Insights Gained From a Re-analysis of Five Improvement Cases in Healthcare Integrating System Dynamics Into Action Research

Background

Healthcare is complex and is characterized by a variety of patients and care pathways as well as multi-professional staff serving these pathways. Time pressure and minimal margins for errors make it challenging to implement new policies or procedures, no matter how desirable they are. ^1-3 May et al ⁴ suggest using questionnaires to assess organisational preparedness for top-down implementation, whereas Øvretveit et al ⁵ and Bååthe and Norbäck ⁶ describe the futility in trying to define context-independent principles of implementation, by referring to the observation that any solution in healthcare needs to be calibrated to local needs and circumstances. The tension between the hierarchical organisation and the professions as described by Mintzberg ⁷ may lead to mistrust and powerplay, ⁸ further complicating change efforts.

System dynamics (SD) offers methods, techniques, and tools that are intended to bring about a shared understanding of reality and its systemic implications. An SD-based modelling process usually encompasses: (1) problem articulation, (2) formulation of a dynamic hypothesis, (3) formulation of a simulation model, (4) model testing, and (5) policy design and evaluation. ⁹ SD requires expertise in the method, as well as in the use of associated software tools. SD simulations allow for the rapid development, testing, and reality-checking of scenarios for coherence and consistency in silico before implementing solutions in real life. Group model building (GMB) is often used in SD. Rouwette ¹⁰ describes how GMB leads to the convergence of mental models (ie, minds) through mutual information exchange between participants in GMB sessions. This convergence is also described as stakeholder alignment (to consolidate potentially competitive perspectives). ¹¹ Voinov et al ¹² further describe tools and methods that can be used at different stages of GMB and Holmström and Elf have scoped group interventions that may be suitable to use with GMB. ¹³ GMB is usually carried out by using prepared sequenced steps in workshops, so called scripted GMB, where participants often are expected to learn the basics of model building and SD terminology. ^14-17 This can pose a challenge for participants, but on the other hand, scripted GMB enables SD modellers less experienced in facilitation to lead workshops.

Action research (AR) is a field of research that is based on actively engaging participants that are willing to share their own perspective on a problem, to collaborate so as to find a mutually acceptable solution to that problem and then to, pragmatically, investigate how this solution actually works in practice. ¹⁸ The process of AR is usually carried out in cycles of Plan, Do, Study, Act, where participants take informed decisions about how to proceed based on collected data and experience from a previous cycle. ¹⁹ In healthcare, AR projects are typically specific to the work context of the participants, focus on their perceived problems, and is oriented towards developing their workplace here and now. People working together meet in order to identify and address a problem and to generate and assess alternative improvement; they iteratively reflect on their current practices, identify possible actions, and then try them out in order to learn from reality. The AR process also leads to shared mental models, but in contrast to SD, these remain as mindsets and are not necessarily explicitly stated in formal models. Consequently, solutions are derived and assessed in reality, which often makes an AR process time consuming. AR is based on deep participant engagement and a clear sense of ownership of the problems as well as their resolutions. “Implementation” is a concept rarely used in the AR literature. Rather AR relies on shared energy in the group that leads to actualization. If the results obtained are not as expected, then the group asks “what have we missed” and works through yet another iteration to find a more functional solution to try out, – a process not unlike the reality checks applied in SD. The process continues until a satisfactory result is reached.

Mingers and Gill ²⁰ suggest that no single methodology can offer a complete view of the complexities facing organisations and their members and suggest that the application of two or more methodologies are more likely to produce a realistic representation of the challenges facing them, which in turn, leads to better decisions. Hesse-Biber ²¹ suggests that mixed methods forges new pathways and provides innovation. Zolfagharian et al ²² found that there is little knowledge about why, when, and how SD is combined with any particular method. Moreover, Howick and Ackermann ²³ similarly conclude that there appears to be little discussion about generic lessons from mixing methods in practice. Since simulation-based analysis allows for timely, in silico testing, and rejection or acceptance of proposed solutions, time may be saved for such an approach in comparison to a process whereby the assessment of solutions takes place in reality such as in pure AR. Keeping in mind, however, that actualization rates of SD interventions have been reported to be around 5%, in spite of the fact that the interventions proposed appear to yield beneficial results in silico, ^24,25 identifying strategies which can increase chances of actualizing identified solutions are needed. Scholl ²⁶ describes how both SD and AR have iterative processes and suggests that what participants learn through SD simulations can be applied in the reality and what they learn in the real world can improve the virtual reality of the simulations. Scholl proposes that SD can facilitate every phase of AR. Little has, however, been reported about experiences from integrating any simulation method into an AR process. ^27-29

Here we analyse five improvement cases in healthcare where SD was integrated into AR processes to support multi-professional groups in finding solutions to work-related challenges. The aim of our study was to explore how this particular integration adapted to each case over time and to identify related work patterns and principles which can provide methodological guidance for practitioners and improve chances of actualisation in the client organization.

Materials and Methods

Materials – the Studied Cases

Five improvement cases in healthcare, carried out over seven years (2004-2011) were studied. A summary of case characteristics is given below, and a detailed overview of each case is given in Supplementary file 1. All cases were selected from the case experience of the first author (the main facilitator and modeller with extensive case experience, from pure AR, from pure SD, as well as from combined approaches) and were included in the re-analysis based on two selection criteria: (1) having addressed change and improvement processes in healthcare including multi-professional groups, (2) having been executed using SD integrated into AR as a choice of method to suggest and study potentially actionable solutions to problems/objectives posed (Figure 1).

Close

Figure 1. Schematic of Possible Degrees of Mix Between Action Research and System Dynamics From a Mixed-Methods Perspective. In contrast to the scientific literature about AR, GMB (where AR is integrated into SD), and SD, integrating SD into AR has to date been little studied. Abbreviations: AR, action research; SD, system dynamics; GMB, group model building.

All cases originated from Swedish healthcare, national population of 10.4 million inhabitants, where 21 regional councils and 290 municipalities manage and provide healthcare. Municipalities are mainly responsible for elderly care and regional councils are responsible for primary care as well as special care. Case 2 was carried out in a small municipality, the other cases in medium to large regional hospitals, catchment areas between 260 000 and 2 000 000 inhabitants. Each of the five cases was carried out within a time frame of 4-6 months with 4-5 group meetings and planning and modelling work in-between. Implementation of solutions in reality was not within the initial scope of any of the cases. The studied groups were composed of members of different professions in healthcare.

Case 1 concerned a stroke ward, where there was a perceived need of additional patient beds and a wish to determine qualitative factors for improving patient survival and health status when their medical treatment was completed. Case 2 was an obstetrics department, where staff and patients were dissatisfied with current scheduling practices. New work practices had been identified that potentially could solve the issues. However, there was strong uncertainty about how to apply those principles without worsening the situation. Case 3 was a dementia care home that was preparing a reorganisation for patient-centred care as well as planning for the adaptation of the premises to modern practices of dementia care. Case 4 involved a paediatrics department where premises were too cramped in periods with high levels of infectious diseases and were expected to be insufficient with respect to number of patients due to the closure of a satellite unit and increased child population. Case 5 concerned an accident and emergencies department with premises crowded by patient flow peaks several times per week to levels where staff perceived loss of full control and risks for patient safety. The department wanted to review work practices prior to planning for new or rebuilt premises.

Method – Case Re-analysis

The re-analysis of the cases was conducted by researchers from several disciplines, all providing different and complementing perspectives. The first author was accompanied by two professors (SD and clinical sciences, respectively), one associate professor (clinical sciences, experienced in information sciences in healthcare) and one PhD (health and care sciences, experienced in AR in healthcare). In the following, we define a work pattern as how steps in a process may be sequenced or related to each other, whereas a work principle is valid throughout an entire process.

Data Processing

All field notes, e-mails and models were revisited and summarized through four major iterations of analysis, using displays according to the methodologies suggested by Miles, Huberman and Saldana ³⁰ and by Eisenhardt. ³¹ The analytical process is described in Supplementary file 2. The approach was abductive, where each iteration began with a provisional adoption of an explanatory hypothesis ³² Each analysis was first carried out individually by each co-author, then in group discussions, comparing and contrasting individual observations. This was repeated until the authors arrived at condensed and coherent descriptions of the material.

Project Stages

To understand the development of each case, we used a modification of Kubr’s five steps and James and colleagues’ five approaches to consultancy assignments and client projects. Kubr ³³ describes the scope of consultancy assignments as consisting of five stages: entry, diagnosis, action planning, implementation, and termination. James et al ³⁴ takes a skills-based approach and describes five approaches to the client project: statistical analysis, modelling of key variables, problem identification, implementation of solutions or to be a sounding board, and to select a consultant with matching skills. We analysed the cases by five stages: diagnosis of problems, analysis of facts, modelling of key variables, action planning, and implementation. We excluded Kubr’s entry and termination as not being part of the processes investigated here and James and colleagues’ sounding board as being a different type of consultancy work. We also carried out bottom-up analysis of each case to uncover and describe workflows by case and define generalized workflow steps. In addition, the implementation stage was structured using Brailsford’s ²⁴ three levels of implementation of simulations. These are categorized in suggested (theoretically proposed by the modellers), conceptualised (discussed with a client organisation), and implemented (actually used in practice). This helped to quantify the degree of implementation and identify potential barriers to immediate implementation.

System Dynamics, Action Research, and Mixed-Methods Perspectives

To understand the overall mixed-method process, the cases were analysed through an AR perspective, using Rowbottom’s ³⁵ four questions regarding what is known, assumed, extant, and requisite about the work situation at each stage of the process, supplemented by corresponding interpretations in the SD context to further understand how the cases moved forward (Supplementary file 3). The framework is an enrichment of Rowbottom’s questions and provides an AR-based structure to study the interaction between AR and SD when comparing the five cases and drawing conclusions regarding general work patterns and principles. The share of time between facilitation and modeling work was estimated in each case.

Results

Work Patterns

Adaptive Workflows

The workflows by case are presented chronologically in Figure 2. The headings in the figure reflects work carried out either in a group, by the modeller, or by both. All cases had the same beginning and end, but had different sequences of intermediate steps.

Close

Figure 2. Chronological Workflows by Case. All cases began with an extensive facilitated problems and objectives inventory, starting with individual reflection and notes, then displayed to the group. The processes were purposively exhaustive so that all participants were satisfied that everything had been listed. Intermediary steps were sequenced differently between cases, based on pragmatic judgements regarding what would best move the process forwards. At all steps, facilitation was carried out to ensure that participants owned the issues, not the modeller. All cases ended in a workshop with an experimental part where a multitude of solutions were tested in silico.

Development of Solutions by Case

Figure 3 shows the modeller’s work by case using the modification of Kubr ³³ and James et al. ³⁴ In all cases, the groups worked through a diagnosis of problems, an analysis of facts and modelling of key variables and their interaction (ie, the systems structure). Differences between cases were primarily found in the parts relating to analysis of facts and modelling of key variables.

Close

Figure 3. The Modeller’s Participation by Project Stage According to the Modification of Kubr and James et al and the Identified Workflow Steps. Note that participation in actual implementation was not included in the scope of any of the cases. However, in Case 2, the modellers participated in the introduction of the solution to the staff prior to the implementation and were invited to the final evaluation.

Divergent and Convergent Phases Towards a Multi-professional Knowledge Repository

All cases were characterized by the same kind of oscillating pattern in the group processes. The processes were divergent and creative when actively stimulating the raising and listing of a variety of problems or suggested solutions. Convergent, when trying to reach consensus, prioritizing and agreeing on what to focus on. All groups went through two major phases of divergence and convergence (Figure 4).

Close

Figure 4. The Two Divergent and Convergent Phases. The starting point of all processes was an initial problems statement, which in a first phase was extended in a divergent and creative process ensuring that all issues were under discussion, and which then was converged by prioritization into a set of mutually agreed problems. The second phase began with divergent/creative work, proposing a variety of solutions, which were tested in the simulation, and which then was converged into a set of mutually agreed solutions to potentially be implemented in reality.

During the first divergent phase, problems and objectives were listed until “everything was on the table” and mutually agreed upon. Instead of participants acting as stakeholders, they were encouraged to work as if they were laying out a jigsaw puzzle, exploring how each person’s knowledge and skills complemented that of the others so as to effectively address the problem and context at hand. Over time, this process resulted in their multiprofessional knowledge repository, a shared point of reference, which led to deeper individual and collective understanding of what went on in the system associated with the problem they studied.

During the second divergent phase, suggestions of possible solutions and policies were encouraged. The simulation model and user interface were adapted to allow for the testing of a multitude of suggestions and clarified what could work in reality. During this phase, the group again converged, this time on a mutually agreed distilled number of solutions or policies which formed their proposal for how to continue. Thus, their multiprofessional knowledge repository became consolidated.

Work Principles

Studying the Interaction of Integrating System Dynamics Into Action Research

The case-specific implications of Rowbottom’s four questions for SD are presented in Table. Each case is positioned in the context of its initial objectives and final results. The original questions are interwoven with process steps, insights, and model phases and end up in descriptions of resulting actions, levels of implementation, and identified barriers to implementation. As noted above, all cases had similar beginnings and ends but had different sequencing of intermediary steps as the processes adapted to the needs of the participants. The table shows how each case unfolded over time (vertically) as well as allows for cross-comparison of cases by stage to see similarities and differences between them (horizontally).

The initial problem statement was dismissed in Case 1 and an uncertainty among members of the group arose. This could not be addressed effectively as neither the manager nor the medically responsible physician were present during meetings. As a consequence, discussions focused on how to stem the inflow of non-stroke patients. In Case 2, the work in the group led to the development of potential solutions and a test in reality, after which some modifications were made in preparation for the final deployment of the solution in which the modellers were asked to take part in. In Case 3, the group process and the causal loop diagram and the simulations laid the foundation for continued action planning and contributed, in part, to the subsequent reorganisation of work. In both Cases 4 and 5, the key issue was cramped facilities, which required a more extensive investigation and had to be addressed in an investment budget process outside the scope of the initial objectives. The models indicated that a possible short-term solution might be found, in Case 4, by spreading patients evenly across the available space and, in Case 5, by a better matching of staff levels to foreseeable patterns of patient arrival times.

In all five cases, the starting points were based on issues or problems where stated work principles did not work as intended. All cases went through an AR-based inventory of problems and objectives. Initial fact-finding was carried out and an initial SD model with narrow scope was built and/or a causal loop diagram was created based on the problem statements. This led to a first overview of the interconnectedness of the issues at hand. In two of five cases, the model did not replicate reality as perceived by the participants prior to this phase. In the next phase, the participants brought in their assumptions in the form of their respective knowledge and experience about their work situation and what was missing in the models. The models were finalized and became extant as they replicated the realities perceived by the participants. In all improvement cases there were forms of aha-moments when the simulations clearly pinpointed problematic situations and their causes. Once the root issues and causes were present in the model, then insights were shared and perceived as clear to all, the participants moved to suggesting solutions. Proposals were tested in silico in all cases, discarding those that were deemed to have none or little effect finally arriving at requisite proposals of solutions.

The Facilitating and Modelling Perspectives

Almost all meeting time was spent in facilitated group discussions and, in all cases, most of the modelling work was conducted between meetings (Supplementary file 4). There was engagement, interaction, and exploratory learning activities throughout each case. Meetings began by the facilitator asking the participants for reflections since the last meeting. Typically, the participants reported that they, based on thoughts after the previous meeting, had seen their work in the light of a somewhat new perspective. Likewise, they were also asked to report on their reflections at the end of each meeting, leading up to ideas about what to observe before the next meeting. Facilitation ensured that the voice of each group member, and profession, was heard at all meetings in that the participants were repeatedly being encouraged to talk in turn. Although it was not made explicit to the participants, AR and SD were combined throughout the entire processes. Initially, the participants were hardly introduced to SD because the focus was on a group process that contributed to the uncovering of problems and to the development of a shared view on the current state of affairs (cf. Rowbottom: manifest and assumed). However, already during the initial steps, the outline of a model was considered by the facilitator/modeller – including which variables to include and what data that were available.

As an overall work principle, the process moves from left to right in Figure 1. As each project progressed, the model and simulations became explicit and took more of the meeting time. During the experimental phase and final workshop, the participants were fully absorbed by the model and simulations, but the ongoing group process was still considered central and was focused by facilitation in line with AR principles. In none of the cases did the participants have to learn the basics of model building or SD terminology. However, they understood sufficiently to see the simulation results as credible and useful.

Group Composition

Group members of each respective case shared the same workplace (Cases 2-5), but four of twelve participants in Case 1 came from an external department (facilities management) and predominantly took part as observers. In all workplaces, patients were handed over from one profession to the next, a process that required coordination between staff members rather than collaboration as the professions had different roles and accountabilities. However, in Case 5, the staff interacted more extensively to the point of collaboration.

The group sizes in Cases 2-5 were close to the ideal 6-7 participants as shown by Sjölund. ³⁶ Those groups also had Coalesced Authority, Power, and Influence (CAPI) required for decision-making. ³⁷ They had high levels of engagement and rapidly moved forward through the process. However, the group in Case 1 was large, had non-active participants, and did not have CAPI as neither the manager nor the medically responsible physician was present. This resulted in the process being impeded.

Discussion

The aim of this study was to explore and identify methodological aspects of how the integration of SD into AR supported multi-professional groups in healthcare to discover workable solutions to work-related challenges. We found that all of the cases studied began with an extensive AR-inspired inventory of problems/objectives and ended with an SD-facilitated experimental phase where mutually agreed upon solutions were tested in silico. The sequence of intermediate steps and activities was case specific, but the majority of cases included the development of causal loop diagrams and graphical user interfaces to the respective SD model. The time was predominantly divided between discussions/reflections during meetings and modeler/modelling work between meetings. There was an overall pattern of continuous oscillation between divergence and convergence, but in two major phases, as each group moved towards a mutually developed point of reference, a consolidated case-specific multi-professional knowledge repository. Work principles ensured that the voice of each participant was heard, inspired engagement, interaction, and exploratory mutual learning activities among both participants and modeller/facilitator.

Divergent and Convergent Phases

Guilford ³⁸ defines divergent thinking as creatively going off in different directions and leading to a diversity of answers, where more than one may be acceptable, whereas convergent thinking narrows down the number of answers. We found that the studied cases went through two major divergent/convergent phases (Figure 4). The first phase focused on generating a comprehensive listing of potential problems connected with the issues at hand. This led to participants understanding problems as seen by others and how all issues fitted together. Divergent phases need to be exhaustive in order to avoid premature convergence, ³⁹ which otherwise could lead to sub-optimal solutions or low acceptance among participants or in the client organization. Fact-finding was carried out as needed by the identified issues. During the second phase, ideas and solutions were brainstormed. All suggestions could immediately be tested in silico and evaluated as potential solutions. Finally, the investigated mixed-methods approach led to acceptance as each participant saw how their work fitted in with others and gave a more comprehensive understanding of their system. These findings are in line with results presented by Parnes. ⁴⁰ He summarizes problem solving in five stages, each divergent and convergent: (1) fact-finding, (2) problem-finding, (3) idea-finding, (4) solution-finding, and (5) acceptance finding, also emphasizing the aspects of creative divergent brainstorming activities and convergent focusing to this problem-solving model.

Multi-professional Knowledge Repository

In this study, we found that the participants brought their reality into the process, used their insights from the meetings to observe and re-assess their realities between meetings, and, in turn, brought their reflections into the next meeting. This “feedback” improved both the insights of the group altogether as well as quality of the simulator under development, and it contributed to the identification of actionable solutions that were tested in the virtual reality of the simulation, as suggested by Scholl. ²⁶ At the same time, the work principles applied assured that the overall group process was nudged towards satisfactory outcomes. The simulations had a catalytic role to play in the overall process and represented an agreed-upon comprehensive repository of knowledge. This study extends the results of an earlier in-depth study of Case 2, ²⁸ which concluded that the participants were not stakeholders in the sense that they held firmly competing opinions and felt compelled to persuade each other. ¹⁰ Instead, it was a co-creative process, where participants explored how each person’s set of knowledge was aligned with that of the others and did so in an experiential ⁴¹ and reciprocal ²⁶ learning process that resulted in the accumulation of the here reused concept of a multi-professional knowledge repository. They thus became shapeholders ⁴² rather than stakeholders.

Integrating

Studying the Interaction of Integrating System

In contrast to scripted GMB, we found that the process of integrating SD into AR in a more fundamental way addressed the concerns of participants and laid the foundation for actualisation. We found that Scholl’s ²⁶ suggestion of reciprocal learning, ie, convergence, between reality and the simulation model contributed to the investigated work process. Both AR and SD influence how people think. SD provided a form of gaming or role-playing platform where experiential learning took place. One of our key observations is that a group process rooted in AR leads to shared mindsets and to the realization that it is the mindsets, rather than a top-down implementation project in itself, that increase chances for actual change. AR processes are, in a sense, self-implementing.

We are in agreement with Mingers and Gill ²⁰ that using several methods indeed provided a richer multi-facetted representation of the problems and potential solutions than what the application of either of the two methods in isolation would have provided. Using the mixed-method approach described here, we found that the groups tested a multitude of potential solutions in a very short time, discarded what did not seem to work, and converged on robust proposals that satisfied objectives and were accepted by all. A “pure” AR process usually has a considerably higher focus on reflexivity, group development, and iterative testing of actions in reality. ⁴³ A “pure” SD process has a focus on the building of a fully formed and rigorous model with a strong emphasis on feedback loops and non-linear interaction effects, and the complex dynamics that result from these. ⁴⁴ For GMB, where AR is integrated into SD, Oyo et al ⁴⁵ further underlines the importance of rigour in the modelling process but also discusses Scholl’s proposal of integrating SD into AR. To the best of our understanding, few studies have otherwise reported on this particular approach. Searches in Scopus, and in conference proceedings of the International System Dynamics Society in December 2020, using different combinations of the search terms “SD and AR” for papers detailing the integration of SD into AR resulted in three papers including Scholl. Holmström et al ²⁸ reports on an in-depth study of one of the cases in this paper, integrating SD into AR, and Walker and Haslett ²⁷ provides a case description of the outcomes of an AR process using SD. The latter, however, does not describe the characteristics of the combined process as such making it difficult to compare their data to the findings reported here. Furthermore, Rosmulder et al ²⁹ reported on AR used with another OR method in healthcare (discrete event simulation) and, similarly as suggested by our study, conclude that integrating simulations with an AR process is a promising combination to promote action in reality.

Facilitation

Group process facilitation is a vital part of the AR process. Vennix ⁴⁶ also notes that group facilitation is one of the most crucial elements in effective GMB in SD, when facing complex and “messy” problems as well as divergent stakeholders. Facilitators are usually transparent about how they are to proceed, but not necessarily explicit about the theory and practise of their work, as this could shift the focus from the work of the group to the facilitator. Straus ⁴⁷ states that facilitators rely on interventions or preventions. The former is applied to keep a meeting on track when obstacles occur, and the latter includes actions taken before or during meetings to eliminate potential barriers to success. Straus ⁴⁷ claims that the most skilful facilitators rely on preventions. As shown in Figure 2, case flows were pragmatically adapted to the facilitator’s perception of “where the group was.” This made his work invisible, so that participants appreciated their own work rather than that of the facilitator. All the cases studied began by facilitating the problem identification (including a process of clarification), where the purpose was to uncover all problems of concern, listen to and take note of the problems described without judgement and, for the group, to gain trust and acceptance in the facilitator and the process. This laid the ground for a way forward whereby participants would feel safe when challenging the modelling process as well as when challenging the simulation results and the subsequent interpretations. The facilitator also needs to be trusted in order to be a to be a “comforter” and to – temporarily – contain the anxieties stirred up by the uncertainty, ambiguity, and potential conflicts in an emergent process. ⁴⁸ The nature of this facilitation process caused the participants to take ownership in the solutions and to take responsibility for their actualization.

Strengths and Weaknesses

The presented re-analyses relied on a large body of contemporaneous notes about the improvement cases that were revisited and interpreted during the research process. The qualitative abductive approach, using initial constructs as a basis for analysis and preparation of displays that were discussed and reinterpreted in a multidisciplinary setting, iteratively lead up to new insights and clarifications that caused the perspectives of the research group to meld. Eisenhardt ³¹ describes how theory evolves in constant iteration, backwards and forwards between steps, and suggests that a priori constructs can shape the initial design of research creating new theory-building from multiple cases. We consider this as one of the strengths of our study. Any retrospective analysis may, however, be subject to recall bias on the part of assessed data and should be considered a possible limitation. More extensive and clearly structured notes and contemporaneous reflections in this study could have contributed to a deeper analysis. However, the support and rigorous questioning from the multidisciplinary group of researchers and the objective handling of the material may well have compensated for this. Had the research documented in this study been planned in a prospective setting, the experiences of the participants could have been recorded through independent observations and interviews. On the other hand, the act of recording, observing, transcribing, and measuring a process may cause an observer effect, affecting the behaviour of the participants and potentially influencing the process. ⁴⁹

As AR is situational, the outcome when integrating SD into AR is likely to be affected by this feature, ie, if the process is repeated it would neither be identical, nor would it produce the exact same results. ²⁶ The presented findings should, therefore, be taken as guidance to others to adapt to the situations that they encounter. It would also be helpful if the act of integrating SD or other OR methods into AR processes were described by others to enable us to understand how the investigated approach performs in other areas of healthcare as well as in other contexts.

Conclusion

Using a mixed-methods approach, where SD is integrated into AR, we found that there is potential to achieve outcomes that are more useful, comprehensive, and robust for the client organization than when applying either approach in isolation. From the investigated improvement cases in healthcare, we found that AR contributed to a high level of engagement among the participants and to the building of confidence in and ownership of the results. SD provided a coherent and consistent systems overview of the complex and complicated structure of each case that we studied, offered causal rigor, and provided ample opportunities for reality checks. The overall AR/SD process came together in two major divergent and convergent phases that stimulated creativity as well as led to a timely congregation of mental models. The process also led to shared points of reference for the problem as such, helped to identify robust solutions, and increased understanding of systemic effects resulting from putting suggested solutions in action. The results were calibrated to local needs and circumstances and created a higher likelihood of sustained actualisation compared to what results from a traditional top-down implementation typically would have done.

Acknowledgements

The original case work was funded as consultancy assignments carried out by the first author. Permission was given to use any results academically but anonymized. The first author is retired and has worked on the manuscript in his own time. CO is the main supervisor of the first author in his PhD studies and is partially funded by Swedish Research Council (2017-01753). The other authors have contributed as supervisors and part of their organizational roles.

An early version of the abstract by the same authors, containing aim, materials and methods was presented as work-in-progress at the 39th International Conference of the System Dynamics Society on July 21, 2020, virtually in Bergen, Norway, with the title “Insights gained from a reanalysis of five model-facilitated change processes in healthcare based on action research approaches.” The abstract can be found at: https://proceedings.systemdynamics.org/2020/abstracts/1192.html.

We would also like to thank our anonymous reviewers for helpful comments to improve previous versions of this manuscript.

Ethical issues

No ethical approval was needed for this study since it builds on already collected case data without personal details about participants.

Competing interests

Authors declare that they have no competing interests.

Authors’ contributions

Conception and design: PH, PD, FB, and CO. Acquisition of data: PH. Analysis and interpretation of data: PH, PD, FB, and CO. Drafting of the manuscript: PH. Critical revision of the manuscript for important intellectual content: PD, FB, CO, and TBE. Supervision: PD, FB, CO, and TBE.

Supplementary files

References

1. Brownson RC, Allen P, Jacob RR. Understanding mis-implementation in public health practice. Am J Prev Med 2015; 48(5):543-551. doi: 10.1016/j.amepre.2014.11.015 [Crossref] [ Google Scholar]
2. Davidoff F. On the undiffusion of established practices. JAMA Intern Med 2015; 175(5):809-811. doi: 10.1001/jamainternmed.2015.0167 [Crossref] [ Google Scholar]
3. Greenhalgh T, Wherton J, Papoutsi C. Beyond adoption: a new framework for theorizing and evaluating nonadoption, abandonment, and challenges to the scale-up, spread, and sustainability of health and care technologies. J Med Internet Res 2017; 19(11):e367. doi: 10.2196/jmir.8775 [Crossref] [ Google Scholar]
4. May CR, Johnson M, Finch T. Implementation, context and complexity. Implement Sci 2016; 11(1):141. doi: 10.1186/s13012-016-0506-3 [Crossref] [ Google Scholar]
5. Øvretveit J, Andreen-Sachs M, Carlsson J. Implementing organisation and management innovations in Swedish healthcare: lessons from a comparison of 12 cases. J Health Organ Manag 2012; 26(2):237-257. doi: 10.1108/14777261211230790 [Crossref] [ Google Scholar]
6. Bååthe F, Norbäck LE. Engaging physicians in organisational improvement work. J Health Organ Manag 2013; 27(4):479-497. doi: 10.1108/jhom-02-2012-0043 [Crossref] [ Google Scholar]
7. Mintzberg H. Structure in Fives: Designing Effective Organizations. Prentice Hall; 1992.
8. Lindgren Å, Bååthe F, Dellve L. Why risk professional fulfilment: a grounded theory of physician engagement in healthcare development. Int J Health Plann Manage 2013; 28(2):e138-157. doi: 10.1002/hpm.2142 [Crossref] [ Google Scholar]
9. Sterman J. System Dynamics: Systems Thinking and Modeling for a Complex World. Boston: Irwin/McGraw-Hill; 2000.
10. Rouwette EA. Group Model Building as Mutual Persuasion [thesis]. Nijmegen: Wolf Legal Publishers; 2003.
11. Ackermann F, Andersen DF, Eden C, Richardson GP. Using a group decision support system to add value to group model building. Syst Dyn Rev 2010; 26(4):335-346. doi: 10.1002/sdr.444 [Crossref] [ Google Scholar]
12. Voinov A, Jenni K, Gray S. Tools and methods in participatory modeling: selecting the right tool for the job. Environ Model Softw 2018; 109:232-255. doi: 10.1016/j.envsoft.2018.08.028 [Crossref] [ Google Scholar]
13. Holmström P, Elf M. Scoping Group Interventions for Suitability in Participatory Modeling. Operations Research Society Conference OR 52; September 2009; Warwick.
14. Andersen DF, Richardson GP. Scripts for group model building. Syst Dyn Rev 1997; 13(2):107-129. doi: 10.1002/(sici)1099-1727(199722)13:2<107::aid-sdr120>3.0.co;2-7 [Crossref] [ Google Scholar]
15. Ackermann F, Andersen DF, Eden C, Richardson GP. ScriptsMap: a tool for designing multi-method policy-making workshops. Omega 2011; 39(4):427-434. doi: 10.1016/j.omega.2010.09.008 [Crossref] [ Google Scholar]
16. Hovmand PS, Andersen DF, Rouwette E, Richardson GP, Rux K, Calhoun A. Group model-building ‘scripts’ as a collaborative planning tool. Syst Res Behav Sci 2012; 29(2):179-193. doi: 10.1002/sres.2105 [Crossref] [ Google Scholar]
17. Scott RJ. Group Model Building: Using Systems Dynamics to Achieve Enduring Agreement. Singapore: Springer; 2018.
18. Reason P, Bradbury H. The Sage Handbook of Action Research: Participative Inquiry and Practice. 2nd ed. London: SAGE Publications; 2008.
19. Deming WE. The New Economics for Industry, Government, Education. 2nd ed. Cambridge, MA: Massachusetts Institute of Technology, Center for Advanced Engineering Study; 1994.
20. Mingers J, Gill A. Multimethodology: The Theory and Practice of Combining Management Science Methodologies. Chichester: John Wiley & Sons; 1997.
21. Hesse-Biber SN. Navigating a turbulent research landscape: working the boundaries, tensions, diversity, and contradictions of multimethod and mixed methods inquiry. In: Hesse-Biber SN, Johnson RB, eds. The Oxford Handbook of Multimethod and Mixed Methods Research Inquiry. 1st ed. Oxford: Oxford University Press; 2015.
22. Zolfagharian M, Romme AGL, Walrave B. Why, when, and how to combine system dynamics with other methods: towards an evidence-based framework. J Simul 2018; 12(2):98-114. doi: 10.1080/17477778.2017.1418639 [Crossref] [ Google Scholar]
23. Howick S, Ackermann F. Mixing OR methods in practice: past, present and future directions. Eur J Oper Res 2011; 215(3):503-511. doi: 10.1016/j.ejor.2011.03.013 [Crossref] [ Google Scholar]
24. Brailsford SC, Harper PR, Patel B, Pitt M. An analysis of the academic literature on simulation and modelling in health care. J Simul 2009; 3(3):130-140. doi: 10.1057/jos.2009.10 [Crossref] [ Google Scholar]
25. Fone D, Hollinghurst S, Temple M. Systematic review of the use and value of computer simulation modelling in population health and health care delivery. J Public Health Med 2003; 25(4):325-335. doi: 10.1093/pubmed/fdg075 [Crossref] [ Google Scholar]
26. Scholl HJJ. Action research and system dynamics: can they benefit from each other? Paper presented at: 37th Annual Hawaii International Conference on System Sciences; 5-8 January 2004; Big Island, HI. 10.1109/hicss.2004.1265243.
27. Walker B, Haslett T. System dynamics and action research in aged care. Aust Health Rev 2001; 24(1):183-191. doi: 10.1071/ah010183 [Crossref] [ Google Scholar]
28. Holmström P, Hallberg S, Björk-Eriksson T. Insights gained from a systematic reanalysis of a successful model-facilitated change process in health care. Syst Res Behav Sci 2021; 38(2):204-214. doi: 10.1002/sres.2724 [Crossref] [ Google Scholar]
29. Rosmulder RW, Krabbendam JJ, Kerkhoff AHM, Houser CM, Luitse JSK. Computer simulation within action research: a promising combination for improving healthcare delivery?. Syst Pract Action Res 2011; 24(5):397-412. doi: 10.1007/s11213-011-9191-y [Crossref] [ Google Scholar]
30. Miles MB, Huberman AM, Saldaña J. Qualitative Data Analysis: A Methods Sourcebook. 4th ed. Los Angeles: SAGE Publications; 2020.
31. Eisenhardt KM. Building theories from case study research. Acad Manage Rev 1989; 14(4):532-550. doi: 10.2307/258557 [Crossref] [ Google Scholar]
32. Peirce CS. Collected Papers of Charles Sanders Peirce. Cambridge: Harvard University Press; 1931.
33. Kubr M. Management Consulting: A Guide to the Profession. 2nd ed. Geneva: International Labour Organization (ILO); 1986.
34. James M, Cotter S, Fairley A, Payne A. How to Use and Choose a Management Consultant. London: The Economist; 1989.
35. Rowbottom RW. Social Analysis: A Collaborative Method of Gaining Usable Scientific Knowledge of Social Institutions. London: Heinemann Educational; 1977.
36. Sjölund A. Gruppsykologi. Vol 2. Stockholm: Rabén & Sjögren; 1979.
37. Adizes I. Mastering Change: The Power of Mutual Trust and Respect in Personal Life, Family Life, Business, and Society. Santa Monica, CA: Adizes Institute; 1992.
38. Guilford JP. Personality. New York: McGraw-Hill; 1959.
39. Basadur M, Pringle P, Speranzini G, Bacot M. Collaborative problem solving through creativity in problem definition: expanding the pie. Creat Innov Manag 2000; 9(1):54-76. doi: 10.1111/1467-8691.00157 [Crossref] [ Google Scholar]
40. Parnes SJ. Guide to Creative Action. New York: Scribner; 1977.
41. Kolb DA. Experiential Learning: Experience as the Source of Learning and Development. 2nd ed. Upper Saddle River, New Jersey: Pearson Education, Inc; 2015.
42. Kennedy MR. Shapeholders: Business Success in the Age of Activism. New York: Columbia University Press; 2017.
43. Kemmis S, McTaggart R, Nixon R. The Action Research Planner: Doing Critical Participatory Action Research. Singapore: Springer; 2014. 10.1007/978-981-4560-67-2.
44. Rahmandad H, Sterman JD. Reporting guidelines for simulation-based research in social sciences. Syst Dyn Rev 2012; 28(4):396-411. doi: 10.1002/sdr.1481 [Crossref] [ Google Scholar]
45. Oyo B, Williams D, Barendsen E. Integrating action research and system dynamics: towards a generic process design for participative modelling. In: 2009 42nd Hawaii International Conference on System Sciences. Waikoloa, HI: IEEE; 2009:1-11. 10.1109/hicss.2009.266.
46. Vennix JAM. Group Model Building: Facilitating Team Learning Using System Dynamics. Chichester: John Wiley & Sons; 1996.
47. Straus DA. Managing meetings to build consensus. In: Susskind LE, McKearnen S, Thomas-Lamar J, eds. The Consensus Building Handbook: A Comprehensive Guide to Reaching Agreement. Thousand Oaks, CA: Sage Publications; 1999:.
48. Prins S. The psychodynamic perspective in organizational research: making sense of the dynamics of direction setting in emergent collaborative processes. J Occup Organ Psychol 2006; 79(3):335-355. doi: 10.1348/096317906x105724 [Crossref] [ Google Scholar]
49. Von Foerster H. Cybernetics of Cybernetics: The Control of Control and the Communication of Communication. 2nd ed. Minneapolis: Future Systems; 1995.