Significance

What is already known on this subject? The MCH workforce faces challenges that are dynamic and complex. Existing workforce approaches commonly take narrow perspectives rather than acknowledging broad dynamics of larger systems.

What this study adds? This is the first known attempt to identify all published research using SD to study MCH topics. The MCH workforce will be able to use this scoping review to (a) learn about the value of SD approaches for their work, (b) identify examples of strong SD approaches in MCH, and (c) consider potential applications of SD approaches in their own practice or research.

Objectives

Maternal and child health (MCH) is a far-ranging field encompassing issues of preterm birth, childhood obesity, sexually transmitted infections, and maternal mortality, among others. A common thread across MCH issues is the fact that they are both “complex and dynamic”, meaning they are caused by a system of interconnected factors (e.g., crossing socio-ecological levels) that continue to change over time (Kroelinger et al., 2014; Meadows & Wright, 2008). The contemporary MCH workforce faces tremendous challenges in responding to such persistent issues, particularly around making best use of limited resources and addressing issues of equity (Fanta, Ladzekpo et al. 2021; Mehta et al., 2021). This complexity requires MCH to work across system boundaries (e.g. organizational, disciplinary, geographic, life-course). Existing MCH strategies and approaches may be enhanced if the workforce embraced a systems perspective and integrated systems thinking tools into current practice (Kroelinger et al., 2014).

System Dynamics (SD) offers a set of tools and approaches for understanding behaviors of dynamic systems surrounding complex problems (Forrester 1961–1969, Meadows, 1999; Sterman, 2000, 2006; Homer & Hirsch, 2006; Meadows & Wright, 2008). SD acknowledges that problematic events or trends are produced because of the underlying system (Fig. 1); effectively addressing the problematic outcome requires an understanding of the system’s structure and the “mental models” of stakeholders keeping the problematic system in place (Maani & Cavana, 2007). Whereas many typical problem-solving approaches assume a consistent, linear relationship between variables, SD tools capture more realistic non-linearities caused by endogeneity (feedback) in causal relationships as systems respond to changes over time and delays in information and production (Haraldsson, 2004; Sterman, 1989). SD requires a more holistic understanding of key determinants of change over time, which aligns with the demands of the MCH environment.

Fig. 1
figure 1

The Iceberg Model to System Thinking. The Iceberg Model is a common framework used to guide system thinking (Maani & Cavana, 2007). The top level (“Event”) represents the visible part of a problem, whereas the lower levels (“Pattern/Trend”, “Structure”, and “Mental Model”) consider more deeply elements of the system that produce the problem and leverage points for change

Full size image

SD offers qualitative (e.g., causal loop diagrams) and quantitative (e.g., simulation models) tools. The application of SD tools can be organized into three approaches: qualitative diagrams, concept models, and tested/analyzed models; these approaches can be used independently or together as part of an iterative process, with qualitative diagrams being the starting point and tested/analyzed models being the end deliverable (Fig. 2). Qualitative diagrams are developed, often with stakeholders, to better understand complex, problematic trends that need to change. They facilitate conversations among diverse stakeholders by providing a tangible language for understanding structures and mental models surrounding persistent challenges. Concept models build upon qualitative diagrams by introducing preliminary (often hypothetical) numbers as model parameters and inputs. The model is then used to test hypotheses and explore impact of system feedback on outcomes of interest. The final iteration of SD approaches is tested/analyzed models. Model parameters are calibrated, often using historical data, until users feel confident in the model’s validity; from there, future trends and evidence can be generated using simulation modeling to test hypotheses, inform decision-making, and holistically study complex challenges.

Fig. 2
figure 2

Figure 2 provides an example of each of the three main approaches seen in SD: Qualitative Diagrams, Concept Models, and Tested/Analyzed Models. The qualitative causal loop diagram (a) and a qualitative stock and flow diagram (b) were used to facilitate conversations among a group of stakeholders in Munar et al. (2015) regarding the impact that limited staff and clinic capacity has on getting children vaccinated. The concept model (c) is adapted from a model published in Minyard et al. (2014) that was used as a teaching tool with state-level policy makers. The concept model (c) along with the six-questions framework (d) were used as part of an iterative process in which policy makers interacted with SD models to “test” the impact of various policy scenarios through simulation modeling. Finally, in Hosseinichimeh et al. (2018), a tested/analyzed model (e) was built to understand and quantify the interactions between depressive symptoms, rumination, and stress in adolescent populations. The model was parameterized (see formulas in Fig. 2e) and primary-collected longitudinal data was inputted to estimate (via simulation modeling) the impact of prior stressors on current levels of depression for 16 different groupings of adolescents (f). *Note The building blocks of all SD models include variables, arrows, polarity, feedback loops, stocks, and flows. Variables are written as nouns or noun phrases with clear meaning when they increase or decrease over time. Thin arrows drawn between variables indicate that a change in the first variable triggers a change in the second variable over time, all else equal. The polarity of causal links is labeled “S” or “ + ” to indicate the variables change in the same direction (e.g., if the value of the first goes up, the value of the second also goes up); they are labeled “O” or “ − “ to indicate that the variables change in opposite directions (e.g., if the value of the first goes up, the value of the second goes down). Feedback loops are closed chains of causal linkages that reinforce (i.e., exacerbate) or balance (i.e., stabilize) changes over time. Stocks depict accumulations of focal variables whose value or level is determined by the balance of inflows and outflows over time; flows are notated using a double arrow with an hourglass and represent rates of change in and out of a stock over time. For more information on SD models, including how to “read” a model, see “Introduction to Systems Thinking” (Kim, 1999)

Full size image

SD is well-suited for practical application by the MCH workforce due to the feasibility of stakeholder engagement in the modeling process (Cilenti 2019). SD approaches encourage ‘group model building’, in which diverse stakeholders come together to create a shared map of the system that maintains a persistent problem (Vennix, 1996, 1999). SD tools and approaches facilitate productive dialogues across diverse stakeholders about causes of persistent MCH issues and possible system responses to different courses of action (i.e. practice/policy changes). For more information on SD, see Sterman (2000).

In the midst of widespread health systems transformation and movement toward equity-focused approaches in public health, MCH professionals have been embracing leadership roles in cross-sector collaborations. The time is ripe for SD to become more common in MCH practice and research as a way for the workforce to fully understand the systems in which they operate, predict unintended consequences of program and policy choices, and lead – informed by tools that enhance understanding of complexity as the field moves towards centering equity practice (Kroelinger et al., 2014). As such, this manuscript presents a scoping review of existing SD literature with application to MCH (Munn, 2018). We organize our findings into three domains: (1) SD approaches applied to MCH research, (2) purposes for which SD was used by MCH practitioners, and (3) MCH topics studied using SD.

Methods

We attempted to identify all existing published research that used SD tools to study MCH topics between January 1958 and July 2018. Publications that met inclusion criteria were identified over four steps in the scoping review. These steps, discussed in detail in Online Appendix 1, were as follows: Step 1 used three different search strategies in the Web of Science Core Citation Indexes (WOS) and PubMed to identify research using SD between January 1958 and July 2018. Works that did not have a health sciences or health services focus were conservatively filtered out based on title-screen in Step 2. In Step 3, pairs of authors reviewed titles and abstracts (if available) of all works meeting our SD and health criteria to identify those that demonstrated an application of SD methods. Finally, in Step 4, pairs of authors reviewed abstracts and full texts to identify works that were relevant to MCH and thus eligible to be included in this review. This review builds on a previously completed systematic search for SD applications in health. Step 1 through Step 3 reflect efforts accomplished as steps in this earlier search and Step 4 reflects efforts specific to this scoping review; two members of the research team (KHL and IG) were among those who participated in the previous search process.

Information from the SD tools and approaches abstraction was double coded by a team of four authors who are experts in SD (KHL, IG, JS, SA). Information abstracted included: SD approach, model purpose, and level of stakeholder engagement. Three authors (DC, AM, SK) specializing in MCH practice, and collectively bring over 80 years of experience, conducted the MCH content abstraction. Information abstracted included: MCH-relevant Healthy People 2020 objectives addressed, MCH-relevant UN Sustainable Development Goals (SDGs) addressed, domestic- or global-focus, and utility for MCH research/policy/practice. All abstracted information was double coded and any discrepancies were resolved with the full abstraction team.

The abstracted information was chosen to be both practically useful to readers interested in seeking out works applying SD to MCH topics, and to gauge the extent to which SD/MCH researchers are studying high-priority MCH research topics. Definitions for SD approach, model purpose, level of stakeholder engagement, and utility for MCH research/policy/practice are provided in Table 1 footnotes. The authors chose the Healthy People 2020 topics to reflect current priority topics in US-domestic MCH research. The UN SDGs were chosen to reflect current international priority topics in MCH research. Two authors identified MCH-relevant Healthy People 2020 topics and SDGs using the project’s definition of MCH (see Online Appendix 2). The group then discussed and approved the included goals. Each work could be coded as studying any number of these goals, including none at all.

Table 1 Characteristics of maternal and child health studies using a system dynamics approach
Full size table

Results

Steps 1–3 identified 663 articles meeting criteria of SD methods applied to health. Of those articles, 521 articles were excluded for MCH irrelevance based on title and abstract review (Fig. 3). Additionally, 41 articles were excluded after full text review because they did not study an MCH population (n = 37), they were not an application of SD to MCH (n = 2), or the description of the SD work was not detailed enough to permit abstraction of the relevant information (n = 2). A total of 101 articles met all inclusion criteria and were included in this review (Table 1).

Fig. 3
figure 3

Results from Step 4 of Search Strategy (Moher et al., 2009)

Full size image

SD Approach

Of the 101 included works, by far the most common SD approach described was tested/analyzed models (n = 67). One example was in Hosseinichimeh et al., (2018), where the authors built, tested, and analyzed a SD model (Fig. 2e) to holistically study the complex relationships among stressors, rumination, and depression. Longitudinal, primary data collected from middle-school students was used an input data to simulate evidence on impact of prior stressors on current levels of depression for an adolescent population (Fig. 2f). This SD approach allowed researchers to better understand feedback created between stressors, rumination and depression, including average time adolescents tend to ruminate after activated by a stressor and corresponding levels of depression associated with lengths of rumination. Findings indicate opportunities to improve clinical interventions targeting pediatric depression. As a second example, in Frerichs et al. (2013), the researchers compared 15 different combinations of interventions to prevent and treat childhood obesity and 6 variations on adult-to-child impact factor ratios for these interventions to identify those with the highest levels of impact over a 10-year time horizon. This paper is a compelling example of the power of a rigorously tested model to deliver insights useful for MCH decision makers.

Of the 101 included works, 27 developed qualitative diagrams. For example, in Munar, et al. (2015) a causal loop diagram (Fig. 2a) and a stock and flow diagram (Fig. 2b) were used to facilitate conversations among stakeholders participating in the Salud Mesoamerica 2015 initiative in Honduras regarding the impact of limited staff and clinic capacity on the number of children vaccinated. Authors note that qualitative SD diagrams provide “tangible” tools that help diverse stakeholders with diverse perspectives articulate complex problems; using such diagrams to guide difficult conversations shifts the focus “from whether one person is right and the other is wrong, to a discussion about whether or not the diagram is correct, captures the relevant relationships, resolves a conflict, and so on” (Munar, et al., 2015). Other topics explored using qualitative diagrams included pediatric asthma management (Gillen et al., 2014), care transitions for children with disabilities (Hamdani et al., 2011), child care (Maital & Bornstein, 2003), neonatal mortality (Rwashana, Nakubulwa et al. 2014), homeless youth policies (Staller, 2004), cross-disciplinary collaboration (Munar et al., 2015; Pieters et al., 2011), and a new implementation evaluation method for programs with complex networks of structures and stakeholders (Fredericks et al., 2008).

Finally, we found 10 examples of concept models. One exemplary instance is a teaching model created for working with policymakers from the state of Georgia on childhood obesity. (Minyard et al., 2014) Georgia policymakers chose the topic of the model, directed the project, and were led through a “learning lab” that allowed them to experiment in the model with a number of different strategies to prevent and reduce childhood obesity (Fig. 2c, 2d). The group of policymakers reported that the learning lab informed the passage of a bill that proposed a unique combination of interventions to prevent childhood obesity. While we only found 10 examples of concept models applied to MCH, we believe concept models offer valuable opportunities for the MCH workforce to engage diverse stakeholders to understand and address complex MCH problems.

SD Purpose

We posited 4 purposes for which SD tools and approaches are utilized: increasing understanding, strategic planning, informing policy, or predicting (Table 2).

Table 2 Purpose of the system dynamics models
Full size table

The most common model purpose we identified was increasing understanding (~ 55% of results). One example, by Moxnes and Jensen (2009), describes the creation of a tested/analyzed model that simulates the user’s blood alcohol concentration (BAC). This model was used by high school students to explore a number of scenarios where teens exceed their intended BAC: drinking with a full stomach, and drinking to attain a particular level of BAC. When compared to students who received written educational materials, students who used the simulation to experiment with different drinking behaviors were better able to learn lessons that might help them avoid future binge drinking. This and another study (Siegel et al., 2011) show a promising use of SD models in helping individuals increase their understanding in order to modify risky behaviors after experimentation in a “learning lab” environment. Another example is Osgood, Dyck, et al. (2011), Osgood, Mahamoud, et al. (2011)), which studied the impact of gestational diabetes on future risk of developing type-2 diabetes for women and their children. Using data from Saskatchewan, they used a tested/analyzed model to trace the population’s progress through different disease states. The second example is more typical of SD models that attempt to increase scientific understanding.

The second most common model purpose identified was strategic planning, which involves comparing effectiveness among interventions or policies to inform decision-making (~ 25% of results). Hirsch et al. (2012) used this type of model to compare the costs and effectiveness of six different types of interventions addressing early childhood caries (tooth decay) singly and in combination. The authors used a tested/analyzed model with a ten-year time horizon, and used national and state data from the Colorado Child Health Survey, the National Health and Nutrition Examination Survey and the Medical Expenditures Panel Survey to make these results Colorado specific.

The third most common model purpose identified was informing policy (~ 11% of results). An example of this type of work is found in Ahmad’s articles (2005 & 2007) on tobacco policies (see Table 1), one of which compared the effects of a United States legal smoking age of 21 versus 18 (Ahmad, 2005a, 2005b). Using a tested/analyzed model with a 50-year time horizon, the author examines three different scenarios for how smoking behaviors (and subsequent health and cost outcomes) might be affected. The input values for the model came from national surveys and the literature, and were tested in a sensitivity analysis.

The final modeling purpose is predicting, where the researcher uses past system behavior to project future system behavior. Nine works (~ 9%) created a model for this purpose. These were used to predict ambulatory health care demand (Diaz et al., 2012), the US urology workforce (McKibben et al., 2016), the prevalence of people with intellectual developmental disorders (Lee et al., 2016), the prevalence of Kawasaki disease (Huang et al., 2013), the Taiwanese pediatric workforce (Wu et al., 2013), the shortage of physicians in Japan (Ishikawa et al. 2013), medical specialists needed in Sri Lanka (De Silva, 2017), nutrition status of the Colombian population (Meisel et al., 2018), and the supply of therapeutic oxytocin in Tanzania (Nadkarni et al., 2018).

MCH Topics

The content abstraction identified a broad range of Healthy People 2020 objectives studied (Table 3). While the topics studied in the 101 works varied, the two Healthy People 2020 objectives addressed most frequently were early and middle childhood, addressed by ~ 30% of the studies, and access to health services, addressed by ~ 26% (Table 3); the topic of early and middle childhood seeks to improve the healthy development, health, safety, and well-being of adolescents and young adults, and the topic of access to health services seeks to improve access to comprehensive, quality health care services. An example of studying the topic of early and middle childhood was found in Liu et al. (2016), where a tested/analyzed model compared three possible interventions to implement a tax on sugar-sweetened beverage to understand the impact on children’s weight over time. Of the works identified focusing on early and middle childhood, obesity and nutrition was most studied. Other examples of early and middle childhood literature included studies on developmental disorders (Bernard et al., 1977; Lee et al., 2016; Sheldrick et al., 2016) and immunization (Rwashana et al., 2009; Schuh et al., 2017). Of those works addressing access to health services literature, several specifically spoke to workforce needs (Barber & Lopez-Valcarcel, 2010, Ishikawa et al., 2013, Wu et al., 2013, Crettenden, McCarty et al. 2014, McKibben et al., 2016, De Silva, 2017) and STI-related services (Evenden et al., 2006; Hontelez et al., 2016; Kok et al., 2015; Viana et al., 2014; Zou et al., 2018).

Table 3 Healthy people 2020 goals
Full size table

We found that fewer works focused on the UN’s SDGs (Table 4). This seems unusual, given that we found 49 of the works focused on global problems and 5 focused on both global and domestic issues. The preponderance of works that did study a SDG focused on the goal to ‘end disease epidemics’ and/or ‘end preventable deaths’. The disease epidemics most commonly addressed were related to STIs.

Table 4 United Nations Sustainable Development Goals (UN SDGs)
Full size table

The majority of works were rated as “high” utility (n = 43) or “low” utility (n = 44) for MCH policy/practice, while 14 were rated as “medium” utility for the field. (Table 1).

Stakeholder Engagement

In addition to topics, methods and purposes, we also noted patterns in the selected studies regarding stakeholder engagement. While the majority of studies we found (n = 53) did not involve stakeholders in the modeling processes, there were 40 studies which included what we considered to be a high level of stakeholder engagement. A prime example of one of these studies was by Bridgewater et al. (2011), which studied youth violence in Boston and engaged stakeholders throughout the model building and analysis. Qualitative causal loop diagrams developed by the community were used as the basis for a tested/analyzed model with a 12-year time horizon to explore a number of interventions to reduce youth violence.

Finally, we found that the number of SD publications on MCH topics has been increasing rapidly in the past decade (see Fig. 4). These works have been spread across 69 publication sources, with the most common being PLOS ONE (n = 7) and Journal of Public Health Management & Practice (n = 7).

Fig. 4
figure 4

Number of MCH/SD articles published by year

Full size image

Conclusions for Practice

The application of SD to MCH topics described here include a broad range of approaches, purposes, topics, and levels of stakeholder engagement. The inventory of articles identified in this review provides guidance and direction to those in the MCH workforce looking to bring systems perspectives to their MCH work; however, many areas and approaches remain unexplored.

Qualitative diagramming studies appear to be underused in MCH/SD research. We see opportunities for future studies to draw on qualitative diagrams to bridge science and practice in support of addressing pressing, persistent MCH problems. Group modeling sessions could be integrated into qualitative studies involving in-depth interviews, focus groups, or ethnographic methods (Bridgewater et al., 2011; Hovmand, 2014). Bridgewater et al. (2011) illustrate how stakeholder-engaged group qualitative diagramming can produce insights about the system on its own. This type of qualitative diagramming can also be a stepping-stone for later modeling work. Weeks et al. (2013) illustrate that ethnographic research on MCH topics could be adapted into qualitative diagrams in order to extend their usefulness as drivers of policy. For MCH researchers wary of the mathematical skills necessary for tested/analyzed models, qualitative diagramming can supply a deeper understanding of complex problems in MCH without the time and skill investment of quantitative modeling.

Another future research direction is simulated life course studies, as typified by Osgood, Dyck, and Grassman’s 2011 study of the intra- and intergenerational impact of gestational diabetes on risk of type-2 diabetes. By using historical data to calibrate and validate the model, theories about the intergenerational cycle of risk can be tested. While such studies can never replace longitudinal cohort studies for testing life course theories, they may be able to rule out intergenerational effects if models containing them cannot replicate historical data using the range of parameters estimated in previous studies.

STI research appears to be at the forefront of MCH in terms of adopting SD approaches, possibly because of the similarities between stock-and-flow models and more traditional infectious disease/compartmental models from epidemiology. Childhood obesity has also been a fruitful area for research crossover between MCH and SD; in this case, researchers may have been more comfortable using SD models because they are more common in the biomedical sciences. Collectively, these two fields of research (STI and childhood obesity) have only contributed thirty-five publications, which means many questions regarding the health and wellbeing of MCH populations remain unstudied. One opportunity for MCH researchers and practitioners to lead the way is to incorporate a greater variety of social determinants of health in SD models.

Studies comparing interventions and policies were common, likely because the ubiquity of other modeling methods in comparative cost-effectiveness research makes transitioning to SD methods more acceptable. Unfortunately, many of these studies did not meet basic guidelines – as outlined by the International Society for Pharmacoeconomics and Outcomes Research’s Consolidated Health Economic Evaluation Reporting Standards—for economic evaluation and comparative cost-effectiveness research in terms of reporting style or validation/sensitivity analyses (2000; Sculpher et al., 2000; Garrison, 2003; Weinstein et al., 2003; Husereau et al., 2013). Given the workforce’s role to prioritize actions that make best use of limited resources, assessing the business case for competing interventions is a valuable application of SD methods in the MCH field. However, future research should draw on existing standards for cost-effectiveness research in order to clearly report higher-quality results and best support decisions on resource-allocation.

Finally, MCH professionals should take advantage of teaching and collaboration opportunities inherent in model building. Several studies in this review created interactive, non-intimidating dashboards for their models that laypeople could use with relatively little training (Minyard et al., 2014; Moxnes & Jensen, 2009; Rauner, 2002; Siegel et al., 2011). For some of these projects, the goal was to allow policymakers and public health leaders to try out a number of policy scenarios and receive graphical or simplified feedback on how these policy decisions might affect key outcomes of interest: costs over time, people cured or reached, people missed or harmed, and unintended consequences. In other projects, the goal was to help patients learn how to manage their own health. Interactive models are a way for policymakers, public health leaders, and stakeholders to experiment using methods that deliver consequence-free and evidence-based results in minutes.

The papers in this review demonstrate the potential for the MCH workforce to use SD to understand complex problems such as STI control, obesity, oral health, substance use disorders, and workforce planning. However, many of the wicked problems facing MCH populations, including equity practices, remain unstudied using SD. Furthermore, few of the SD applications described here were then translated into significant action to address the problem under study. These tools have untapped potential. In this critical period of health transformation, SD can produce a better understanding of the varied, multilevel forces interacting to produce the complex problems facing MCH professionals and policymakers.