Background

Group model building (GMB) is a participatory research method from the field of system dynamics that has provided new insight into the etiology of complex problems across different scientific domains [1]. In GMB, researchers facilitate specifically guided group discussion, with understanding of the problem studied, but without introducing their views on the problem [2, 3]. In these group discussions, that are conducted with topic experts (often key stakeholders of the problem), GMB captures their shared view on how the contributing factors of a complex problem interact by jointly constructing a model [2, 4,5,6]. This model summarizes the etiology of a complex problem. Traditional research methods do not provide this overview because they approach problems in a linear way, often identifying several interactions between factors without depicting their total coherence and impact [7]. GMB models also show how underlying mechanisms work and help to explore effective interventions [7]. Geriatric medicine is facing many complex problems, such as geriatric syndromes, dementia, emergency department (ED) visits, and the rehabilitation of patients with functional decline [8,9,10,11,12]. The complexity of these issues lies in the extensive interaction of many contributing factors from different scientific domains and the fact that these interactions are undefined [9, 10, 13,14,15]. GMB might help to address these complex problems.

The last decade has seen an increased application of GMB in healthcare, for example to address patient flow and workforce demand, and to understand syndromes like obesity, HIV/AIDS, and Alzheimer disease [9, 16,17,18]. Uleman et al. highlighted the potential of GMB in their comprehensible overview of Alzheimer disease and identified potentially important underlying mechanisms. However, despite these valuable results, studies have not investigated how to use GMB to address specific geriatric problems and examples of GMB in geriatric medicine are scarce [1, 9, 16, 17].

The COVID-19 pandemic has forced GMB studies to move online. This has advantages and disadvantages [19, 20]. Such as the advantage of logistics (planning and data collection in particular) and the disadvantage of more formal communication that potentially limits the sharing of ideas. Furthermore, facilitating online discussions requires a different set of skills and tools than in-person discussions do, but few studies have described these tools [19, 20]. A description of how to implement the GMB process online is needed. The implementation of GMB in the field of geriatric medicine has also not been well described [1, 9, 16, 17]. To address these gaps, this study aimed to describe the methodology of online GMB using a geriatric case study.

Methods

GMB compared with traditional research methods

GMB differs methodologically from traditional research methods in several ways, which make it useful for addressing complex problems. In comparison to in-depth interviews and the Delphi method, GMB involves group discussions, which give experts the opportunity to exchange views on the problem in person. This face-to-face interaction is essential for exploring a shared view together, especially if problems have an interdisciplinary character [5, 21]. Unlike focus groups, experts participate actively in forming a graphical depiction of the groups shared view and in doing so develop shared commitment [5, 21, 22]. In addition, GMB is facilitated by scripted activities, which optimize the use of different cognitive tasks, and step-by-step capture of the complexity of the problem in a model [3]. Main common features and differences between GMB and traditional research methods in researching the etiology of problems are described in Table 1.

Table 1 Main common features and differences between GMB and traditional research methods in researching the etiology of problems
Full size table

Types of models created by GMB

Models created through GMB are system dynamic models [7, 27]. System dynamics is a simulation modelling approach addressing complex issues in various application domains and has been used extensively in healthcare [1, 16,17,18]. Different system dynamics models can be constructed, such as a causal loop diagram (CLD) or a stock and flow model (SFM). A CLD is a model type that provides overview of a complex problems most important causative mechanisms and is often the first step in system dynamics modeling of a problem [7, 28]. It does so by capturing only the key contributing variables, relations, and underlying mechanisms (feedback loops) of the problem [7, 29, 30]. In a CLD, variables are connected by arrows to illustrate a causal relation. When a closed circle of variables connected by arrows is formed, an underlying mechanism arises, informing the reader of the CLD of a hidden enhancing or balancing effect as the result of variable interaction [7]. An example of a CLD and the insights it can provide is given in Fig. 1. A SFM translates the same variables and relations into ‘stocks’ (elements that accumulate over time and can be measured) and ‘flows’ (elements that change over time)[28]. It brings a CLD closer to a computerized model, but can provide less overview of a problem’s underlying mechanisms and make capturing a problem more complex [28]. Once a CLD or SFM is constructed, the model is quantified and validated via literature and data [31]. This article describes the construction of a CLD that has been described in a previous case study [32].

Fig. 1
figure 1

Example of a CLD on Alzheimer’s disease, from Uleman et al. [9]. This figure visualizes the role interactions between key causative factors can play in the etiology of a complex disease, such as Alzheimer’s. In this figure, physical activity and depressive symptoms, for example, are interconnected via feedback loop RD7. This feedback loop shows, that an increase in depressive symptoms will lead to a decrease of physical activity that can further increase depressive symptoms. In contrast to traditional research methods, the coherence of key causative factors and the feedback loops they are involved in tell us what effects need to be taken into account when exploring effective interventions. No changes to the figure were made. Copyright license: http://creativecommons.org/licenses/by/4.0/

Full size image

Case study

We conducted a case study between February and May 2021 to better understand why people older than 65 years of age visit the ED in Amsterdam at a population level without excluding any subcategories of older adults (see Appendix 1). The results of this case study have been published separately [32]. In the present study, we used an online GMB research design to capture the views of an interdisciplinary expert group on the interactions of most important contributing factors to these older adults’ ED visits as a population in a CLD. The CLD depicted the year 2019 to exclude the effects of the COVID-19 pandemic. These sessions were done online because of COVID-19 restrictions. OS and IB both designed as well as facilitated the GMB sessions, HW facilitated the GMB sessions and clarified output, and ER gave expert advice on design. Further information on the researchers’ backgrounds can be found in the section on authors’ information. All methods were carried out in accordance with relevant guidelines and regulations.

GMB protocol

Sessions, scripts, and adaptations

To design a GMB protocol that fitted to the case study, we consulted the literature [4, 5, 30, 33, 34], Scriptapedia (an open access online book containing guidelines and scripts for evidence-based GMB [35]), and our own expertise. The protocol was designed in February, March, and April 2021 and the sessions, goals, scripts, and preparation are summarized in Fig. 2. All sessions and scripts were adapted for online use and these adaptations are included in Table 2.

Fig. 2
figure 2

Overview of GMB design, including session themes, scripts, goals, and preparation

Full size image
Table 2 GMB preparation assignments and scripts, including name, description, goal, motivation for selection, and adaptation
Full size table

The GMB protocol was designed as four 1.5-hour sessions conducted a few days apart. This was done for several reasons. First, 1.5-hour sessions allowed the experts to focus on a limited number of activities and thereby on each step of the model build. Second, conducting sessions a few days apart allowed us to give the experts information and assignments to prepare for the sessions, thereby enhancing session performance. Third, time between sessions allowed experts to consolidate the knowledge they gained during sessions and to develop new perspectives [36]. Four, several shorter sessions made it easier for experts to fit sessions into their schedule. Last, time between sessions allowed us to process the session output and improve its visualization before the next session.

The sessions addressed the following themes: identifying key factors (session one), exploring interactions (session two), forming the CLD (session three), consolidating the CLD, and testing scenarios (session four).

Sessions were structured using GMB activities called scripts. Scripts were compiled on Scriptapedia and included 38 scripts [cited 1st of June 2022] [35]. The scripts were divided into three categories: established (n = 21), promising (n = 12), or under development (n = 5). These scripts have been validated by multiple independent teams and produce consistent results (26). Scripts were also divided by the type of cognitive tasks they entail; these were introductory/presentation (designed to educate or update experts), divergent (designed to produce different ideas and interpretations), convergent (designed to cluster and categorize ideas and interpretations), and concluding/evaluative (designed to rank and choose between options and ideas).

Only established scripts from Scriptapedia were selected for the sessions. These scripts included ‘hopes and fears’, ‘presenting the reference mode’, ‘variable elicitation’, ‘dots’, ‘initiating and elaborating a CLD’, ‘model review’, ‘action ideas’, and ‘next steps and closing’[35]. Additional scripts were designed by the researchers to enhance session performance. These included ‘entrance’, ‘welcome and introduction’, ‘introduction and recap’, and ‘closing’ and checked the experts’ understanding of the content, methodology, and medium (Zoom and Miro) as well as their perspectives on the CLD.

Adapting the GMB protocol to an online format included altering the organization, visualization, and communication of the scripts. We used a recent article on conducting GMB online as a guideline [19]. For communication, Zoom (video communication [37]), Miro (an online whiteboard [38]), and email were used. To maintain focus on the model under construction, experts were given ‘read-only access’ to Miro for reflection and were only allowed to speak after using the raise-hand function in Zoom (see Tables 2 and Appendix 2 for more details). To minimize the chance of potentially limiting the sharing of ideas by experts and bias in their response as a result of the online format, we aimed to create a safe, non-judgmental, informal setting (however with clear communication protocol) and gave every expert a chance to share ideas or interrupt using the raise hand function. Experts were given preparative information and assignments before each session to stimulate thoughts on the process and enhance session performance (see Table 2).

Roles and facilitation manuals

Sessions were led by three facilitators (OS, IB, HW, see the section on authors’ information for their background). To facilitate the scripted GMB process, researchers take on different roles [4]. A minimum of five essential roles are needed: facilitator, modeler/reflector, process coach, recorder, and gatekeeper [4]. In our study, OS (facilitator one) focused on group facilitation, knowledge elicitation, and initial drafts of the structure. IB (facilitator two) fulfilled the recorder role by collecting data and conceptualizing the system. OS and IB fulfilled the modeler/reflector role. HW (facilitator three) took on the roles of gate keeper and process coach. HW evaluated the group dynamics and helped to frame the problems discussed in the first session. In later sessions, HW reflected on group and facilitator team dynamics. We used facilitation manuals to design and implement the GMB process. These included an in-depth description of the sessions’ objectives, roles, scripts, and agendas [35] and are presented in Appendix 2.

Expert selection

We purposefully recruited experts based on a predetermined essential profiles list in order to achieve the research aim, account for optimal online discussion group size as suggested by ER and Wilkerson et al., as well as to reduce the risk of selection bias. In GMB, expert selection involves multiple experts that are central to a topic and experts are selected using different methods [6]. Expert profiles included key stakeholders of the older adults’ patient journey leading to ED visits in Amsterdam, who were seen as an expert by colleagues and had at least five years of job experience. Nine essential expert profiles were identified for our case study, including seven local health care professionals (district nurse, ED physician, general practitioner, geriatrician, geriatrics physician, nurse specialist geriatrics, nurse transfer coordinator), one patient representative and one healthcare insurance data analyst. Nine experts were recruited and formed a fixed participant group during the GMB study. Expert selection was described in detail previously [32]. A summary of their characteristics can be found in appendix 3.

Data collection, data analyses, and model validation

Data were collected by video recording the sessions and capturing expert discussion on the online whiteboard. Video records were transcribed verbatim, anonymized, and checked for accuracy. As in line with Scriptapedia guidelines [35], data were analyzed and validated as part of the scripts and between sessions. All output, analyses and validations were discussed with the expert group. More details are presented in Table 2. and Fig. 2. Validation of the model was previously described [32].

Results

Session implementation and adjustments

All sessions were held in April and May 2021 and produced a CLD that visualized the combined expert view on why older people visit the ED in Amsterdam. Sessions one to four lasted 1.5 h. Sessions were adjusted because experts found it challenging to define unambiguous contributing factors. Two non-scripted sessions were added to clarify the CLD and establish consensus on a final version. The final protocol is shown in Fig. 3 and all script adjustments are presented in Table 3. The experts’ reflections on implementation are summarized in Appendix 4.

Fig. 3
figure 3

Overview of GMB implementation, including session themes, scripts, goals, and preparation

Full size image
Table 3 Adjusted GMB preparation assignments and scripts
Full size table

Three adjustments were made to help experts better define contributing factors. First, all facilitators used their own expertise to paraphrase factors in an unambiguous way during the sessions and these paraphrases were approved by the experts. These discussions resulted in unambiguous definitions for most factors. Second, the modelers categorized the suggested factors using the Miro whiteboard and these categories were approved by the experts. This facilitated group discussion on ambiguity and contributed to unambiguous definitions for several factors. It also helped the researchers to organize the CLD in a clear way. Third, the facilitators planned iterative plenary reflection on suggested factors at the beginning of each session, which also sharpened the definitions.

Extra time was scheduled for scripts that defined contributing factors, including ‘initiating and elaborating on a CLD’ and ‘model review’. Time was saved by processing some of the suggestions made by experts in the ‘initiating and elaborating a CLD’ script after the session.

A fifth non-scripted facultative session was added to discuss the clarity of the CLD. In GMB, CLD clarity is often checked by the modelers [1, 16, 35]. We invited the experts to join the check because defining contributing factors was challenging. The ED physician, geriatric nurse, and district nurse participated. This clarified the three factors and added two relations.

A sixth non-scripted mandatory session was also added to establish consensus among the experts on the final CLD. This final presentation of the CLD is often not considered part of the GMB [1, 16, 35]. Because definitions were challenging, a mandatory session was organized with all experts to check if the CLD clearly depicted the experts’ views. Experts established consensus on the CLD formed in session five (see Fig. 3 and Appendix 2 for more details).

Discussion

This study describes a methodological GMB approach in geriatric medicine. It was challenging to implement four qualitative online GMB sessions on why older adults visit the ED. We adjusted the protocol to help the experts better define contributing factors. These adjustments included reserving extra time for discussion, paraphrasing definitions, categorizing definitions, and reflecting with experts on suggested factor definitions. Communication was promoted by giving every expert the chance to speak combined with a clear communication protocol. Six sessions were held altogether, which resulted in a clear CLD.

To the best of our knowledge, no other study has reported difficulty using GMB methodology to define factors contributing to why older adults visit the ED. We believe that the challenges we observed were caused by the characteristics of the ED visits, such as frailty, which are poorly defined, hard to measure, and the result of multiple contributing factors [12, 39, 40]. The CLD formed in this study underlines this hypothesis [32]. Traditional GMB methodology [4, 5, 30, 33,34,35] did not provide the tools to unambiguously define these characteristics so we developed alternative ways to use the methodology. ED visits by older adults have similar characteristics to many problems in geriatric medicine, such as geriatric syndromes and the rehabilitation potential of patients with functional decline [8, 11]. Therefore, defining contributing factors for these many problems is expected to be challenging.

Paraphrasing, categorizing, iterative plenary reflection and reserving extra time for defining contributing factors helped our experts to better define these factors. These tools clarify and organize the conceptualization of contributing factors [36, 41, 42]. In line with these findings, the Scriptapedia guidelines describe paraphrasing as an important tool in the GMB process, but the skills required for paraphrasing are not described [35]. In this study, we found that a good clinical background in geriatric medicine was essential for paraphrasing definitions effectively. To our knowledge, categorization has not been reported as a tool to help experts define contributing factors in GMB studies. We found that extensive knowledge on the problem under discussion was essential for effective categorization. Reserving extra time and plenary reflection have often been used in GMB, but not for defining contributing factors. In summary, we advise using these four tools when applying GMB in geriatric medicine and having at least one researcher on the team with a clinical background in geriatrics and extensive knowledge on the problem being studied.

Scripts have to be adapted for online use. Wilkerson et al. have produced practical guidelines for adjusting scripts to an online format and we followed these guidelines when designing the present study [19]. We made a few additions to Wilkerson’s guidelines, such as giving every expert a chance to speak. This, combined with a clear communication protocol, improved the discussions and we recommend these measures in future online GMB studies.

A strength of this study is the detail of process description. The rationale of GMB studies is often limited to script adaptation [1], which means valuable insights into effective implementation are missed. An additional strength is that this study was conducted online effectively and therefore represents an example of the logistical benefits of online GMB.

Limitations

A shortcoming of this study is that no GMB methodology expert facilitated the sessions. We addressed this by consulting a methodology expert (ER) in advance about the study design and by using our own expertise. Furthermore, paraphrasing and categorizing the definitions of contributing factors may have introduced bias. We minimized this risk by asking the experts if they approved these changes. Finally, by including only one expert from a different scientific discipline (healthcare insurance data analyst/economics) we may have introduced a bio-medical bias on response. However, healthcare professionals working in geriatrics, diagnose and treat psychosocial problems every day as a part of their profession and see the effects of laws as well as financial problems contribute to older persons’ ED visits. Furthermore, these problems were extensively visualized in the CLD.

This study contributes to both research and clinical practice by offering an example of how online GMB can be used to better understand complex problems in geriatrics. This example can help both clinician, researcher as policy maker to use GMB for addressing complex geriatric problems. These complex problems can be biopsychosocial, organizational or intertwined. In order to secure a holistic approach to these problems, representativeness of different scientific disciplines, such as sociology, anthropology, psychology, economics, mathematics, law, philosophy etc., in GMB expert selection should be taken into consideration. More online GMB studies are needed in geriatric medicine to validate the challenges and possible solutions we have identified in this methodological approach.

Conclusions

In sum, we have described the methodological approach for applying GMB to unravel complex problems in geriatric medicine. We also tested alternative ways of using the methodology to help experts overcome the challenge of defining contributing factors in a geriatric case study. These ways included paraphrasing and categorizing the definitions, offering plenary reflection, and reserving extra time for defining contributing factors. Giving every expert the chance to speak combined with a clear communication protocol also promoted orderly communication. Since the characteristics of this geriatric case study are similar to many geriatric problems, the insights from this study may improve the application of GMB in geriatric medicine.