While thousands of machine-learning models are developed and published every year, only a small fraction are implemented into clinical care, and even fewer are successful. Through a successful use case of implementation of a patient triage model for patients with dizziness in a tertiary care center, we discuss some common pitfalls and present a methodology to maximize the chances of a successful development and implementation of an AI model into clinical practice. We discuss how workflow analysis, appropriate accuracy metrics to the clinical problem, model deployment, change management and pilot design apply to the implementation of AI solutions into clinical practice.
Santiago Romero-Brufau, MD, PhD is Assistant Professor of Healthcare Systems Engineering and ENT at Mayo Clinic, where he leads the AI initiatives in the Department of ENT. He is also Adjunct Assistant Professor and a member of the Executive Committee for the Master's in Health Data Science at the Harvard T.H. Chan School of Public Health, where he teaches how to implement machine-learning models into the clinical workflow.