BACKGROUND:
Patient privacy protection is a core ethical requirement in healthcare and a critical element of women’s, maternal, and child health rights. Privacy risks are especially prominent in electronic medical records (EMRs) involving pregnancy, reproductive health, sexually transmitted diseases, and pediatric care, where inappropriate disclosure may result in stigma or discrimination. Traditional rule-based privacy protection methods are often technically complex and lack clinical flexibility. Recent advances in large language models (LLMs) offer new opportunities for adaptive and clinician-accessible privacy protection.

OBJECTIVES:
To develop and evaluate a no-code, LLM-based framework for identifying and protecting sensitive privacy information in Chinese EMRs, with a focus on pregnancy-related, maternal, adolescent, and pediatric health data.

METHODS:
Disease-specific EMR text data from Peking Union Medical College Hospital were analyzed, covering approximately 160,000 patients and over 4 million clinical text entries. Through prompt engineering, an LLM automatically annotated sensitive information and classified privacy risk levels across seven categories, including pregnancy-related and sexually transmitted disease data. A tiered protection strategy was applied according to sensitivity level. The workflow relied entirely on medical natural language prompts using the Qwen model, without additional programming. Performance was evaluated using precision and recall, with internal testing and external validation at another tertiary hospital.

RESULTS:
The framework achieved an average precision of 97% and recall of 95%. High-risk pregnancy-related and sexually transmitted disease data demonstrated similarly robust performance, with consistent results in external validation.

CONCLUSIONS:
This codeless LLM-based framework enables ethical and practical privacy protection in maternal, adolescent, and child healthcare, supporting women’s health rights while preserving clinical data utility.