Transparent Decision Support in ever-changing healthcare contexts

Abstract

Every day physicians make choices on clinical treatment that directly influence patients’ well-being. To deal with these critical decisions and avoid treatment errors and costs, physicians show a growing interest in Clinical Decision Support Systems (IDSS). Current CDSSs, however, suffer from limited transparency and flexibility, which gives rise to ethical concerns when applied in healthcare environments. A novel CDSS approach overcomes these limitations: Behavioral Artificial Intelligence (BAIT). However, a BAIT-based CDSS is not yet capable of incorporating new clinical developments to keep its recommendations accurate over time. Because healthcare decision-making is highly dynamic, the static BAIT-based CDSS needs a transformation into a dynamic BAIT-based CDSS that retains its accuracy and, therefore, clinical relevance over time. However, the preferences regarding a dynamic BAIT-based CDSS vary among healthcare decision-making contexts. Therefore, CDSS developers need an architecture that illustrates how to create and customize a dynamic BAIT-based CDSS that matches physicians' preferences in a particular healthcare context. Because designing these architectures is challenging, this research aims to formulate design principles that guide the design of dynamic BAIT-based CDSS architectures. By following the Action Design Research (ADR) framework, this research identified and tested a set of architecture requirements by building an architecture in a situated problem context. The generalization of the requirements that worked in a situated context resulted in ten design principles guiding the design of a dynamic BAIT-based CDSS architecture. The greater part of the design principles is specific to the design of a dynamic BAIT-based CDSS architecture. By providing these novel insights, the design principles contribute to the CDSS architecture design knowledge base. The research also contributes to CDSS architecture design knowledge because it tackles the challenges of designing a dynamic BAIT-based CDSS architecture. By doing so, the research outcomes eliminate the barriers to design such an architecture and lay a foundation to continue the work on transparent and dynamic CDSSs. The findings highlight new design challenges for further research, like designing a dynamic BAIT-based CDSS architecture with the ten design principles in another sector to investigate how to modify the design principles, so the principles are useful outside the healthcare sector.