Use a customized large language model (LLM) agent to enhance Post-Abortion Care (PAC): A qualitative phenomenological study of user experiences among Chinese women (Preprint)

Abstract

The integration of artificial intelligence (AI) into healthcare, particularly in sensitive areas such as post-abortion care (PAC), remains under-explored. Traditional PAC services often struggle with efficiency, accessibility, consistency, and privacy concerns.

To explore the lived experiences of Chinese women receiving PAC supported using an in-house developed, customized large language model (LLM) agent, understand the practical benefits and limitations of using the AI-driven tool in a highly sensitive healthcare setting, and to inform future quantitative research and improvement initiatives.

We created a customized LLM AI agent using a Chinese LLM platform and a specialized knowledge base, and integrated it into the PAC services offered at a major healthcare institution in Southwest China. The agent was introduced to women patients on the PAC program. We collected qualitative data from a purposive sample of women using semi-structured interviews by phone. Data were analyzed using Colaizzi’s thematic analysis method to extract key themes and insights.

Of the 167 women introduced to the LLM agent, 37 (22.2%) engaged actively in using the agent and reported meaningful interactions. 12 participants were interviewed before data saturation was achieved. Analysis of interview transcripts revealed three main themes, including varied utilization of the LLM agent as a chat companion and information resource, unique opportunities that encouraged frequent use, and recognized strengths and lingering skepticism towards the technology. While many appreciated the 24/7 availability, ease of use, and privacy that the AI agent offered, concerns persisted about data security and the accuracy of the AI’s responses.

The study confirms the potential of AI to enhance the accessibility and quality of post-abortion care, providing emotional support and reliable information outside traditional settings. However, to foster broader acceptance and optimize the effectiveness of AI tools in healthcare, improvements in the AI tool, increased transparency about data handling, and enhanced user education on interacting with AI are necessary. Future research should focus on quantitative assessments of patient outcomes and satisfaction to further validate and refine the AI applications in healthcare.