---
title: "Medical Context Variation — paper by Sinha et al. (2026)"
description: "This case study asks whether that stylistic variation weakens pre-trained language models on medication-event context… 17 references."
authors: ["Aman Sinha", "Marianne Clausel", "Mathieu Constant", "Xavier Coubez"]
year: 2026
venue: "BioNLP 2026"
source: "https://aclanthology.org/2026.bionlp-1.70/"
doi: "10.18653/v1/2026.bionlp-1.70"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/sinha-2026-medical-context-variation-source-impairment/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/sinha-2026-medical-context-variation-source-impairment/)

[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/36/) / Medical Context Variation: A source of impairment for Event classification

# Medical Context Variation: A source of impairment for Event classification

**Authors:** [Aman Sinha](https://md.research.elamur.ai/free/en/authors/aman-sinha/), Marianne Clausel, [Mathieu Constant](https://md.research.elamur.ai/free/en/authors/mathieu-constant/) and Xavier Coubez · **Year:** 2026 · **Venue:** [BioNLP 2026](https://md.research.elamur.ai/free/en/venues/bionlp/) ([BioNLP 2026](https://md.research.elamur.ai/free/en/venues/bionlp/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.bionlp-1.70/) · [PDF](https://aclanthology.org/2026.bionlp-1.70.pdf) · [DOI 10.18653/v1/2026.bionlp-1.70](https://doi.org/10.18653/v1/2026.bionlp-1.70) · [OpenAlex](https://openalex.org/W7166802037)

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

## Summary

Clinical notes are mostly free text, and each clinician writes them differently. This case study asks whether that stylistic variation weakens pre-trained language models on medication-event context classification. The authors use the CMED corpus, which contains 500 annotated notes on Type II Diabetes patients from two hospitals. They fine-tune BERT, ModernBERT, ClinicalBERT and PubmedBERT to label sentences describing medication changes across five context dimensions. For each sentence they extract 220 linguistic features with the BEA tool and use OLS to relate those features to whether the ensemble predicted correctly. Many features, mainly lexico-semantic and syntactic ones, are significantly associated with errors. This suggests the models respond to surface style rather than clinical meaning.

To test and reduce this effect, they prompt GPT-3.5-Turbo to rewrite sentences along seven selected features, producing counterfactuals. These rewrites shift predictions in the expected direction in about 60% of cases. A manual risk audit shows that some rewrites distort clinical content. The work matters because it shows that robustness to writing style cannot be assumed for clinical NLP, and that LLM-based style normalization is helpful but needs clinical checking.

## Key points

- **Data and tasks:**
  - Data: the CMED dataset, derived from the n2c2 contextual medication event extraction task.
  - Labels: sentences with a disposition (medication change) are classified for Action, Negation, Temporality, Certainty and Actor.
- **Feature extraction:**
  - Tool: the BEA extractor (Lee and Lee, 2023), run with SciSpacy (en_core_sci_sm).
  - Output: 220 sentence-level features covering surface, syntax, lexico-semantics and discourse.
  - Feature families: discourse features dominate for Actor and Negation, while lexico-semantic and syntactic ones dominate elsewhere.
- **Selected features:** seven features were used for the counterfactual rewrites, including:
  - #unique_nouns
  - simple_verbs_variation
  - root_verbs_variation
  - #named_entities_person
  - simple_type_token_ratio
  - avg_syllables_per_word
- **Effect of the rewrites:**
  - Expected behavior occurred in roughly 60% of cases.
  - Negation stayed largely stable.
  - Actor and Certainty were the most sensitive, under simple_verbs_variation and simple_type_token_ratio respectively.
- **Risk audit:**
  - Scope: 350 rewritten sentences (50 sentences × 7 features), each rated low, moderate or high risk.
  - Safest feature: root_verbs_variation, with 98% low-risk rewrites.
  - Riskiest features: #named_entities_person and simple_verbs_variation, each with 20.4% high-risk rewrites.
  - The authors suggest that low- and moderate-risk rewrites could serve as data augmentation.

*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

- [Centre National de la Recherche Scientifique](https://md.research.elamur.ai/free/en/labs/centre-national-de-la-recherche-scientifique/) — Research institute, Paris, France · 168 works · 62,369 citations
- [Université de Lorraine](https://md.research.elamur.ai/free/en/universities/universite-de-lorraine/) — University, Nancy, France · 20 works · 945 citations
- [Analyse et Traitement Informatique de la Langue Française](https://md.research.elamur.ai/free/en/labs/analyse-et-traitement-informatique-de-la-langue-francaise/) — Research institute, Nancy, France · 6 works · 390 citations

Also on the byline: [Institut Élie Cartan de Lorraine](https://ror.org/04rvw8791) · [Institut Strauss](https://ror.org/03k1y0r37).

📚 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 · [💭 Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/) · 1,552 works · [💭 Social Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/social-psychology/) · 306 works

## Cite

Cite as: Aman Sinha, Marianne Clausel, Mathieu Constant and Xavier Coubez. 2026. Medical Context Variation: A source of impairment for Event classification. BioNLP 2026.

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

```bibtex
@inproceedings{sinha2026medical,
  title = {{Medical Context Variation: A source of impairment for Event classification}},
  author = {Aman Sinha, Marianne Clausel, Mathieu Constant and Xavier Coubez},
  booktitle = {BioNLP 2026},
  year = {2026},
  doi = {10.18653/v1/2026.bionlp-1.70},
  url = {https://doi.org/10.18653/v1/2026.bionlp-1.70}
}
```

- [← previous MedEvi-NS at ArchEHR-QA 2026: Using Clinical Reasoning Principles to Improve Zero-shot Capabilities of Large Language Models in Evidence Alignment](https://md.research.elamur.ai/free/en/papers/sun-2026-medevi-ns-archehr-qa-2026/)
- [next → Medical Summarization in Practice: Design, Deployment, and Analysis of a Clinical Summarization System for a German Hospital](https://md.research.elamur.ai/free/en/papers/rauf-2026-medical-summarization-practice-design-deployment/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/36/)

## References

17 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Jørgen Bansler, Erling Havn, Kjeld Schmidt, Troels Mønsted, Helen Petersen, and Jesper Svendsen. 2016. Cooperative epistemic work in medical practice: An analysis of physicians’ clinical notes. Computer Supported Cooperative Work, 25. [find](https://scholar.google.com/scholar?q=J%C3%B8rgen+Bansler%2C+Erling+Havn%2C+Kjeld+Schmidt%2C+Troels+M%C3%B8nsted%2C+Helen+Petersen%2C+and+Jesper+Svendsen%2E+2016%2E+Cooperative+epistemic+work+in+medical+practice%3A+An+analysis+of+physicians%E2%80%99+clinical+notes%2E+Comput)
2. Justin Cadwallader, Kenneth Spry, Justin Morea, AL Russ, Jon Duke, and Michael Weiner. 2013. Design of a medication reconciliation application. Applied clinical informatics, 4(01):110–125. [find](https://scholar.google.com/scholar?q=Justin+Cadwallader%2C+Kenneth+Spry%2C+Justin+Morea%2C+AL+Russ%2C+Jon+Duke%2C+and+Michael+Weiner%2E+2013%2E+Design+of+a+medication+reconciliation+application%2E+Applied+clinical+informatics%2C+4%2801%29%3A110%E2%80%93125%2E)
3. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2018. [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/). CoRR, abs/1810.04805.
4. Božena Džuganová. 2019. Medical language-a unique linguistic phenomenon: Engleski. Jahr–European Journal of Bioethics, 10(1):129–145. [find](https://scholar.google.com/scholar?q=Bo%C5%BEena+D%C5%BEuganov%C3%A1%2E+2019%2E+Medical+language%2Da+unique+linguistic+phenomenon%3A+Engleski%2E+Jahr%E2%80%93European+Journal+of+Bioethics%2C+10%281%29%3A129%E2%80%93145%2E)
5. Neville W Goodman and Martin B Edwards. 2014. Medical writing: a prescription for clarity. Cambridge University Press. [find](https://scholar.google.com/scholar?q=Neville+W+Goodman+and+Martin+B+Edwards%2E+2014%2E+Medical+writing%3A+a+prescription+for+clarity%2E+Cambridge+University+Press%2E)
6. Yu Gu, Robert Tinn, Hao Cheng, Michael Lucas, Naoto Usuyama, Xiaodong Liu, Tristan Naumann, Jianfeng Gao, and Hoifung Poon. 2021. [Domain-specific language model pretraining for biomedical natural language processing](https://md.research.elamur.ai/free/en/papers/gu-2020-domain-specific-language-model-pretraining/). ACM Trans. Comput. Healthcare, 3(1).
7. Kexin Huang, Jaan Altosaar, and Rajesh Ranganath. 2019. [Clinicalbert: Modeling clinical notes and predicting hospital readmission](https://md.research.elamur.ai/free/en/papers/huang-2019-clinicalbert-modeling-clinical-notes-predicting/). arXiv preprint arXiv:1904.05342.
8. Arun K Kuchibhotla, Lawrence D Brown, Andreas Buja, and Junhui Cai. 2019. All of linear regression. arXiv preprint arXiv:1910.06386. — [arXiv:1910.06386](https://arxiv.org/abs/1910.06386) [find](https://scholar.google.com/scholar?q=Arun+K+Kuchibhotla%2C+Lawrence+D+Brown%2C+Andreas+Buja%2C+and+Junhui+Cai%2E+2019%2E+All+of+linear+regression%2E+arXiv+preprint+arXiv%3A1910%2E06386%2E+%E2%80%94+%5BarXiv%3A1910%2E06386%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F1910%2E06386%29)
9. Bruce W. Lee and Jason Lee. 2023. LFTK: Handcrafted features in computational linguistics. In Proceedings of the 18th Workshop on Innovative Use of NLP for Building Educational Applications (BEA 2023), pages 1–19, Toronto, Canada. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Bruce+W%2E+Lee+and+Jason+Lee%2E+2023%2E+LFTK%3A+Handcrafted+features+in+computational+linguistics%2E+In+Proceedings+of+the+18th+Workshop+on+Innovative+Use+of+NLP+for+Building+Educational+Applications+%28BEA+2023%29)
10. Leonardo Campillos Llanos, Dhouha Bouamor, Pierre Zweigenbaum, and Sophie Rosset. 2016. Managing linguistic and terminological variation in a medical dialogue system. In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC’16), pages 3167–3173. [find](https://scholar.google.com/scholar?q=Leonardo+Campillos+Llanos%2C+Dhouha+Bouamor%2C+Pierre+Zweigenbaum%2C+and+Sophie+Rosset%2E+2016%2E+Managing+linguistic+and+terminological+variation+in+a+medical+dialogue+system%2E+In+Proceedings+of+the+Tenth+Inter)
11. Diwakar Mahajan, Jennifer J. Liang, and Ching-Huei Tsou. 2022. Toward Understanding Clinical Context of Medication Change Events in Clinical Narratives. AMIA Annual Symposium Proceedings, 2021:833– 842. [find](https://scholar.google.com/scholar?q=Diwakar+Mahajan%2C+Jennifer+J%2E+Liang%2C+and+Ching%2DHuei+Tsou%2E+2022%2E+Toward+Understanding+Clinical+Context+of+Medication+Change+Events+in+Clinical+Narratives%2E+AMIA+Annual+Symposium+Proceedings%2C+2021%3A833%E2%80%93+84)
12. SLM Mitchell. 2007. Medical writing: A prescription for clarity. models. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 6707–6723, Online. Association… [find](https://scholar.google.com/scholar?q=SLM+Mitchell%2E+2007%2E+Medical+writing%3A+A+prescription+for+clarity%2E+models%2E+In+Proceedings+of+the+59th+Annual+Meeting+of+the+Association+for+Computational+Linguistics+and+the+11th+International+Joint+Con)
13. Zhuo Xu, Yue Wang, Lu Bai, and Lixin Cui. 2022. Writing style aware document-level event extraction. arXiv preprint arXiv:2201.03188. — [arXiv:2201.03188](https://arxiv.org/abs/2201.03188) [find](https://scholar.google.com/scholar?q=Zhuo+Xu%2C+Yue+Wang%2C+Lu+Bai%2C+and+Lixin+Cui%2E+2022%2E+Writing+style+aware+document%2Dlevel+event+extraction%2E+arXiv+preprint+arXiv%3A2201%2E03188%2E+%E2%80%94+%5BarXiv%3A2201%2E03188%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F2201%2E03188%29)
14. A Dong Nguyen, Laura Rosseel, and Jack Grieve. 2021. On learning and representing social meaning in nlp: a sociolinguistic perspective. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human language technologies, pages 603–612. [find](https://scholar.google.com/scholar?q=A+Dong+Nguyen%2C+Laura+Rosseel%2C+and+Jack+Grieve%2E+2021%2E+On+learning+and+representing+social+meaning+in+nlp%3A+a+sociolinguistic+perspective%2E+In+Proceedings+of+the+2021+Conference+of+the+North+American+Chap)
15. Trang Pham, Truyen Tran, Dinh Phung, and Svetha Venkatesh. 2017. Predicting healthcare trajectories from medical records: A deep learning approach. Journal of biomedical informatics, 69:218–229. [find](https://scholar.google.com/scholar?q=Trang+Pham%2C+Truyen+Tran%2C+Dinh+Phung%2C+and+Svetha+Venkatesh%2E+2017%2E+Predicting+healthcare+trajectories+from+medical+records%3A+A+deep+learning+approach%2E+Journal+of+biomedical+informatics%2C+69%3A218%E2%80%93229%2E)
16. More Information for CMED Dataset The CMED dataset contains clinical notes that came from the Partners HealthCare Electronic Medical Records (EMR). EMR at Partners HealthCare comprises a platform shared by two large academic tertiary hospitals – Massachusetts General Hospital (MGH) and Brigham and W… [find](https://scholar.google.com/scholar?q=More+Information+for+CMED+Dataset+The+CMED+dataset+contains+clinical+notes+that+came+from+the+Partners+HealthCare+Electronic+Medical+Records+%28EMR%29%2E+EMR+at+Partners+HealthCare+comprises+a+platform+shar)
17. Action Actor Certainty Negation Temporality What is the change discussed? Who initiated the change? How likely is this change to have occurred / will occur? Is the change being discussed negated? When is this change intended to occur? Alexis Ross, Ana Marasović, and Matthew Peters. 2021. [Explaining…](https://md.research.elamur.ai/free/en/papers/ross-2021-explaining-nlp-models-minimal-contrastive/)

*Updated 1 November 2026*