Systematic Review of Large Language Models in Maternal and Child Health: A New Milestone in Medical AI Applications

This article summarizes a systematic review on the application of large language models (LLMs) in the field of maternal and child health. It focuses on exploring specific scenarios of LLMs in clinical practice, medical education, and clinical decision-making, analyzes their impact on patient safety, and looks forward to the future development direction of medical AI. This review provides a reference for understanding the current status and challenges of LLMs in the field of maternal and child health.

Maternal and child health is a sensitive and complex field in the medical system, covering multiple links from pre-pregnancy consultation to postnatal care, which requires precise and personalized information support. Traditional medical information systems are difficult to meet dynamic needs, while the natural language understanding and generation capabilities of LLMs can fill this gap.

This review adopts a rigorous systematic literature search, covering multiple academic databases and gray literature, focusing on four core issues: 1. Specific scenarios where medical personnel use LLMs in the field of maternal and child health; 2. Actual effects of applications; 3. The role of LLMs in medical education; 4. Impact on patient safety and potential risks.

Clinical Practice Support: Assist in medical history collection to generate structured medical records, provide personalized diagnosis and treatment suggestions (e.g., gestational diabetes management), and translate medical terms into plain language to facilitate patient communication. Medical Education: Simulate clinical scenarios for trainees to practice core skills, and generate realistic case discussion materials to cultivate clinical thinking. Clinical Decision Support: As a source of "second opinion", it assists in the diagnosis and treatment ideas and evidence synthesis of rare/difficult cases, supplementing doctors' judgments.

Potential for Improvement: Standardized information processing reduces human errors, real-time monitoring and early warning of risks, and assists in checking drug doses and interactions to ensure medication safety. Potential Risks: LLM "hallucinations" may generate incorrect information; medical data privacy and security issues; training data bias leading to algorithmic bias that exacerbates medical inequality.

Technological Improvements: Combine Retrieval-Augmented Generation (RAG) to reduce hallucinations, enhance decision interpretability, and improve personalized recommendation capabilities (combining individual information such as genomics). Regulation and Ethics: Establish technical standards and access thresholds; define the boundary of responsibility between AI assistance and doctors; improve the patient informed consent mechanism.

The application of LLMs in the field of maternal and child health is in a stage of rapid development. They are powerful assistants for medical staff but cannot replace human professional judgment and humanistic care. The ultimate goal of the technology is to serve maternal and child health, so that every mother and newborn can receive safe and high-quality medical care.