SHM Converge Scientific Abstract ePoster Gallery-Assessing an Artificial Intelligence-Assisted Discharge Prediction Tool

Pdf Summary

This study evaluated the implementation and workflow impact of the 48-Hr Discharge Prediction Tool (48DPT), an AI-assisted tool designed to predict discharge readiness 48 hours in advance and help interdisciplinary teams begin discharge planning earlier.<br /><br />Researchers used two approaches at a large urban academic hospital: direct observation of interdisciplinary rounds on two medical units and semi-structured interviews with stakeholders including hospitalists, social workers, case managers, nurse managers, and unit medical directors. The interviews were guided by RE-AIM and CFIR implementation frameworks.<br /><br />Over 30 observation dates, 170 patients with positive 48DPT scores were followed. The tool was routinely used by case managers in 95.9% of cases, and discharge planning was discussed for 97.3% of these patients. However, when clinicians explicitly commented on the tool, they agreed with its positive prediction only 53% of the time.<br /><br />Interviews with 14 clinicians and team members showed limited reach and adoption among some staff, especially hospitalists and less experienced clinicians. Main barriers included low awareness of the tool, concerns about its accuracy, and limited transparency about how the algorithm worked. Another major issue was that many end-users were not involved in the tool’s development or implementation.<br /><br />Overall, the 48DPT was widely used operationally, especially by case management, but clinicians had limited engagement and only moderate confidence in its predictions. The study concluded that AI and machine learning tools are more likely to succeed when a broad range of users are involved early, when onboarding is repeated over time, and when ongoing user feedback is built into implementation.

Asset Subtitle

Tuyet-Trinh Truong

Meta Tag

Author List Andrew Dunn, Susanna Schmer, Tuyet-Trinh Truong

Category Research

Concept 48-Hour Discharge Prediction Tool

Concept Discharge Planning

Concept Clinician Acceptance

Concept Direct Observation

Concept Semi-Structured Interview

Distinguished Non-Finalist

Presenter Organization Icahn School of Medicine at Mount Sinai

Presenting Author Tuyet-Trinh Truong

Track Technology in Hospital Medicine

Keywords

48-Hr Discharge Prediction Tool

AI-assisted discharge planning

implementation workflow

interdisciplinary rounds

case managers

hospitalists

RE-AIM framework

CFIR framework

algorithm transparency

clinical adoption

48-Hour Discharge Prediction Tool

Discharge Planning

Clinician Acceptance

Direct Observation

Semi-Structured Interview