Integrating Gender-Sensitive Data into Clinical AI Systems: A Proof of Concept for Inclusive Healthcare

Abstract

Current IT tools in the medical system often lack gender-sensitive design, which can compromise both diagnosis and treatment. Since clinical evidence shows that men and women may respond differently to medications, this study developed an AI-based system to incorporate gender-specific data into clinical decision-making. A Retrieval-Augmented Generation (RAG) pipeline was implemented to analyze a curated dataset of 20 scientific articles using three distinct Large Language Models (LLMs): kronos483/Llama-3.2-3B-PubMed, deepseek-v2, and mistral-small-3.2. The evaluation revealed distinct trade-offs between the models. Mistral-small-3.2 achieved the highest average F 1 score, while deepseek-v2 delivered the highest average Precision but failed to assign gender-sensitivity scores. The performance of Llama-3.2 was comparable to Mistral’s, but its responses occasionally included faulty information. The findings confirm that this RAG-based approach could be a feasible method for generating promising, gender-sensitive clinical insights, demonstrating a path toward more equitable and personalized AI-supported healthcare.

Ähnliche Arbeiten