Balancing Efficiency and Empathy: Healthcare Providers' Perspectives on AI-Supported Workflows for Serious Illness Conversations in the Emergency Department

Paper Title

Balancing Efficiency and Empathy: Healthcare Providers' Perspectives on AI-Supported Workflows for Serious Illness Conversations in the Emergency Department

Publication Info

• Topic area: AI integration in healthcare communication workflows, specifically for Serious Illness Conversations in emergency departments.
• Keywords: Serious Illness Conversations, emergency department, artificial intelligence, healthcare workflows, empathy, efficiency, EHR, patient-provider communication, documentation, conversational agents.

Background and Problem

• Problem / challenge: Serious Illness Conversations (SICs) are rarely conducted in Emergency Departments (EDs due to fragmented patient information, time constraints, lack of conversational guidance, and burdensome documentation. Current workflows lack structured protocols and tools tailored to ED-specific constraints.
• Significance: SICs improve care alignment, reduce unnecessary interventions, and enhance patient outcomes, yet their absence in EDs leads to misaligned care and increased suffering for patients with life-threatening conditions.
• Motivation and related work: Prior research has explored SICs in non-ED settings and AI tools for healthcare communication but has not addressed the unique challenges of SICs in EDs. This paper investigates how ED providers conduct SICs, the barriers they face, and their perspectives on AI integration.

Solution

• Proposed approach: AI-supported workflows for SICs in EDs, designed to balance efficiency and empathy through ambient and peripheral systems that assist providers without undermining human connection.
• Novelty:
  1. Development of a four-phase SIC workflow tailored to ED settings: Identification, Preparation, Conduction, and Documentation.
  2. Empirical analysis of challenges and needs at each workflow phase.
  3. Design guidelines for integrating AI into SIC workflows while preserving clinical autonomy and relational empathy.
• Procedure and key techniques:
  • Semi-structured interviews with 11 ED providers to map SIC workflows and challenges.
  • Thematic analysis to identify barriers and opportunities for AI support.
  • Development of design guidelines for AI systems based on findings.

Results

• Concrete findings:
  • SIC workflow consists of four phases: Identification, Preparation, Conduction, and Documentation.
  • Key challenges include fragmented EHR data, lack of preparation time, difficulties in initiating and sustaining conversations, and burdensome documentation practices.
  • AI opportunities include summarizing patient information, generating personalized conversational cues, providing real-time support, automating documentation, and offering post-SIC feedback.
• Advantage over baselines: AI systems could reduce cognitive load, improve workflow efficiency, and enhance emotional readiness for providers, while preserving the human connection central to SICs.
• Experiments / evaluation: Semi-structured interviews with 11 ED providers from tertiary academic medical centers in the U.S., analyzed using thematic analysis to identify workflow stages, challenges, and AI opportunities.
• Limitations and future work:
  • Limited sample size (11 participants) and U.S.-specific healthcare context.
  • Lack of patient and caregiver perspectives.
  • Future work includes participatory design, prototype development, and cross-departmental evaluation of AI-supported SIC systems.

Summary

This paper identifies the unique challenges of conducting Serious Illness Conversations in Emergency Departments and explores how AI systems can support providers while preserving empathy. Through interviews with ED providers, the authors mapped a four-phase SIC workflow and highlighted barriers such as fragmented EHR data, time constraints, and emotional labor. Proposed AI solutions include tools for information synthesis, conversational scaffolding, and automated documentation, designed to operate peripherally and enhance workflow efficiency without undermining human connection. The findings contribute to the design of AI systems that balance efficiency and empathy in high-stakes clinical environments.