Classifying clinical work settings using EHR audit logs: a machine learning approach.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Clinical Workflow Analysis

Background:

• Anesthesiologists often work across diverse clinical settings, including surgical intensive care units (ICUs) and operating rooms.
• Accurately classifying these work settings is crucial for understanding physician activities and workload.
• Electronic health record (EHR)-based audit logs contain rich data on clinical activities.

Purpose of the Study:

• To classify the work settings of anesthesiology physicians using EHR-based raw audit logs.
• To evaluate the performance of supervised machine learning classifiers for this task.
• To identify specific clinical activities indicative of working in a surgical ICU.

Main Methods:

• An observational study included 24 attending anesthesiologists working in surgical ICUs in 2019.
• Time-stamped EHR audit log events were processed into a term frequency-inverse document frequency matrix.
• Multiple supervised machine learning classifiers were trained and evaluated to predict physician work settings.

Main Results:

• A random forest classifier achieved high discriminative performance (AUC-ROC: 0.88; AUC-PR: 0.72).
• Physicians performed a median of 2545 EHR actions weekly and worked a median of 5 weeks in surgical ICUs.
• Activities like signing clinical notes and updating diagnoses predicted surgical ICU work.

Conclusions:

• A random forest classifier accurately predicted anesthesiologist work settings based on EHR audit log data.
• This approach enables automated assessment of clinician activities and work contexts.
• Findings support using audit logs for workload assessment in different clinical environments.