This research evaluates whether synthetic clinical notes can be a viable privacy-preserving alternative to de-identified real notes for research purposes.

• De-identification alone may not provide adequate privacy protection for clinical data
• Synthetic notes generated by large language models offer a promising alternative approach
• While synthetic notes preserve some utility for research tasks, they still face challenges in completely replacing real clinical data
• Synthetic data approach provides stronger privacy guarantees than traditional de-identification methods

For healthcare organizations, this research offers important insights into balancing data utility with patient privacy when sharing clinical information for research purposes.

De-identification is not enough: a comparison between de-identified and synthetic clinical notes