Journal article · 2026 · Open access
Assessment of differentially private fine-tuning of large language models for synthetic clinical note generation
Intelligent Systems with Applications, 30, 200659 · Published 7 April 2026 · DOI 10.1016/j.iswa.2026.200659
In brief
“Assessment of differentially private fine-tuning of large language models for synthetic clinical note generation” is a 2026 journal article by Atiquer Rahman Sarkar and 4 co-authors, published in Intelligent Systems with Applications. OpenAlex counts 1 citations as of 19 November 2026. It is open access under a CC BY-NC-ND licence: the full text is free to read at Intelligent Systems with Applications.
- Citations
- 1
- Most cited in
- 2026 (1)
- Field-weighted impact
- 1.08
- References
- 34
- Authors
- 5
Abstract
The sharing of clinical notes is limited by privacy regulations such as HIPAA, PIPEDA, and GDPR. Synthetic data generation (SDG) has emerged as a promising approach, but it lacks quantifiable privacy guarantee. This study investigates three differentially private (DP) fine-tuned large language models for generating synthetic clinical notes, employing a recently proposed context-aware note generation technique. We investigate the impact of differential privacy bound on two key aspects: (i) the reappearance of protected health information (PHI), and (ii) utility preservation, measured via linguistic and semantic similarity metrics in the generated notes. We experimented with various privacy budgets and generation settings. Our results, obtained on the I2B2-2014 de-identification corpus, suggest that differential privacy finetuning does not automatically provide privacy guarantees. Some synthetic notes were observed to leak protected health information. Consequently, the synthetic notes generated from the DP-finetuned LLMs are not compliant with the HIPAA privacy rule for de-identification standards. Rank-based statistical analyses demonstrate that the expected positive monotonic association between the privacy budget ϵ and leakage is not consistently significant when comparing DP-finetuned models (i.e., ϵ ∈ { 1 , 4 , 8 } ). More importantly, the data indicate that, while leakage increases when moving to a non-private baseline ( ϵ = ∞ ), the leakage difference shows a substantial baseline of PHI reappearance even under the strict privacy setting (e.g., at ϵ = 1 , 27 of 1304 notes are affected in the I2B2-2014 corpus). This suggests that a significant portion of leakage is an inherent artifact of the fine-tuning process, which is not fully mitigated by the DP mechanism, positioning ϵ -dependent leakage as only one component of the overall risk. The findings contribute to advancing the discourse on privacy-aware data sharing in healthcare research.
Cite this paper
Formatted from the catalogue record. Check names, capitals and page numbers against the paper before you submit.
APA 7
Sarkar, A. R., Sarmin, F. J., Mouheb, D., Fung, B. C. M., & Mohammed, N. (2026). Assessment of differentially private fine-tuning of large language models for synthetic clinical note generation. Intelligent Systems with Applications, 30, 200659. https://doi.org/10.1016/j.iswa.2026.200659
MLA 9
Sarkar, Atiquer Rahman, et al. “Assessment of differentially private fine-tuning of large language models for synthetic clinical note generation.” Intelligent Systems with Applications, vol. 30, 2026, p. 200659. https://doi.org/10.1016/j.iswa.2026.200659.
Chicago
Sarkar, Atiquer Rahman, Fatima Jahan Sarmin, Djedjiga Mouheb, Benjamin C. M. Fung, and Noman Mohammed. 2026. “Assessment of differentially private fine-tuning of large language models for synthetic clinical note generation.” Intelligent Systems with Applications 30: 200659. https://doi.org/10.1016/j.iswa.2026.200659.
Harvard
Sarkar, A. R. et al. (2026) ‘Assessment of differentially private fine-tuning of large language models for synthetic clinical note generation’, Intelligent Systems with Applications, 30, p. 200659. doi:10.1016/j.iswa.2026.200659.
IEEE
A. R. Sarkar, F. J. Sarmin, D. Mouheb, B. C. M. Fung, and N. Mohammed, “Assessment of differentially private fine-tuning of large language models for synthetic clinical note generation,” Intelligent Systems with Applications, vol. 30, p. 200659, 2026, doi: 10.1016/j.iswa.2026.200659.
BibTeX
@article{sarkar2026assessment,
title = {{Assessment of differentially private fine-tuning of large language models for synthetic clinical note generation}},
author = {Sarkar, Atiquer Rahman and Sarmin, Fatima Jahan and Mouheb, Djedjiga and Fung, Benjamin C. M. and Mohammed, Noman},
journal = {Intelligent Systems with Applications},
year = {2026},
volume = {30},
pages = {200659},
doi = {10.1016/j.iswa.2026.200659},
url = {https://doi.org/10.1016/j.iswa.2026.200659}
}
Download for a reference manager: BibTeX (.bib) · RIS for Zotero, Mendeley and EndNote (.ris)
Topic and keywords
Physical Sciences › Computer Science › Artificial Intelligence › Privacy-Preserving Technologies in Data
- differential privacy
- large language models
- synthetic data generation
- privacy leakage
- protected health information
- de-identification
- HIPAA
- privacy budget
Work with this paper
Most-cited papers in Computer Science
- ImageNet classification with deep convolutional neural networksAlex Krizhevsky, Ilya Sutskever, Geoffrey E. Hinton · Communications of the ACM · 2017 · 114,726 citations
- Long Short-Term MemorySepp Hochreiter, Jürgen Schmidhuber · Neural Computation · 1997 · 109,097 citations
- A Mathematical Theory of CommunicationClaude E. Shannon · Bell System Technical Journal · 1948 · 91,736 citations
- Deep learningYann LeCun, Yoshua Bengio, Geoffrey E. Hinton · Nature · 2015 · 90,226 citations
- Faster R-CNN: Towards Real-Time Object Detection with Region Proposal NetworksShaoqing Ren, Kaiming He, Ross Girshick and 1 more · IEEE Transactions on Pattern Analysis and Machine Intelligence · 2016 · 74,143 citations
Questions about this paper
How do I cite this paper in APA style?
Sarkar, A. R., Sarmin, F. J., Mouheb, D., Fung, B. C. M., & Mohammed, N. (2026). Assessment of differentially private fine-tuning of large language models for synthetic clinical note generation. Intelligent Systems with Applications, 30, 200659. https://doi.org/10.1016/j.iswa.2026.200659
How many times has this paper been cited?
1 times, according to OpenAlex on 19 November 2026. Google Scholar and Scopus count different sets of sources, so their numbers differ.
Is this paper free to read?
Yes. It is open access at Intelligent Systems with Applications: sign in and open the full text from this page.
Who wrote this paper?
Atiquer Rahman Sarkar, Fatima Jahan Sarmin, Djedjiga Mouheb, Benjamin C. M. Fung, Noman Mohammed.
What is the DOI of this paper?
10.1016/j.iswa.2026.200659, permanently at https://doi.org/10.1016/j.iswa.2026.200659. Sign in and open it from this page.
Bibliographic data from OpenAlex (CC0), last checked 19 November 2026. ResearchGuru does not host this paper; once you are signed in, the links open it at the publisher or the repository that does.