Designing an Artificial Intelligence Transcription Tool to Support Multimodal Serious Illness Communication: Institutional Experience and Best Practices

Abstract

Background:

Communication during serious illness is often complex, emotionally charged, and cognitively demanding, particularly for patients and families with limited health literacy, sensory impairments, or language barriers. Existing communication supports inadequately address diverse information-processing needs, underscoring the need for thoughtfully designed, technology-enabled interventions.

Local Problem:

Within inpatient palliative care settings, family meetings frequently convey high-stakes information without a durable, accessible record of the conversation for patients and caregivers. This limitation may exacerbate disparities in understanding, recall, and decision making, especially for individuals who benefit from visual or multimodal communication supports.

Methods:

We employed a multiphase, iterative development and rapid refinement process to design an artificial intelligence (AI)-enabled communication support tool that generates written transcripts of palliative care family meetings. Development was guided by ethical principles, user-centered design, and early engagement with institutional data security, privacy, and regulatory leadership. The interdisciplinary development team included palliative care physicians, gerontologists, and a physician with training in public health who led the computational development.

Interventions:

The intervention involved the development of a multimodal communication tool that produces AI-generated written transcripts of clinician-led family meetings, designed to supplement—not replace—verbal communication. Development occurred in collaboration with institutional IT, privacy, and security leadership to enable future integration into existing clinical workflows and governance structures, while supporting transparency, clinician oversight, and institutional compliance.

Results:

The development process demonstrated that generation of high-quality, AI-generated written transcripts of family meetings is feasible within inpatient palliative care settings and institutional data security architectures. Early co-development with clinical leadership and regulatory stakeholders streamlined implementation while maintaining ethical and institutional safeguards.

Conclusions:

In the context of the rapid evolution of AI technologies, clinically led, ethically grounded development of AI-generated written transcripts for serious illness communication, involving institutional IT, privacy and security leadership, clinicians, patients, and caregivers, is feasible and may enhance accessible, patient-centered communication.