Skip to main content

Implementing AI in Healthcare: Bridging the Gap Between Technical, Cognitive, and Sociotechnical Perspectives

Implementing AI in Healthcare: Bridging the Gap Between Technical, Cognitive, and Sociotechnical Perspectives

The implementation of clinical decision support (CDS) tools is notoriously challenging. Safely, effectively, and equitably incorporating AI into CDS implementations holds the promise of improving precision and appropriateness of CDS, providing improved information synthesis, and reducing clinical cognitive burden. AI-specific barriers exacerbate CDS implementation hurdles. The effectiveness of AI-enhanced CDS tools in actual clinical settings remains underexplored, with limited evidence: (i) few randomized clinical trials have assessed the impacts of AI-enabled CDS in clinical care, and (ii) there are few no studies of employing AI to enhance the design and analysis of cognitive-socio-technical systems relevant to conventional CDS implementations. Yet, clinical trials of AI-enabled CDS will not provide sufficient information for implementation. Sociotechnical approaches are crucial to understand the complex personal, technical, work system, and broad societal implications of AI in clinical care. Proposed sociotechnical strategies include seeking clinician feedback from design through implementation and sustainment phases for AI-based approaches in CDS systems.

SCOPE

There's a notable gap in sociotechnical research focusing on the clinician's experience with AI systems, particularly regarding the integration of data visualizations to support their decision-making. Many sociotechnical studies of AI-based systems have focused on patient end users, but there is a need for additional clinician studies for CDS users that evaluate information synthesis needs, effective incorporating data summarization and visualizations, and assessments of the changes in the data and clinical environment that the CDS itself produces beyond the targeted action. Effective interventions that support better decision-making can be bolstered with design and evaluation that incorporates cognitive theories.as well as social theories that recognize healthcare data as one part of a complex organizational system. A commonly used One example from cognitive psychology is dual-process theory which categorizes thinking processes – that which are intuitive and fast (system 1) or deliberate and slow (system 2). Dual-process theory does not map well to neurological processes, for that, approaches related to cognitive control may be better. Social science theories may also have an important role to play in helping us to conceptualize modern healthcare environments as complex dynamic systems that are actively shaped by tool users, tool designers, and tools themselves. In addition to these complexities, it is critical that AI-assisted CDS also be considered, assessed, and continuously evaluated for fairness and equity within clinically and culturally relevant sub-groups that may have differential care goals and needs.

TOPICS

  • Healthcare AI implementation and sustainability
  • Approaches to distributed cognition
  • AI-driven early warning scores and acceptability
  • Individual differences and implications for AI implementation - attention and motivation
  • Sociotechnical & cognitive evaluation approaches to AI models
  • AI in context
  • Workflow for AI incorporation
  • Evaluating the sustainability of clinical use cases and organizational settings for specific AI-based CDS solutions
  • Trust in AI and sociotechnical design
 

PROGRAM

Time

Format

Topic

Speakers

8 - 8:05 am

Coffee and Settle

8:05 - 8:20 am

Podium Presentation

Welcome Address

Jorie M. Butler

8:20 - 8:50 am

Podium Presentation

Opening Keynote Implementation and Translation

Michael Matheny

8:50 - 9:05 am

Research Presentations

Implementation of a Risk Prediction Tool for Severe COVID-19 Outcomes to Improve Vaccination Rates among Veterans 

Evan Carey 

9:05 - 9:25 am

Research Presentations

Augmenting Healthcare: Examining the Impact of Surgeon-Robot Congruence on Clinical Effectiveness and Surgeon Learning

Christos Makridis

9:25 - 10 am

Interactive Panel

Q&A on Research Topics

Panelists: Michael Matheny, Evan Carey, Christos Makridis

Moderator: Peter Taber

10 - 10:15 am

Coffee Break

10:15 - 10:35 am

Research Presentations

Improved Interpretability without Performance Reduction in a Sepsis Prediction Risk Score 

Adam Kotter 

10:35 - 10:55 am

Research Presentations

Utilizing Large Language Models to Identify Reddit Users Considering Vaping Cessation for Digital Interventions  

Sai Vuruma 

10:55 - 11:20 am

Research Presentations

Framework to Support the Design and Development of AI Systems for Clinical Settings (20 min) 

Ayomede Owoyemi

11:20 - 11:50 am

Interactive Panel

Q&A on Research Topics

Panelists: Adam Kotter, Sai Vuruma, Ayomede Owoyemi

Moderator: Jorie M. Butler

11:50 - 12:15 pm

Discussant

Synthesizing Remarks Closing

Peter Taber

scientific paper PROGRAM Committee

Peter Taber, PhD
Research Assistant Professor, Biomedical Informatics University of Utah
Jorie Butler, PhD
Associate Professor, Biomedical Informatics University of Utah
Michael E. Matheny, MD, MS, MPH, FACMI
Professor, Department of Biomedical Informatics, Biostatistics, and Medicine Vanderbilt University Medical Center
Aref Smiley, PhD
Research Assistant Professor, Biomedical Informatics University of Utah