Robust privacy amidst innovation with large language models through a critical assessment of the risks
Bibliographic record
Abstract
OBJECTIVE: This study evaluates the integration of electronic health records (EHRs) and natural language processing (NLP) with large language models (LLMs) to enhance healthcare data management and patient care, focusing on using advanced language models to create secure, Health Insurance Portability and Accountability Act-compliant synthetic patient notes for global biomedical research. MATERIALS AND METHODS: The study used de-identified and re-identified versions of the MIMIC III dataset with GPT-3.5, GPT-4, and Mistral 7B to generate synthetic clinical notes. Text generation employed templates and keyword extraction for contextually relevant notes, with One-shot generation for comparison. Privacy was assessed by analyzing protected health information (PHI) occurrence and co-occurrence, while utility was evaluated by training an ICD-9 coder using synthetic notes. Text quality was measured using ROUGE (Recall-Oriented Understudy for Gisting Evaluation) and cosine similarity metrics to compare synthetic notes with source notes for semantic similarity. RESULTS: The analysis of PHI occurrence and text utility via the ICD-9 coding task showed that the keyword-based method had low risk and good performance. One-shot generation exhibited the highest PHI exposure and PHI co-occurrence, particularly in geographic location and date categories. The Normalized One-shot method achieved the highest classification accuracy. Re-identified data consistently outperformed de-identified data. DISCUSSION: Privacy analysis revealed a critical balance between data utility and privacy protection, influencing future data use and sharing. CONCLUSION: This study shows that keyword-based methods can create synthetic clinical notes that protect privacy while retaining data usability, potentially improving clinical data sharing. The use of dummy PHIs to counter privacy attacks may offer better utility and privacy than traditional de-identification.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.003 | 0.005 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.002 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".