Patient empowerment through improved health literacy is essential to better health outcomes, yet electronic health record, or EHR, notes remain difficult for patients to understand because of complex clinical language. This talk presents an AI-driven framework for patient empowerment that addresses two goals: improving comprehension of health information and supporting positive behavior change.
We first introduce NoteAid, a natural language processing system that enhances EHR note readability by linking medical jargon to plain-language definitions. Randomized trials with crowdsourced participants and patients with diabetes in community hospitals showed that NoteAid significantly improved EHR note comprehension.
Building on this work, we developed Chatbot-NoteAid, which shifts patient education from passive reading to interactive dialogue. Using a multiagent architecture with supervised fine-tuning and reinforcement learning, the chatbot delivers personalized, incremental education. In Turing test evaluations, it outperformed nonexpert humans in educating users about EHR notes, highlighting the promise of conversational AI for health literacy.
To reduce health disparities among Hispanic populations, we developed Spanish-NoteAid, a low-resource medical translation system based on meta-learning. Our Chain-of-Dictionary approach, which integrates UMLS knowledge, enabled a 14-billion-parameter Phi-4 model to outperform GPT-4o in English-to-Spanish medical translation. Finally, we present ChatThero, a large language model-based therapist assistant designed to support behavior change and address the nationwide shortage of mental health professionals.
Together, these systems demonstrate how AI can empower patients by improving access to health information and supporting informed, sustainable lifestyle changes, ultimately enhancing patient engagement and health outcomes.