UNSTRUCTURED
While AI is making inroads into the clinical practice of medicine, its progress is impeded by a lack of interaction between diagnostic and prognostic predictive model developers and healthcare administrators, providers and staff. While developers judge the success of such models through performance measures of discrimination and calibration, healthcare workers are concerned with clinical significance. While the research literature abounds with papers describing all manners of model development, a paucity of guidance is available on how to implement such models after prioritizing those with the greatest chance of optimizing local clinical processes and outcomes. This manuscript describes a four-factor framework to estimate the potential for the successful implementation of a predictive model in a health system to aid health system leaders, providers in planning such interventions in the clinical workflow. This framework addresses the lack of interaction between model developers and users by planning for meaningful interactions through the implementation and evaluation process. As such, it addresses a critical problem at the planning table, setting the stage for future collaboration between the parties.