A Natural Language Processing Model to Identify Confidential Content in Adolescent Clinical Notes

Area of Science:

• Medical Informatics
• Natural Language Processing
• Adolescent Health

Background:

• The 21st Century Cures Act requires immediate electronic health information release.
• Adolescent confidentiality requires special consideration during information sharing.
• Detecting confidential content in clinical notes can support adolescent privacy efforts.

Purpose of the Study:

• To evaluate a natural language processing (NLP) algorithm's ability to detect confidential content in adolescent clinical progress notes.

Main Methods:

• A logistic regression model was trained on 1,200 manually annotated adolescent progress notes.
• The model provided sentence-level and note-level probability estimates for confidential content.
• Prospective validation was performed on 240 notes, followed by a pilot intervention.

Main Results:

• The NLP model achieved high accuracy (90% and 88% AUC in test and validation cohorts).
• Prevalence of notes with confidential content was 21-22%.
• Pilot deployment demonstrated efficiency gains over manual review.

Conclusions:

• NLP algorithms can accurately identify confidential content in clinical notes.
• NLP deployment supports operational efforts to maintain adolescent confidentiality.
• This approach aids compliance with information blocking mandates while protecting privacy.