Literature was searched using the Scopus and PubMed databases and used the following search terms: causal loop diagram*, complex*, system* thinking, method*, tool, approach, research, and public health. Inclusion criteria were 1) public health research, 2) peer reviewed journal article, 3) described or created a CLD as a research method, and 4) published in English from January 2018 to March 2021. The key objective was to find state-of-the-field examples of CLDs, therefore, extensive hand searches of references was completed. It is important to note that piloting this search strategy uncovered numerous articles that only mentioned CLDs and did not explicitly meet the criteria of “described or created a CLD as a research method.” While we set out to use PRISMA guidelines we deemed it unnecessary given the search strategy quickly became one of including all articles that meet our inclusion criteria.

Study selection was conducted by one author (LBL) while appraisal and duplicate independent data extraction and validation was conducted by two authors (LBL and CH). CN provided input throughout the study and facilitated discussion about any differences. Data extraction followed these six categories:

1) Research aim,

Two authors (LBL and CH) extracted verbatim text that aligned with the extraction categories and these were saved to a spreadsheet. Both authors reviewed the spreadsheet in its entirety, discussed individual articles to gain clarity, and wrote summary paragraphs to identify high level themes. Following this, for each article, summary statements were written for the six extraction categories and a table was created. The two authors reviewed each other’s summaries for accuracy and revisions were made. Finally, directed content analysis was used to interpret extracted data “through systematic classification of coding and identifying themes and patterns” [25].

Although the literature addressed a range of public health topics, non-communicable disease prevention was most frequently addressed (15/23) and of those, seven were focused on obesity prevention. Table 2 provides a list of research topics.

In terms of research aims found in the 23 articles, four themes emerged: 1) to examine the complexity of a public health topic and illustrate complex systems thinking [26–34]; 2) to discuss the complexity of a public health intervention [35–40]; 3) to describe study protocol and how CLDs were created [41–44]; and 4) to illustrate how CLDs can be used to monitor and track initiatives to improve population health or evaluate impact of interventions [45–48].

Complex systems thinking was discussed in terms of systems, problems, interventions, and key concepts that drive this type of approach. Several articles indicated that the systems they were studying were complex, for example:

Following on this, feedback loops in complex systems were explicitly discussed in all articles to some extent. Jalali et al. [38] described these in terms of “causal chains of multiple variables in which changes in each variable could be traced back to its historical values.” They go on to define the difference between reinforcing and balancing feedback loops.

Another way complex systems thinking was described was with respect to complex problems and interventions. Burrell et al. [36] discussed community violence in terms of embedded contexts and the lack of holistic understanding of such “dynamic complexity.” Complex problems and interventions were often discussed together. The need to move away from “isolated intervention thinking” to systemic interventions to study systems change was highlighted by Knai et al. [30].

All articles built upon the descriptions reported above in some manner when discussing complex systems thinking. Some articles described this as providing “the opportunity to understand, test, and revise our understanding of how the different components in a system work together” [31] and “to study complex problems as the manifestation of dynamic interactions among their constituent parts” [36]. Furthermore, a few articles expanded the discussion to include such concepts as boundary judgement [38, 43, 47], that is, “establishing boundaries to the system is a fundamental starting point to efforts to change systems” [47].

CLDs were mostly seen as a means or a tool to examine feedback at play in public health issues. Some articles were explicit [28, 32, 33, 40, 43, 44] while others implied this. Both Riley et al. [43] and Parmar et al. [40] labeled this as “causal loop analysis” and the resulting CLDs were a means to understand systems and potential “programming.” Using a CLDs was a new tool for some [42, 46] and as one article related, “business as usual” was not working to address obesity [47]. CLDs were also considered a tool to help tell a story. For example, a CLD was thought to support the development of “a concise narrative about a particular problem” [42] and Brereton et al. [28] stated that “every causal loop tells a story that links cause and effect through feedback.”

There were many combinations of methods used to create CLDs. In this section we present this diversity in terms of 1) data sources, 2) processes, 3) data analysis, 4) frameworks, and 5) diagramming (Table 3).

There were nine ways that CLDs were intended to be used and these are identified in Table 4. The following provides examples of each intended use.

We learned that some articles described in detail theoretical orientations with respect to complex systems thinking while others gave brief explanations. The most frequent concepts regarding complex systems were the inherent dynamic interactions among many entities, factors, variables that illustrate whole system structure and behavior. This is consistent with other public health literature on the topic [52–54]. The difference in descriptions was more a matter of comprehensiveness than definitions. For example, boundary judgement was not well articulated in the articles. According to Ulrich [55], drawing boundaries builds in selectivity and partiality and therefore transparency is important in study design. Therefore, we recommend that attention be given to defining boundaries to signal a specific endogenous perspective and a unique, snap-shot-in-time diagram of feedback loops of system behavior [56].

Some articles had strong theoretical coherence with respect to complex systems thinking that was demonstrated in discussions about the reasons for choosing, creating, and using CLDs. We learned that articles were most coherent when they first discussed feedback loops from a theoretical perspective and then carried this through to creating CLDs and to using them to identify leverage points for systems change (see for example 30). Overall, the descriptions of feedback in the articles were aligned with the idea that CLDs are “the applications of the loop concept underlying feedback and mutual causality” and that feedback loops are “powerful unifying notions that illuminate the structure of arguments, explanations, and causal views” [56]. Meadows [51] is well-known for explaining that disrupting or amplifying feedback loops can be effective leverage points in systems change. Therefore, we recommend that future research be designed with this theoretical coherence in mind.

Lewin’s famous statement that “there is nothing so practical as good theory” was salient for what we learned [57]. Few articles used theoretical frameworks in research design or discussed the need to advance theory (i.e., complexity, systems) in public health research. The articles that used frameworks appeared to be more robust especially with respect to embedding theoretical constructs in the resultant CLD (see for example 35). While we appreciate that theory is emerging, we recommend that this be given more emphasis to help continue to build a solid foundation for furthering the application of CLDs in public health research.

Knai et al. [30] pointed out that current public health research “concentrates mainly on a system’s elements rather than the interconnections within it, and this is beginning to reveal its intrinsic limitations.” Some articles described data analysis methods to identify variables and examine interconnections to draw CLDs, however, others lacked clear descriptions of the often highly iterative methods and therefore it was difficult to follow a data trail and assess the resultant CLD. We recommend that more clarity be provided as to how researchers innovate in qualitative data analysis to further develop the art and science of creating CLDs.

We found a range of research methods used to create CLDs. Ozawa et al. [58] state that mixed methods research is important

We hypothesize that mixed methods may produce more robust CLDs, however, this needs to be examined. We recommend that future research be undertaken to assess the strengths, limitations, and benefits of using mixed methods and determine what methods create greater confidence in the variables and feedback loops illustrated in CLDs.

We found there was a wide range of who was involved in creating CLDs, from researchers only to multiple group model building sessions with stakeholders and community members. We see the latter methodology embedded in the traditions of action research [59] and/or community-based participatory research (CBPR) [60]. The CBPR approach involves “a commitment to conducting research that shares power with and engages community partners in the research process” and is intended “to increase knowledge and understanding of a given phenomena and integrate knowledge gained with interventions and policy and social change” [60]. There was little discussion of CBPR in the articles. We recommend that greater engagement with participatory action research literature be undertaken to embed the theory and philosophy of genuine participation and empowerment in research and action.

There was limited discussion regarding how exactly CLDs were to be used to enhance evidence-informed policy and practice. Few articles explicitly discussed incorporating knowledge users or those able to use research results. As Sturmberg [61] relates, this requires users who are “deeply interested in understanding the highly interconnected and interdependent nature of the issues.” This led us to think about the importance of knowledge translation (KT) and how to strengthen the use of CLDs. Haynes et al. [6] state that KT needs to be conceptualized as not “a discrete piece of work within wider efforts to strengthen public health, but as integral to and in continual dialogue with those efforts.” We recommend that future public health research using CLDs should articulate KT plans that articulates knowledge user engagement in defining outcomes for strengthening public health policies and practices.

We conceptualize public health research to be guided by principles of social justice and human rights to address the goal of reducing health inequities through action on the determinants of health. Although many articles discussed determinants of health, the goal of reducing health inequities was largely absent. Baum et al. [62] discuss the concept of path dependency as “the tendency of institutions to retain policy directions and preferences rather than change or reform them.” They further suggest that disrupting “path dependency that exacerbates health inequities” is critical and we see how CLDs could uncover path dependencies. We recommend that CLDs in public health research should include the examination of leverage points for pro-equity policy and practice.

Senge [63] states that “reality is made up of circles” but often arguments and explanations are linear, therefore, CLDs can provide “a language of interrelationships” to uncover deep patterns in systems. Studying the interrelationships and explanations of each CLD was outside the scope of this paper, however, we learned about some basic elements of reader friendly CLDs. We recommend that the following questions could be used assess CLDs: Are established conventions [56] used effectively for drawing the CLD (e.g., labeling, positive and negative arrows, reinforcing and balancing loops)? Does the diagram illuminate the most significant variables, feedback loops or leverage points? How well does the diagram function as an effective medium for presenting findings to knowledge users? How well does the CLD tell a story of what’s going on in a system?

In terms of limitations, the 23 articles were not considered to be comprehensive. Since completing the study, we found that Mui and others [64] published an article on a community-based system dynamics approach and suggests solutions for improving healthy food access in a low-income urban environment. We also found Savona et al. [65] identified the views of adolescents regarding the causes of obesity and used CLDs. While this can be considered a limitation, we hope to see a continual building of knowledge and skill in using CLDs in public health research. A strength of this paper is that 23 recent articles were identified that used CLDs and the depth and breadth of discussion in the articles provided good representation. Having three authors conduct the literature review is also a strength because this afforded a high degree of confidence in reporting results and transparency in search strategy and data extraction, analysis and synthesis. Together the results and recommendations can contribute to informing global public health research by highlighting key considerations to help design research and address public health issues through complex systems thinking.