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

Menglin Zhao, Zhuorui Yong, Ruijia Guan, Kai-Wei Chang, Adrian Haimovich, Kei Ouchi, Timothy Bickmore, Zhan Zhang, Bingsheng Yao, Dakuo Wang, Smit Desai

TL;DR

This paper investigates how artificial intelligence can support Serious Illness Conversations (SICs) in the high-stress environment of Emergency Departments (EDs). By interviewing 11 healthcare providers, the authors map a four-stage clinical workflow and propose design guidelines for ambient, peripheral AI systems that enhance operational efficiency without disrupting the crucial empathetic connection between doctors and patients.

What Problem Does This Paper Address?

Serious Illness Conversations (SICs)—discussions regarding values, goals, and care preferences for patients facing life-threatening conditions—are critical for aligning medical interventions with patient wishes. While early SICs are proven to reduce unnecessary suffering and improve care quality, they rarely happen in Emergency Departments (EDs).

Unlike planned outpatient care, ED providers operate under extreme time constraints, manage multiple concurrent crises, and often treat unfamiliar patients without accessible medical histories. The paper addresses this gap by exploring the unique workflow and barriers ED providers face when conducting these conversations, and examines how AI can be thoughtfully integrated to support clinicians without dehumanizing the patient experience.

Methods

The authors conducted semi-structured interviews with 11 ED providers (9 emergency medicine physicians and 2 nurses) working at tertiary academic medical centers in the United States. Using iterative thematic analysis, they synthesized a novel, ED-specific four-stage workflow for SICs: Identification, Preparation, Conduction, and Documentation. They then systematically analyzed the specific challenges at each stage and gathered provider perspectives on the opportunities, requirements, and risks of introducing AI tools into this high-stakes environment.

The study procedure used by the authors to extract workflows, challenges, and design guidelines.

Datasets

Not discussed in the paper. (As a qualitative interview study, no machine learning datasets were trained or evaluated).

Key Insights and Ideas

  • The Empathy-Efficiency Paradox: A core conceptual contribution of the paper. While providers fear that technology might make interactions feel mechanical, they simultaneously request AI features to offload cognitive burdens (like searching fragmented records or typing notes). The authors argue that reducing cognitive load actually creates the mental and emotional space required for genuine human empathy.
  • Ambient and Peripheral AI Design: To avoid disrupting eye contact and trust during the sensitive Conduction phase, AI tools must act as a "quiet listener." This means using ambient audio processing and displaying subtle cues on peripheral screens rather than forcing clinicians to look at a computer terminal.
  • Rationale-Driven Documentation: Standard AI summarization often captures final decisions (e.g., "Do Not Intubate") but misses the why (the patient's core values, such as wanting to remain communicative for family). Providers emphasize that downstream care teams need the underlying rationale to maintain longitudinal care continuity.
  • Breaking the Post-SIC "Negative Loop": Due to a lack of formal training, providers often experience anxiety or a sense of failure after emotionally draining conversations, creating a dread of future SICs. Immediate post-hoc AI feedback can act as a micro-learning engine to break this loop and build clinical confidence.

Key Results

The paper maps out the current challenges against AI-supported opportunities across the four synthesized phases of the clinical workflow:

Comparison of current ED workflows versus proposed AI-supported workflows for Serious Illness Conversations.

Based on these findings, the authors formulated four core Design Guidelines (DG) for future healthcare AI systems:

  1. DG1 (Identification & Preparation): Provide at-a-glance information scaffolding. AI should automatically synthesize fragmented Electronic Health Record (EHR) data and explicitly highlight missing information gaps.
  2. DG2 (Conduction): Offer non-intrusive ambient and peripheral conversational support. AI should passively monitor the room and offer lightweight, non-evaluative phrasing suggestions during conversational impasses.
  3. DG3 (Documentation): Generate rationale-driven, provider-curated notes. AI should draft summaries that emphasize the patient's underlying values and track longitudinal progress across departments.
  4. DG4 (Feedback Mechanism): Provide immediate post-SIC feedback. AI should analyze the interaction to offer concise, actionable alternative phrasing to help clinicians reflect and improve.
Summary of the key challenges, AI opportunities, and design guidelines mapped across the SIC workflow.

Limitations and Future Work

The study relies on a small sample size of 11 providers, all based within the U.S. healthcare system at tertiary academic medical centers, which may limit generalizability to community hospitals or international contexts. Furthermore, the study focuses exclusively on healthcare providers; the perspectives of patients and their families were not captured.

As a next step, the authors plan to transition these guidelines into concrete system designs. They intend to engage ED providers through participatory design to build a functional prototype and evaluate its efficacy through usability testing in real-world, high-pressure ED environments.

Summary updated

Abstract

Serious Illness Conversations (SICs), discussions about values and care preferences for patients with life-threatening illness, rarely occur in Emergency Departments (EDs), despite evidence that early conversations improve care alignment and reduce unnecessary interventions. We interviewed 11 ED providers to identify challenges in SICs and opportunities for technology support, with a focus on AI. Our analysis revealed a four-stage SIC workflow (identification, preparation, conduction, documentation) and barriers at each stage, including fragmented patient information, limited time and space, lack of conversational guidance, and burdensome documentation. Providers expressed interest in AI systems for synthesizing information, supporting real-time conversations, and automating documentation, but emphasized concerns about preserving human connection and clinical autonomy. This tension highlights the need for technologies that enhance efficiency without undermining the interpersonal nature of SICs. We propose design guidelines for ambient and peripheral AI systems to support providers while preserving the essential humanity of these conversations.
BibTeX
@inproceedings{zhao2026balancing,
  author = {Menglin Zhao and Zhuorui Yong and Ruijia Guan and Kai-Wei Chang and Adrian Haimovich and Kei Ouchi and Timothy Bickmore and Zhan Zhang and Bingsheng Yao and Dakuo Wang and Smit Desai},
  title = {{Balancing Efficiency and Empathy: Healthcare Providers' Perspectives on AI-Supported Workflows for Serious Illness Conversations in the Emergency Department}},
  booktitle = {CHI},
  year = {2026},
  doi = {10.1145/3772318.3791301},
}
Figure