A Proof-of-Concept Study Using Generative AI for Structured PGHD Extraction in Remote Patient Monitoring and Mobile Health

This paper presents a Generative AI approach to streamline patient self-reporting in Remote Patient Monitoring (RPM) and mHealth. Patients submit free-text or spoken health updates, which are automatically converted into structured data. To ensure privacy, synthetic narratives were generated using WizardLM-2. Eight LLMs were evaluated for data extraction accuracy. Claude 3.5 Sonnet, GPT-4o, and DeepSeek Chat outperformed others, showing promise for real-world use. This approach could reduce patient burden, improve data completeness, and enhance engagement. This work serves as a proof-of-concept to identify high-performing models for structured PGHD extraction, laying the groundwork for future validation on real-world patient narratives and eventual clinical integration.