Engaging Hospital Staff to Identify Levers for Adoption of Clinical Decision Support: Protocol for a Single-Site Case Study Using System Dynamics Group Model Building.

Background: Clinical decision support (CDS) tools that provide patient-specific and evidence-based information to clinicians and care managers regarding patient risk for adverse outcomes have been a part of health care for decades. However, modern CDS, which consists of automated predictions based on complex machine learning models and hundreds of complex input variables, faces obstacles to adoption related to health care providers' perceptions of lack of transparency and utility. Often, the expertise of data scientists and clinical end users is not well integrated, creating implementation gaps from CDS development to adoption and ongoing implementation. Objective: This protocol describes the use of group model building (GMB) from the field of system dynamics to engage health system staff in identifying dynamic facilitators and barriers to implementing 1 class of CDS-early warning scores (EWSs)-in general medical-surgical wards. We aim to produce a causal model that reflects the insights and feedback shared during these sessions. We will also evaluate the GMB process as a potential strategy for CDS implementation and adoption more generally. Methods: The protocol consists of 3 sequential GMB sessions designed to elicit key variables for inclusion in the model, understand how changes in variable behavior over time affect adoption, and develop a causal loop diagram. Pre- and postsession questionnaires assess changes in perceived acceptability, appropriateness, and feasibility of the EWS and collect feedback on the GMB process. A stock-and-flow simulation model will be developed from the causal loop diagram to quantify how feedback loops influence variables over time and test assumptions. Results: The project was funded in 2022-2025, and 3 GMB sessions and qualitative causal loop diagrams were completed during that time. Data analysis is ongoing. This analysis consists of translating the causal loop diagram from the GMB into a stock-and-flow simulation model to quantify how feedback loops influence variables over time. Results will include a causal loop diagram accompanied by a detailed narrative that together tell a story about system behavior surrounding EWS adoption that is supported by session transcripts, the simulation model and test results, and the data on GMB participants' views about the EWS itself and the modeling process. Conclusions: These findings will have broader applicability beyond just EWSs. Future work will build on the EWS system dynamics model by incorporating multiple clinical use cases to fully capture multilevel factors that determine real-world adoption and sustainability of machine learning CDS.