Using the PARAFAC2 tensor factorization on EHR audit data to understand PCP desktop work.

Area of Science:

• Health Informatics
• Computer Science
• Clinical Workflow Analysis

Background:

• Electronic Health Record (EHR) audit logs are crucial for privacy and security.
• These logs also offer potential insights into clinical desktop workflows.

Purpose of the Study:

• To determine if EHR audit log data can be processed to derive primary care provider (PCP) workflow measures.
• To investigate the utility of PARAFAC2 tensor factorization for identifying PCP tasks and workflow variations.

Main Methods:

• Analyzed 578,394 time-stamped records from 876 PCPs across 17,455 ambulatory care encounters.
• Applied PARAFAC2 tensor factorization to identify clusters of audit log records representing PCP tasks without ground-truth labels.
• Interpreted results using PARAFAC2 factors for task definitions and task frequency per encounter.

Main Results:

• PARAFAC2 automatically identified 4 common clinical encounter tasks: medications access, notes access, order entry access, and diagnosis modification.
• Discovered significant variation in task completion, including 9 distinct variants for notes access, explaining 77% of data variation.
• Mapped discovered variants to known workflows and identified distinct PCP user groups based on notes access methods.

Conclusions:

• EHR audit log data can be efficiently processed to generate higher-level features representing time-stamped PCP tasks.
• This approach enables objective measurement and understanding of clinical workflows from audit data.