Patient Perceived Empathy of an AI Chatbot for Atrial Fibrillation Education

Background:

Atrial fibrillation is a the most common arrhythmic disorder in the United States, with significant morbidity and healthcare cost burden. The number of patients with atrial fibrillation is expected to reach 12.1 million in 2030. AFib related annual incremental healthcare costs were estimated to be $6 to $26 billion based on 2010AF prevalence projections.

Lifestyle modification is one of the four primary pillars of atrial fibrillation care, and patient education is paramount to foster behavioral change. There is also evidence that early rhythm control using medications and ablation therapy can mitigate the overall burden of atrial fibrillation and improve patient quality of life among patients with other cardiovascular risk factors and/or heart failure. However, social determinants of health including health literacy significantly impact patient management, referral practices for procedural interventions, and ultimately patients' clinical outcomes. Anticoagulation prescription for prevention of thromboembolic complications, and adherence to anticoagulation are also known to be correlated with patients' medical literacy and disease awareness. These data point to three important unmet needs. First, there is a need for more robust and intentional patient education to foster behavioral change including lifestyle modification. Second, patient awareness and understanding of the disease is paramount to facilitate anticoagulation adherence and seeking specialist referral which in turn could mitigate AFib related morbidity and healthcare costs. Third, patient education must be delivered in empathetic language, at a (6th-8th) grade reading level, and translated to languages other than English per patient preference.

In that context, LLMs (large language models) such as ChatGPT (Open AI, San Francisco), with built in capabilities for language simplification and translation could play a foundational role in modern medicine. Current data show that LLMs(Chat GPT) are capable of answering patient queries in an empathetic and knowledgeable manner. Data showing improved readability of surgical consent forms simplified using ChatGPT, while maintaining medical information and medicolegal integrity demonstrate the potential of using LLMs for consent processes and shared decision making. However, it is important to note that providing appropriate guardrails to minimize probabilistic guess responses and fine tune the content for accuracy and empathy requires human interaction with the software. The availability of such prompt-engineered technology for patient education could have a profound impact on patient behavior, clinical outcomes and provider-patient relationship.