Mining tasks and task characteristics from electronic health record audit logs with unsupervised machine learning.

Area of Science:

• Health Informatics
• Clinical Workflow Analysis
• Human-Computer Interaction

Background:

• Clinician interaction with electronic health record (EHR) systems significantly impacts time spent and workload.
• Understanding clinician activities within EHRs is crucial for optimizing healthcare efficiency.

Purpose of the Study:

• To characterize clinician activities as distinct tasks within EHR systems.
• To develop novel, data-driven metrics for analyzing EHR task complexity and prevalence.
• To identify opportunities for optimizing EHR workflows.

Main Methods:

• Utilized unsupervised learning on EHR audit logs to identify and categorize clinician tasks.
• Developed metrics for event prevalence and repetition to define 4 task complexity profiles.
• Applied Mann-Whitney U tests to compare performance time, event types, and clinician prevalence across task complexities.
• Employed process mining and clinical annotations to validate identified tasks.

Main Results:

• Analyzed 57,234 EHR sessions from 33 nurses, identifying 81 distinct tasks.
• Found significant differences in performance time across the 4 task complexity profiles.
• Observed no significant differences in clinician prevalence or event modification frequency between task complexities.
• Presented expert-validated task workflows demonstrating clinical relevance.

Conclusions:

• EHR audit logs offer a valuable resource for investigating clinician activities.
• The developed metrics and task characterization can assist hospitals in optimizing EHR workflows.
• Further analysis can lead to improved clinician efficiency and reduced workload.