Developing an Approach for Integrating Multi-Stakeholder Causal Loop Diagrams: A Case Study of Agitation and Restraint Use in t

Thursday, July 16, 2026, 8:00 PM

Session: SOC Poster Session (Virtual 2) (Virtual)

Introduction/Problem Statement System dynamics (SD) thinking, including the development and presentation of causal loop diagrams (CLDs), offers an attractive methodology for identifying leverage points across complex systems. Not surprisingly, SD is therefore increasingly being used in public health research to identify such points of intervention to better address complex public health issues and promote positive health outcomes (Baugh Littlejohns et al., 2021; Uleman et al., 2024). Addressing these complex problems requires diverse perspectives from various stakeholders. Group model building (GMB) allows investigators to efficiently capture and visualize the mental models of these stakeholders as CLDs. However, where multiple GMB sessions are conducted across different stakeholder groups, limited methodological guidance currently exists on how to systematically combine the separate CLDs into a single integrated model. In response to this gap, Ryan et al. (2021) proposed several approaches for integrating CLDs, including triangulation, grounded theory, and synthesis. These approaches do not, however, provide the practical guidance investigators need to actually merge CLDs (Rajah & Kopainsky, 2025). Rajah & Kopainsky (2025) therefore recently proposed practical steps for integrating CLDs while emphasizing that the process should not rely only on combining visual diagrams, but should also consider the narratives shared by GMB participants to minimize unintentional reinterpretation of the CLDs by modelers during integration. This extended abstract describes an ongoing effort to integrate CLDs developed from separate stakeholder groups involved in a study investigating the complex dynamics of the emergency department (ED) at a major university hospital in Connecticut and their influence on patient agitation, workplace violence, restraint use, and staff burnout. While individual CLDs provide valuable insight into stakeholder-specific perspectives, combining the CLDs is necessary to better understand dynamics across the ED. Existing literature on CLD integration informed our approach, particularly the synthesis approach described by Ryan et al. (2021), the novel qualitative systems exploration model by Hulme et al. (2026), along with Rajah & Kopainsky’s (2025) emphasis on basing CLD integration decisions in participant discussions and narratives, rather than relying only on researcher interpretation of the diagrams themselves. Overall, this work aims to contribute practical guidance for transcript-informed integration of stakeholder-generated CLDs in participatory SD research. Background Several GMB sessions were conducted with different ED stakeholder groups, including attending physicians, resident physicians, nurses and technicians, security personnel, and patients with lived experience. Participants discussed factors related to patient agitation, workplace violence, restraint use, and staff burnout within the ED. Discussions were guided by several facilitators with SD and GMB expertise. Sessions were recorded and transcribed in Dedoose, a mixed-methods software program (Version 10.0.59, SocioCultural Research Consultants, LLC, Redondo Beach, CA, USA). Separate CLDs were developed for each stakeholder group to capture how each group understood the system and its underlying feedback structures. Preliminary CLDs produced during the GMB sessions were further refined by the research team upon review of the stakeholder-specific transcripts. A total of 8 CLDs were created. CLDs were initially developed in Vensim® PLE (Version 10.3.2, Ventana Systems, Inc., Harvard, MA, USA) and later transferred to Stella Architect (Version 4.1.1, isee systems, Inc., Lebanon, NH, USA) to take advantage of features that improved feedback loop identification and tagging. Current Status The attending physician and resident physician CLDs were first developed individually, including identification and preliminary naming of the most relevant and interpretable feedback loops within each CLD. All identified feedback loops from the attending physician CLD were then transferred into a new CLD. Feedback loops from the resident physician CLD were then added to this combined CLD. Some areas of overlap between the two stakeholder groups were straightforward to integrate due to similar variables and feedback structures. However, other loops required merging or slight revision of variables and causal relationships to better align overlapping concepts across the two CLDs. Throughout this process, we repeatedly referenced the original GMB transcripts to confirm that revisions and merged relationships remained consistent with the stories and experiences that participants described during the sessions, consistent with the recommendations from Rajah & Kopainsky (2025). An inductive coding process was then used to organize related feedback loops into broader thematic groupings. Preliminary loop names served as an initial guide, but loop names and themes were refined iteratively throughout the coding process. These thematic groupings were then visually distinguished within the combined CLD using color coding to improve interpretability and highlight related structures across the combined model (Figure 1). Notably, four similar themes emerged across both GMB sessions, including (1) ED burnout and turnover culture, (2) ED staffing constraints and resulting operational pressures, (3) patient agitation and restraint use in the ED, and (4) ED staff safety shaped by violence prevention and team dynamics. In an earlier iteration of the resident physician CLD, staff safety influenced by violence prevention and team dynamics were initially envisioned as two separate themes. However, attending physicians discussed these concepts in a more interconnected manner, frequently linking team dynamics, workplace violence prevention efforts, and perceptions of staff safety within the same feedback structures. As a result, these concepts were ultimately combined into a single integrated theme within the merged CLD. Additionally, a fifth theme related to trust and empathy as drivers of ED burnout and patient agitation emerged only within the resident physician CLD, reflecting residents’ recognition of interpersonal dynamics as important contributors to both patient and staff outcomes. Although this theme was only identified within the resident physician CLD, its related variables and relationships appeared across feedback loops present within the four other themes, becoming deeply interwoven throughout the integrated model (Figure 1), and ultimately supporting its retention as a distinct theme (Ryan et al., 2021). Next Steps Next steps include refining and integrating CLDs from the remaining stakeholder groups, specifically nurses, technicians, security personnel, and patients, using the methodology that we have developed. Following full integration, the combined CLD will likely undergo an iterative refinement process, adapting current guidance (Hulme et al., 2026; Rajah & Kopainsky, 2025; Ryan et al., 2021) to further simplify the model and identify the most important variables, feedback loops, and system themes contributing to patient agitation, workplace violence, restraint use, and staff burnout in this ED setting. Further, transparency in CLD development is much needed and for GMB sessions in particular, the process used to synthesize group input should be thoroughly discussed (Jalali & Beaulieu, 2024). This discussion should also transparently describe the efforts undertaken to integrate various CLDs, as necessary. As such, we also intend to further document and refine the CLD integration methodology described herein. By further refining and documenting this CLD integration process, this work will improve transparency and reproducibility in participatory SD studies involving multiple stakeholder-generated CLDs.