SHM Converge Scientific Abstract ePoster Gallery-Development of a Data-Driven Dashboard for Visualizing Barriers to Hospital Discharge

Pdf Summary

This project developed a real-time, data-driven dashboard to identify and visualize barriers to hospital discharge using electronic health record (EHR) data. The goal was to help care teams recognize actionable issues that prolong length of stay, increase costs, and strain hospital capacity.<br /><br />The team created an algorithmic, rules-based model using diagnoses, procedures, medications, nursing flowsheets, and keyword searches in clinical notes. Twenty-six discharge barriers were defined by Care Management Physician Advisors based on years of chart-review experience and published literature. Model performance was evaluated by comparing automated results with manual chart reviews of 1,200 patient encounters with an average length of stay of 20 days from May 2023 to May 2024.<br /><br />Overall model performance was strong, with accuracy of 0.92, sensitivity of 0.94, specificity of 0.92, positive predictive value (PPV) of 0.69, and negative predictive value (NPV) of 0.99. However, performance varied considerably by barrier, with sensitivity ranging from 0.94 to 0.42 and PPV from 0.69 to 0.18. These differences suggest some barriers are detected much more reliably than others.<br /><br />The discharge barriers were displayed in a Tableau dashboard, which is currently being piloted by care teams. Feedback from users is being incorporated into the next version. The main limitation is that the system cannot fully use all free-text clinical note content, which likely reduces PPV. Future work will explore integrating a large language model to improve interpretation of clinical notes and strengthen both overall and barrier-specific performance.<br /><br />Overall, the study demonstrates the feasibility of an EHR-driven dashboard for surfacing discharge barriers in real time.

Asset Subtitle

Carlton R. Moore

Meta Tag

Author List Anna Soloway, Carlton R. Moore, Christopher A. Caulfield, Escher Howard-Williams, Jarrod Ambrose, Jaydeep Lamba, Jeff Francki, John Downs, Leonardo Marucci, Patrick O'Shea, Robert W. Lampman, Udobi Campbell

Category Innovations

Concept Discharge Barrier

Concept Electronic Health Record

Concept Rules-Based Algorithm

Concept Medically Ready Patient

Concept Prolonged Hospitalization

Distinguished Finalist

Presenter Organization University of North Carolina School of Medicine

Presenting Author Carlton R. Moore

Track Transitions of Care

Keywords

real-time dashboard

hospital discharge

EHR data

discharge barriers

length of stay

care management

algorithmic model

clinical notes

Tableau dashboard

large language model

Discharge Barrier

Electronic Health Record

Rules-Based Algorithm

Medically Ready Patient

Prolonged Hospitalization