Identifying Pediatric Intensive Care Unit Bedside Teams and their Stability Using Electronic Health Record Audit Log Data

Poster Number: P186
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Mining, Healthcare Quality, Machine Learning, Information Retrieval, Workflow, Administrative Systems, Critical Care, Clinical Decision Support
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Inpatient care teams often consist of varying combinations of clinicians that are not explicitly documented and unique to patients, making them difficult to identify. This may be beneficial, but introduces risk of impaired communication and diminished trust among unfamiliar team members. Current methods of studying inpatient teams are too effort-intensive to develop actionable insights at scale.

We propose two methods for identifying and measuring the stability of patient-centric teams using EHR audit logs: (1)a rule-based, heuristic approach; (2)an unsupervised clustering approach.

We collected audit log data from 194 pediatric intensive care unit (PICU) patient encounters and 3121 clinicians from calendar year 2022 at a large academic children’s hospital. We used these data to identify clinician-teams directly involved in each patient's care each day of their encounter. For the rules-based approach, we identified clinicians with specific audit log actions indicative of bedside presence. For the clustering-based approach, we developed features that captured proportionality of audit log actions performed, embedded them to a 2D vector, performed clustering, and selected team members based on clusters. We computed stability scores for team compositions identified by these approaches.

We successfully identified team compositions that yielded stability scores aligned with clinical expectations. We demonstrated the feasibility of using audit logs to identify and characterize inpatient teams at scale. This methodology enables more complex team-related measures that lead to high/low-quality inpatient care to be developed and studied. Our next steps are to conduct a direct observation study to validate these measures against a gold standard.

Speaker(s):
Liem Nguyen, Undergraduate
Stanford University

Author(s):
Liem Nguyen, Undergraduate - Stanford University; Seunghwan Kim, MS - Washington University in St. Louis; Thomas Kannampallil, PhD - Washington University School of Medicine; Stefanie Sebok-Syer; Daniel Tawfik, MD, MS - Stanford University School of Medicine;