Integrating a Machine Learning System Into Clinical Workflows: Qualitative Study.

Area of Science:

• Clinical Informatics
• Artificial Intelligence in Medicine
• Healthcare Quality Improvement

Background:

• Machine learning (ML) models show promise for enhancing diagnostic accuracy in acute conditions.
• Successful integration of ML tools into routine clinical practice remains a challenge.
• Understanding clinician perspectives is crucial for effective ML implementation.

Purpose of the Study:

• To explore factors influencing the integration of a machine learning sepsis early warning system (Sepsis Watch) into emergency department workflows.
• To identify barriers and facilitators to the adoption of ML-based clinical decision support tools.

Main Methods:

• Semistructured interviews were conducted with 15 emergency department physicians and rapid response team nurses.
• A modified grounded theory approach was used to analyze qualitative data from interviews.
• The study focused on participants involved in the Sepsis Watch quality improvement initiative.

Main Results:

• Three dominant themes emerged: perceived utility and trust, implementation processes, and workforce considerations.
• Clinician trust in ML models was influenced by perceived accuracy and personal experience.
• Effective implementation was supported by user-friendly interfaces and clear communication strategies, while information flow and knowledge gaps posed barriers.

Conclusions:

• Frontline clinicians' perceptions of ML models are shaped by trust, utility, and implementation factors.
• Insights gained can guide future strategies for implementing ML interventions in clinical settings.
• Addressing knowledge gaps and improving information flow are key to maximizing adoption of ML tools.