Extending a Clinical Pediatric Growth Chart App Using a Large Language Model

Monitoring a child’s growth is vital for early detection of disorders. This study explores integrating a large language model (LLM) agent into the SMART on FHIR Growth Chart App to assist pediatricians in identifying growth abnormalities. Using a User-Centered Design (UCD) approach, we gathered pediatrician feedback to refine an AI tab analyzing synthetic patient data. The system was implemented using the OpenAI Assistants API with Retrieval Augmented Generation (RAG) and tested for usability and functionality with three pediatricians in an evaluation of model responses for growth abnormality detection, patient history analysis, recommended specialist referral, differential diagnosis and executive summary. Results showed that the agent can achieve high levels of usability when integrated into a clinical setting. However, while the agent accurately analyzed three of five synthetic patients, its responses to differential diagnoses and specialist referrals were insufficient. This proof of concept highlights the potential of AI tools in pediatrics but also underscores the need for improved accuracy in future developments. We have open-sourced the agentic app.