The Philadelphia case study was one of four such childhood obesity prevention studies commissioned by the National Collaborative on Childhood Obesity Research (NCCOR) from 2006 to 2010 to identify possible contributors to declines in childhood obesity (29, 30).³ We used systems mapping to incorporate systems thinking perspectives into the use of the GTE framework to more explicitly capture reinforcing synergistic effects and better illustrate how, in a multi-intervention context, any one strategy might impact a subsystem embedded within the web of factors interacting to facilitate the childhood obesity declines in Philadelphia. Being able to illustrate dynamic interactions, or feedback, within a theory of change can be helpful for post hoc analysis and, by extension, intervention planning and implementation.

The study context, methods and findings of the Philadelphia CODP are available from a detailed report on the NCCOR website, supplemented by a series of articles providing further methodological details, analyses, and perspectives (31–33). Research sites for CODP were selected on the basis of reports of statistically significant declines in child obesity prevalence associated with concurrent or recent community-wide initiatives. The Philadelphia site, with ~190,000 enrolled 5- to 18-year-old students in the School District of Philadelphia (SDP)—the eighth largest school district in the U.S.—was highly relevant from a health equity/health disparities perspective. When the baseline data were collected in 2006–2007, the majority of students in the SDP were African American (62.9%) or Hispanic (16.6%), with nearly half from families eligible for free and reduced-price school meals (48.9%) (34). These percentages were similar or, for free and reduced priced meal eligibility, higher (57.4%) among SDP students in 2009–2010. National data for the period between 2006 and 2009 indicated higher prevalence of obesity among children and adolescents in the non-Hispanic Black and Hispanic categories compared to the non-Hispanic White category. These disparities were observed in one or both sexes overall and in some cases when stratified by low-, middle, and high-income (13, 35).

Obesity prevalence across the CODP study period was estimated from annual height and weight measurements collected for about 60% of the SDP students (similar but varied from year to year) with a similar demographic breakdown to the overall SDP population. Overall, in the data for children in priority populations, childhood obesity and risk reduction declines in prevalence were small but statistically significant, a 4.8% relative decrease for obesity (21.5%−20.5%; p < 0.001) overall and a 7.7% decrease (8.5%−7.9%; p < 0.001) for severe obesity, and were significant for Hispanic and African American children in one or both sexes and for obesity in non-Hispanic white males (34).

The CODP research approach used an adaptation of the Systematic Screening and Assessment methodology (36) to collect narrative data including key informant interviews about PSE relevant initiatives in the Philadelphia community during the pre-study period beginning as early as 1991 through the period of data collection between 2006–2007 and 2009–2010. The narratives described both the initiatives and how they were implemented. The matrix of strategies identified in the main report included 30 policies and programs (29). For the present analysis, data were extracted for the four nutrition-related strategies that key informants interviewed for the case study viewed as having been focal in contributing to the obesity declines—likely to have reached both the largest number of children and the children at highest obesity risk. These strategies occurred within or were linked to the schools concurrent with the CODP study time frame: a Universal Feeding Pilot (UFP) for School Meals; a policy, Eat Right Now, that provided for nutrition education delivered in both school and community settings; a ban on sugary drinks on school premises; and a comprehensive, district-wide School Wellness Policy. Extracted details for these strategies are provided in Supplementary Table 1. The descriptions included details of approaches, barriers, facilitators and partner roles that permitted inferences about pathways through which the interventions were linked and supported the systems mapping. We also consulted a government report on the early experience with the UFP for additional clarity about how the Philadelphia SDP administered and implemented the program (37).

Shown graphically in Figure 1, the icons at the top of the framework represent foundational principles for applying the program using an equity-sensitive perspective—an “equity lens,” using a community-engaged approach, and maintaining a “people-centered perspective” that considers the socioeconomic and sociocultural circumstances, values, priorities and needs of the people in the settings or environments of interest. The circle in the center has four quadrants with examples of possible associated actions or equity-related aspects in the callouts for each quadrant. Numbering of the quadrants is for ease of reference, although the intended use of the framework is iterative and not necessarily sequential. The top two quadrants relate specifically to PSE change approaches. Quadrant 1 focuses on a main policy or programmatic change of interest and its relevance to an equity focus on the setting and population. Quadrant 2 prompts for factors that might work against the effectiveness of this approach in the setting and population of interest. The bottom two quadrants relate to enhancing individual (Quadrant 3—facilitating resources that can assist with social and economic needs linked to participating in and benefitting from the intervention) and community resources and capacity (Quadrant 4—identifying and enhancing community level assets, health promotion contexts, and resources to foster uptake and sustainability). Most pertinent here, the scales of justice in the center prompt for systems thinking, underscoring the potential for synergistic effects on equity from a combination of multidimensional, mutually reinforcing elements in the four quadrants. A more detailed explication of the GTE framework is available as an online user toolkit (38).

Logic models are often the tool of choice in illustrating the types of multifactorial intervention processes that would be modeled using the GTE framework, which requires inputs from multiple domains and different types of activities to achieve a desired outcome. They can represent the sequential set of inputs, activities, outputs, and outcomes that need to cumulatively occur in order to achieve the desired goal or outcome (39, 40). For example, activities across GTE quadrants 1 and 2 can be viewed as operating on an additive scale, that is: the combined impact of policy or programmatic change and the factors working against that change can be understood through traditional linear approaches (i.e., actions to improve obesity minus deterrents that operate counter to those actions equal the net intervention impact). However, the linear theory of change presented in logic models is limited in capturing synergies between activities or actions that emerge through feedback relationships between factors (39, 40). Feedback or mutually reinforcing processes, such as those capturing the multidimensional activities in quadrants 3 and 4 of the GTE framework, require tools that can illustrate cumulative, non-linear effects, and the ways in which they amplify impacts on desired change. These activities occur in a context that can better be understood as a system.⁴ Box 1 details the key terminology used to describe systems and their properties. Within a systems lens, the non-linear effects characteristic of quadrants 3 and 4 are produced by the pattern of interrelationships, feedback, and time delays that accumulate to produce a particular behavior over time (see Box 1). In this case, the system is the set of actions that can lead to equity in the impact of addressing childhood obesity. Within the systems toolkit, which includes a wide array of approaches to analyzing systems (42, 43), we chose a systems mapping approach because of its detailed visualization tools. These tools are helpful in illustrating the specific inputs, activities, outputs, and outcomes, similar to logic model diagrams, but also to visualize interrelationships between factors and with contextual variables, such as community and social context that characterize the scales of justice at the center of the GTE framework. The diagrammatic visualizations from systems mapping are helpful for facilitating shared understanding and action planning.

The GTE framework guided analyses in which selected observations from the CODP data for Philadelphia were classified according to the GTE quadrants (see Section 2.4), and then conceptualized from a dynamic systems perspective using causal loop diagramming (CLD)—a qualitative systems mapping approach—to illustrate how the multi-component strategies undertaken in Philadelphia during the study period might have interacted to lead to the observed declines in childhood obesity (see Section 2.2). We use CLDs because they are a systems mapping approach that provide a useful balance of visualizing not only the elements in the system (i.e., variables) but also how they work together (i.e., causal connections, direction of causality, delays) while remaining largely accessible to public health audiences.

For the initial GTE-based systems mapping, we focused on the four interventions discussed in Section 2.3 that focused on increasing access to and consumption of healthier foods and beverages in schools (see Figure 2). The Universal Feeding Pilot was mapped to Quadrant 1 but was also recognized as relevant to Quadrant 3 because the program intrinsically provided financial benefits to families, in addition to its main goals of addressing food insecurity and improving access to healthy food options through meals. The SNAP Ed “Eat Right Now” program was mapped to Quadrant 4 because it was interpreted as reaching the broader community, but was also mapped to Quadrant 1 as a school-based educational program.

Figure 3 shows the generalized model used to anchor our systems analysis of the CODP case study, developed from standard system structures (39, 44). The model starts by defining the intervention target in a goal/gap structure where the current state of the outcome of interest differs from the desired goal of that outcome creating a gap that leads to action being taken (often in the form of an intervention; Figure 3A). That intervention then develops inputs, activities, outputs, and outcomes to influence the current state such that it grows closer to the desired state. In Figures 3B, C, we use the same systems diagramming approach to represent the GTE framework; successively building out the reinforcing feedback pathways that link the quadrants together. We first created basic connections between the quadrants of the framework that reflected their descriptions (Figure 3B). Next, we add in the synergistic reinforcing relationships between the quadrants (Figure 3C). This format also allows us to represent the long-term time delay (illustrated by the hash marks on the arrows representing an accumulation) between the cause and effect impacts of successful interventions to promote healthy options. In particular, as the intervention is successful and shifts the current state of obesity to the ideal state of obesity, the deterrents are then less influential, and there is more capacity to improve other social and economic resources and build on community capacity to address increases in obesity prevalence. Additionally, the social and economic resources that families accumulate over time, allow them to contribute to the community's capacity to respond to obesity reduction strategies and activities. Using the three systems diagrams as our analytic heuristic, we applied them to the CODP case study to depict the contributions that systems mapping can add to understanding success in a dynamic, equity-oriented context.

We used causal loop diagramming, building on the core feedback framework for intervention and program planning or—in this analysis, retrospective—evaluation, illustrated in Figure 3 to, analyze the case study described in Section 2.3. We first illustrated reinforcing feedback in the core GTE model as an anchor for synthesizing activities and factors present in the case study. Then, using the core heuristic in Figure 3, we built in the CODP-specific activities, factors, and interrelationships that were identified as driving successful outcomes in Section 2.2.

The four focal initiatives defined the boundary of our analysis around initiatives that occurred in the same/similar contexts, i.e., were school related. We extracted details about the four focal initiatives (see Supplementary Table 1) and iteratively reviewed the strategy descriptions to identify variables or factors reported as part of the success of each strategy, the causal connections between factors, and the direction of these causal relationships. We assessed whether a change in the causal variable created a change in the effect that went in the same direction (positive relationship) or whether a change in the causal variable created a change in the effect in the opposite direction (negative relationship).

After identifying variables, causal connections, and the direction of the causal connections, we used Vensim software version 10.1.4 (Ventana Systems, Inc., Harvard, MA) to create a diagram detailing the systems map. After mapping these connections, we used causal loop diagram guidelines to identify feedback loops throughout the causal connections and assess whether these feedback loops were reinforcing—creating more of the behavior that the loop created in the first place—or balancing—counteracting the behavior that the loop created in the first place. We also classified variables and loops based on the quadrants described in Figure 2 and coded them with different line patterns accordingly: thin solid lines for increasing healthy options, thick dashed lines for reducing deterrents, thin dashed lines for improving social and economic resources, and thick solid lines for building on community capacity. As a research team, we verified our interpretations and adjudicated any differences through discussion within the team, which included drawing on direct observations and subjective insights from the team member (SK) who was aware of and in some case involved in the childhood obesity prevention efforts during the pre-study and active study periods. Through these discussions, we identified an additional important system process that was not explicitly represented in the GTE model, and integral to the success of each initiative; we identified unintended consequences and co-benefits as key parts of system structure and used a dotted line to classify them. Throughout this process, we iteratively refined the map of the system for both parsimony and accuracy.

In Quadrant 1 (Figure 4A), loop B1 represents the system structure of the intervention that CODP identified as improving healthy food options for Philadelphia public school children. The UFP was implemented to address food insecurity, as illustrated in loop B2 in Figure 4B, but was also expected to improve children's healthy food consumption because of the nutrition standards for school meals and other foods available in school environments and the concurrent nutrition education program which are illustrated in depth in Quadrant 4. Increasing access to healthy food for a substantial proportion of public-school children could potentially contribute to closing the gap between observed and targeted obesity prevalence in this population. Stigma associated with having a free- or reduced-price meal was identified previously as a barrier to participation in the National School Meals program when individual children qualified based on a poverty-threshold. The UFP addressed this potential deterrent, as indicated in loop R1, to program participation by shifting eligibility to the school- rather than individual child level. Making all children in a qualifying school eligible to receive free school meals avoided the need to stigmatize some children as “poor” and increased participation. In turn, the increased participation, led to higher “tray” counts which provided direct financial benefits for the SDP as illustrated in loop R7, based on the per meal reimbursement rate.

The objective in Quadrant 2 is to identify and reduce deterrents to effectiveness of the intervention of interest. Figure 4B shows the capacity strain that was present for school staff before the UFP pilot. They were required to enroll students in the School Meals program individually and manage individual families' applications. Staff had to review paperwork, manage eligibility criteria, and process applications on a constant basis since eligibility status was in flux with most families. These capacity strain factors were alleviated, as illustrated in loop R7, due to the pilot's school-wide eligibility and streamlined process of counting and tracking the number of meals served. At the start of the UFP, an additional challenge was the availability of sugar-sweetened beverages (SSBs) in some schools through vending machines. These vending machines had unhealthy food options that students could access if they found the vending options more appealing or palatable than the food served in the school meals. However, an intervention to remove SSBs from vending machines in some schools was implemented prior to the implementation of the UFP in anticipation of this being a detrimental deterrent.

Our analyses identified several unintended adverse consequences and positive side effects (co-benefits) that, in retrospect, provided some of the bigger challenges or were integral to UFP success (Figure 5). In particular, there were three main positive side effects that built up to subsequent greater gains for later program implementation. Loop R2 depicts families' benefit by: (1) spending less time completing enrollment; (2) having the option of spending less time in food preparation; and (3) having more money to spend on food for other purposes or other things when their children ate school breakfast and lunch. As we will explore more in Quadrant 3, we interpreted this as having important implications for families' ability to improve their social and economic resources as well.

On the school side, the removal of responsibility for identifying and differentiating children who were eligible for a free or reduced-price meal vs. paying for meals or tracking and collecting lunch payments alleviated administrative burden and was a major positive side effect of the UFP. Also, rolling out a meals program on a larger scale was anticipated to increase capacity strain but this was mitigated by the greater saving of staff time, as described in loop R3, as well as the potential for increased funding associated with more students taking school meals, as depicted in R6 and R8.

However, there were challenges associated with three unintended consequences. First, while food was more available through meals, as explored in loops B9–14, students reported that the food was unappealing and thus they still relied on alternative, often unhealthy, food sources during the day rather than taking the school meal. These sources included corner stores near the schools, which were a known deterrent to healthy eating even before UFP implementation. A concurrent pilot program that targeted corner stores near schools to improve food options and choices (through in-store education) served as a counterbalance to this challenge. Second, for students who did take the school meal but found it unappealing, both not consuming the school meal (i.e., plate waste) and reverting to unhealthy foods were problematic, as demonstrated by the processes in loops B9-B14. The SDP would count these as reimbursable meals but the program objective of students actually consuming healthier meals would not be achieved in those cases—resulting in a waste of program resources. The third challenge was that most schools did not have the in-house capacity to cook and/or serve meals. This increased the reliance on food delivery that was not always healthy or cost effective through third party vendors and would also have impacted the meal quality, as depicted in loops B5–B8.

Interventions in both Quadrants 1 and 2 and the positive side effects discussed above set the stage for amplifying the social and economic resources that families could leverage (Figure 6). In particular, because families did not need to feed students at home, they both had more money for other financial needs (loops B15–18) and more time to attend to other things (loops R13–18). This positive side benefit for families aligned with the objectives of the partners who saw the UFP as a way to address poverty and food insecurity. It led to an additional, positive co-benefit of increasing multisector support for food and nutrition programming in communities, depicted in loops R13–R18, and collaboration between stakeholders that enhanced social capital through willingness to partner (loops R19–R21) and ability to mobilize. In turn, the enhanced willingness to partner further amplified the ability to advocate for additional policies, particularly supporting the SSB ban district-wide as depicted in loops R9–R12. Furthermore, social capital gains fed back on themselves to amplify the multi-sector support.

The accumulated gains from the co-benefits and the social and economic resource improvement both impacted and were further amplified by community capacity enhancement efforts through a few key programs (Figure 7). The Eat Right Now program was active in, but not limited to, schools; it reached children and adults in the Philadelphia community at large. This education, combined with collaboration gains discussed above set the stage for the Comprehensive School Wellness Policy which, in turn, impacted the nutrition standards overall, as we illustrate in loops B19–B24. Finally, over time, these efforts to leverage community assets and enhance capacity fed back to influence norms and resources for food access that further shaped the foundation from which food and nutrition policy and broader health promotion efforts were approached going forward and potentially accelerated the trajectory toward closing the child obesity gap in Philadelphia, which we illustrate in loops R22 and R23.

Figure 8 presents the full systems map for the CODP assessment of how the four focal interventions in Figure 2 intersected to result in cumulative, dynamic processes. This map illustrates how retrospectively operationalizing the synergistic and mutually reinforcing pathways using systems mapping provided a practical way to understand the cumulative processes that had to operate in concert to achieve even a modest gain in obesity reduction among public school children in Philadelphia. Furthermore, by clearly illustrating actions from each quadrant, we illustrate that factors across quadrants do not have to be balanced in number in order to result in equitable outcomes.

One strength of this retrospective analysis is that it focuses on a PSE approach, universal school meals, which has an intrinsic focus on equity and has become established practice. The information gleaned from CODP reports and manuscripts provides details on four focal initiatives, their engagement with the context and priority populations, and the ways in which the programs evolved over time within the schools and communities. Additionally, the CODP documentation included data collection during both the pre-study and active study periods, which helped with the development of the systems map.

This analysis also has notable limitations. Given that the case study was an associational study, the data collection and declines in obesity prevalence were concurrent but were not able to be linked to certain schools, individual children or students' food consumption. Also, there might have been other initiatives or external factors occurring concurrently in the community outside of the set boundaries that may have potentially impacted the outcomes and ultimately, the assumptions made in the systems mapping component (e.g., other benefits impacting a family's economic status such as child tax credits). Lastly, there were no CODP data available to analyze students' dietary intake or that addressed contextual factors not directly related to food or obesity that might have affected a student's need or willingness to receive free meals.