Workflow Integration of Artificial Intelligence in Clinical Practice

Artificial intelligence (AI) is increasingly being integrated into healthcare to support clinicians, reduce workloads, and improve workflow efficiency. AI is particularly beneficial in image-based and data-driven medical fields due to its pattern recognition capabilities. While extensive research has been done to explore the potential of AI under experimental conditions, the knowledge about the intricacies of its real-world implementation in clinical settings remains scarce. Consequently, a human factors approach that considers healthcare's complexity as a sociotechnical system is essential. <br> In this dissertation an examination of the AI integration into clinical workflows is presented, offering a comprehensive view of human-AI interaction in healthcare's complex environment. The theoretical background of this work is the System Engineering Initiative for Patient Safety Model, with the related Conceptual Model of Workflow Integration and the Technology Acceptance Model. Three research projects – one systematic review and two use cases – are presented in this dissertation. <br> The systematic review involved an assessment of AI's impact on efficiency, clinician outcomes, and workflows in medical imaging, revealing a positive effect on time for tasks and leading to the identification of different AI-augmented workflows. A novel framework was also introduced to categorize the level of AI implementation in the studies. The first considered use case involved an AI implementation in a radiology department, where the AI tool under study was not yet fully integrated into the routine workflow. While the users who participated in the study initially had a positive attitude, the poor fit and longer reading times for complex cases led to workarounds and frustration. The second considered use case, a fully implemented AI tool in human genetics, highlighted that usability and organizational factors were key to successful adoption, as most users incorporated the tool into their daily routines. <br> The work for this dissertation combined various study designs across different medical specialties. By identifying multiple variations of AI-facilitated clinical workflows, it was emphasized that the context of AI implementation is often unique and AI implementation requires local adaptation. Moreover, recommendations drawing upon identified facilitators and barriers are proposed which should be considered for safe and effective future implementation processes of AI in clinical care.