Res%2E+2)
2. Wiest IC, Ferber D, Zhu J, Treeck M van, Meyer SK, Juglan R, et al. From Text to Tables: A Local Privacy Preserving Large Language Model for Structured Information Retrieval from Medical Documents [Internet]. medRxiv; 2023 [cited 2024 Oct 24]. p. 2023.12.07.23299648. Available from: https://www.medrxiv.org/content/10.1101/2023.12.07.23299648v1 [↗](https://doi.org/10.1101/2023.12.07.23299648v1)
3. Lenz S, Ustjanzew A, Jeray M, Ressing M, Panholzer T. Can open source large language models be used for tumor documentation in Germany?—An evaluation on urological doctors’ notes. BioData Min. 2025 July 24;18(1):48. [find](https://scholar.google.com/scholar?q=Lenz+S%2C+Ustjanzew+A%2C+Jeray+M%2C+Ressing+M%2C+Panholzer+T%2E+Can+open+source+large+language+models+be+used+for+tumor+documentation+in+Germany%3F%E2%80%94An+evaluation+on+urological+doctors%E2%80%99+notes%2E+BioData+Min%2E+2025+Ju)
4. Federal Institute for Drugs and Medical Devices (BfArM). ICD-10-GM: International Statistical Classification of Diseases, German Modification [Internet]. 2024 [cited 2025 Mar 7]. Available from: [https://www.bfarm.de/EN/Code-systems/Classifications/ICD/ICD-10-GM/_node.html](https://www.bfarm.de/EN/Code-systems/Classifications/ICD/ICD-10-GM/_node.html) [↗](https://www.bfarm.de/EN/Code-systems/Classifications/ICD/ICD-10-GM/_node.html)
5. Holland J. Bekanntmachung - Aktualisierter einheitlicher onkologischer Basisdatensatz der Arbeitsgemeinschaft Deutscher Tumorzentren e. V. (ADT) und der Gesellschaft der epidemiologischen Krebsregister in Deutschland e. V. (GEKID). Bundesanzeiger. 2021 May 10;BAnz AT 12.07.2021 B4. [find](https://scholar.google.com/scholar?q=Holland+J%2E+Bekanntmachung+%2D+Aktualisierter+einheitlicher+onkologischer+Basisdatensatz+der+Arbeitsgemeinschaft+Deutscher+Tumorzentren+e%2E+V%2E+%28ADT%29+und+der+Gesellschaft+der+epidemiologischen+Krebsregiste)
6. Federal Institute for Drugs and Medical Devices (BfArM). ICD-O-3: International Classification of Diseases for Oncology [Internet]. 2024 [cited 2025 Mar 7]. Available from: [https://www.bfarm.de/EN/Code-systems/Classifications/ICD/ICD-O-3/_node.html](https://www.bfarm.de/EN/Code-systems/Classifications/ICD/ICD-O-3/_node.html) [↗](https://www.bfarm.de/EN/Code-systems/Classifications/ICD/ICD-O-3/_node.html)
7. Soroush A, Glicksberg BS, Zimlichman E, Barash Y, Freeman R, Charney AW, et al. Large Language Models Are Poor Medical Coders — Benchmarking of Medical Code Querying. NEJM AI. 2024 Apr 25;1(5):AIdbp2300040. [find](https://scholar.google.com/scholar?q=Soroush+A%2C+Glicksberg+BS%2C+Zimlichman+E%2C+Barash+Y%2C+Freeman+R%2C+Charney+AW%2C+et+al%2E+Large+Language+Models+Are+Poor+Medical+Coders+%E2%80%94+Benchmarking+of+Medical+Code+Querying%2E+NEJM+AI%2E+2024+Apr+25%3B1%285%29%3AAIdbp23)
8. Federal Institute for Drugs and Medical Devices (BfArM). ICD-10-GM Alphabetical index [Internet]. 2024 [cited 2025 Mar 7]. Available from: [https://www.bfarm.de/EN/Code-systems/Classifications/ICD/ICD-10-GM/Alphabetical-index/_node.html](https://www.bfarm.de/EN/Code-systems/Classifications/ICD/ICD-10-GM/Alphabetical-index/_node.html) [↗](https://www.bfarm.de/EN/Code-systems/Classifications/ICD/ICD-10-GM/Alphabetical-index/_node.html)
9. Federal Institute for Drugs and Medical Devices (BfArM). German procedure classification (Operationen-und Prozedurenschlüssel - OPS) [Internet]. 2024 [cited 2025 Mar 7]. Available from: [https://www.bfarm.de/EN/Code-systems/Classifications/OPS-ICHI/OPS/_node.html](https://www.bfarm.de/EN/Code-systems/Classifications/OPS-ICHI/OPS/_node.html) [↗](https://www.bfarm.de/EN/Code-systems/Classifications/OPS-ICHI/OPS/_node.html)
10. Bukharin A, Li S, Wang Z, Yang J, Yin B, Li X, et al. Data Diversity Matters for Robust Instruction Tuning. In: Al-Onaizan Y, Bansal M, Chen YN, editors. Findings of the Association for Computational Linguistics: EMNLP 2024 [Internet]. Miami, Florida, USA: Association for Computational Linguistics; 2024 [cited 2025 Mar 11]. p. 3411–25. Available from: https://aclanthology.org/2024.findings-emnlp.195/ [↗](https://aclanthology.org/2024.findings-emnlp.195/)
11. Vannata B, Zucca E. Primary extranodal B-cell lymphoma: current concepts and treatment strategies. Chin Clin Oncol. 2015 Mar;4(1):10–10. [find](https://scholar.google.com/scholar?q=Vannata+B%2C+Zucca+E%2E+Primary+extranodal+B%2Dcell+lymphoma%3A+current+concepts+and+treatment+strategies%2E+Chin+Clin+Oncol%2E+2015+Mar%3B4%281%29%3A10%E2%80%9310%2E)
12. Qwen Team. Model card of Qwen2.5-7B-1M on Hugging Face [Internet]. 2025 [cited 2025 Aug 15]. Available from: [https://huggingface.co/Qwen/Qwen2.5-7B-Instruct-1M](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct-1M) [↗](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct-1M)
13. Meta, Inc. Model card of Llama 3.1 8B on Hugging Face [Internet]. 2024 [cited 2025 Jan 15]. Available from: [https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct](https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct) [↗](https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct)
14. Qwen Team. Model card of Qwen3-8B on Hugging Face [Internet]. 2025 [cited 2025 Aug 19]. Available from: [https://huggingface.co/Qwen/Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B) [↗](https://huggingface.co/Qwen/Qwen3-8B)
15. Jiang A, Sablayrolles A, Tacnet A, Kothari A, Roux A, Mensch A, et al. Model card of Mistral Nemo on HuggingFace [Internet]. 2024 [cited 2024 Sept 25]. Available from: [https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407](https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407) [↗](https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407)
16. Qwen Team. Model card of Qwen3-14B on Hugging Face [Internet]. 2025 [cited 2025 Aug 15]. Available from: [https://huggingface.co/Qwen/Qwen3-14B](https://huggingface.co/Qwen/Qwen3-14B) [↗](https://huggingface.co/Qwen/Qwen3-14B)
17. Qwen Team. Model card of Qwen3-32B on Hugging Face [Internet]. 2025 [cited 2025 Aug 15]. Available from: [https://huggingface.co/Qwen/Qwen3-32B](https://huggingface.co/Qwen/Qwen3-32B) [↗](https://huggingface.co/Qwen/Qwen3-32B)
18. Jiang AQ, Sablayrolles A, Mensch A, Bamford C, Chaplot DS, Casas D de las, et al. Model card of Mixtral 8x7B on Hugging Face [Internet]. 2025 [cited 2025 Jan 15]. Available from: [https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1](https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1) [↗](https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1)
19. Meta AI. Model card of Llama 3.3 70B on Hugging Face [Internet]. 2024 [cited 2025 Aug 19]. Available from: [https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct](https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct) [↗](https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct)
20. Lenz S, Ustjanzew A, Jeray M, Ressing M, Panholzer T. Can open source large language models be used for tumor documentation in Germany? - An evaluation on urological doctors’ notes [Internet]. arXiv; 2025. Available from: http://arxiv.org/abs/2501.12106 [↗](https://arxiv.org/abs/2501.12106)
21. Wolf T, Debut L, Sanh V, Chaumond J, Delangue C, Moi A, et al. [Transformers: State-of-the-Art Natural Language Processing](https://md.research.elamur.ai/free/en/papers/wolf-2020-transformers-state-art-natural-language/). In: Liu Q, Schlangen D, editors. Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations [Internet]. Online: Association for Computational Linguistics; 2020 [cited 2024 Sept 5]. p. 38–45. Available from: https://aclanthology.org/2020.emnlp-demos.6 [↗](https://aclanthology.org/2020.emnlp-demos.6/)
22. Hu EJ, Shen Y, Wallis P, Allen-Zhu Z, Li Y, Wang S, et al. [LoRA: Low-Rank Adaptation of Large Language Models](https://md.research.elamur.ai/free/en/papers/hu-2021-lora-low-rank-adaptation-large/). In: International Conference on Learning Representations (ICLR) 2022 [Internet]. 2022 [cited 2025 Mar 11]. Available from: [https://openreview.net/forum?id=nZeVKeeFYf9](https://openreview.net/forum?id=nZeVKeeFYf9) [↗](https://openreview.net/forum?id=nZeVKeeFYf9)
23. Yang A, Li A, Yang B, Zhang B, Hui B, Zheng B, et al. [Qwen3 Technical Report](https://md.research.elamur.ai/free/en/papers/yang-2025-qwen3-technical-report/) [Internet]. arXiv; 2025 [cited 2025 May 22]. Available from: http://arxiv.org/abs/2505.09388 [↗](https://arxiv.org/abs/2505.09388)
24. Kreuzthaler M, Pfeifer B, Schulz S. Secondary Use of Clinical Problem List Descriptions for Bi-Encoder Based ICD-10 Classification. AMIA Annu Symp Proc. 2025 May 22;2024:620–7. [find](https://scholar.google.com/scholar?q=Kreuzthaler+M%2C+Pfeifer+B%2C+Schulz+S%2E+Secondary+Use+of+Clinical+Problem+List+Descriptions+for+Bi%2DEncoder+Based+ICD%2D10+Classification%2E+AMIA+Annu+Symp+Proc%2E+2025+May+22%3B2024%3A620%E2%80%937%2E)
25. Wiest IC, Ferber D, Zhu J, van Treeck M, Meyer SK, Juglan R, et al. Privacy-preserving large language models for structured medical information retrieval. Npj Digit Med. 2024 Sept 20;7(1):1–9. [find](https://scholar.google.com/scholar?q=Wiest+IC%2C+Ferber+D%2C+Zhu+J%2C+van+Treeck+M%2C+Meyer+SK%2C+Juglan+R%2C+et+al%2E+Privacy%2Dpreserving+large+language+models+for+structured+medical+information+retrieval%2E+Npj+Digit+Med%2E+2024+Sept+20%3B7%281%29%3A1%E2%80%939%2E)
26. Mittermeier A, Aßenmacher M, Schachtner B, Grosu S, Dakovic V, Kandratovich V, et al. [Automatic ICD-10 coding : Natural language processing for German MRI reports]. Radiol Heidelb Ger. 2024 Oct;64(10):793–800. [find](https://scholar.google.com/scholar?q=Mittermeier+A%2C+A%C3%9Fenmacher+M%2C+Schachtner+B%2C+Grosu+S%2C+Dakovic+V%2C+Kandratovich+V%2C+et+al%2E+%5BAutomatic+ICD%2D10+coding+%3A+Natural+language+processing+for+German+MRI+reports%5D%2E+Radiol+Heidelb+Ger%2E+2024+Oct%3B64%2810)
27. Devlin J, Chang MW, Lee K, Toutanova K. [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/) [Internet]. arXiv; 2019 [cited 2024 Oct 17]. Available from: http://arxiv.org/abs/1810.04805 [↗](https://arxiv.org/abs/1810.04805)
28. Böhringer D, Angelova P, Fuhrmann L, Zimmermann J, Schargus M, Eter N, et al. Automatic inference of ICD-10 codes from German ophthalmologic physicians’ letters using natural language processing. Sci Rep. 2024 Apr 19;14(1):9035. [find](https://scholar.google.com/scholar?q=B%C3%B6hringer+D%2C+Angelova+P%2C+Fuhrmann+L%2C+Zimmermann+J%2C+Schargus+M%2C+Eter+N%2C+et+al%2E+Automatic+inference+of+ICD%2D10+codes+from+German+ophthalmologic+physicians%E2%80%99+letters+using+natural+language+processing%2E+Sci+)
29. Mikolov T, Chen K, Corrado G, Dean J. arXiv.org. 2013 [cited 2025 Mar 19]. [Efficient Estimation of Word Representations in Vector Space](https://md.research.elamur.ai/free/en/papers/mikolov-2013-efficient-estimation-word-representations-vector/). Available from: https://arxiv.org/abs/1301.3781v3 [↗](https://arxiv.org/abs/1301.3781)
30. Lewis P, Perez E, Piktus A, Petroni F, Karpukhin V, Goyal N, et al. [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://md.research.elamur.ai/free/en/papers/lewis-2020-retrieval-augmented-generation-knowledge-intensive/) [Internet]. arXiv; 2021 [cited 2025 Mar 19]. Available from: http://arxiv.org/abs/2005.11401 [↗](https://arxiv.org/abs/2005.11401)
31. Klang E, Tessler I, Apakama DU, Abbott E, Glicksberg BS, Arnold M, et al. Assessing Retrieval-Augmented Large Language Model Performance in Emergency Department ICD-10-CM Coding Compared to Human Coders [Internet]. medRxiv; 2024 [cited 2024 Oct 22]. p. 2024.10.15.24315526. Available from: https://www.medrxiv.org/content/10.1101/2024.10.15.24315526v1 Acknowledgements We thank Dr. Meike Ressing for the discussion of selected difficult coding cases in the dataset, and we thank Marco Jeray for providing legal guidance on matters related to data extraction and analysis. Funding [↗](https://doi.org/10.1101/2024.10.15.24315526v1)

*Updated 29 October 2026*