Findings from this study yield several insights with respect to participants’ perspectives on systems that produce and maintain disparities in food access and diet in Philadelphia: First is that participants placed relatively lower emphasis on downstream variables and greater emphasis on upstream variables. The most highly connected downstream variable was diet disparities, the main outcome that participants were asked to explore. Other highly connected downstream variables were those related to both physical and economic access to affordable, healthy, and culturally appropriate foods. Key upstream variables included those related to the social, policy, and political systems in which minoritized urban communities are embedded. Among the most highly connected variables were those related to racism (e.g., White supremacy, racial oppression, racism, structural racism, redlining, and residential segregation) and community power (e.g., community land control, community investment, and empowerment).

Importantly, a variable with a high degree centrality is not necessarily a leverage point or a place in a system where a relatively small change can have an outsized impact on system behavior [30]. Identification of leverage points requires examination of the strength and directionality of causal relationships, as well as the overall structure of the system. However, several of the most highly connected variables in the synthesis CLD—and particularly the upstream variables related to racism and community power—were included in feedback structures identified during the workshops and prominent in discussions of system structure. For example, variables related to residential segregation (i.e., segregation, redlining, gentrification) were discussed in all three workshops and were involved in a high number of feedback loops. These variables warrant further qualitative and quantitative exploration to understand their role in generating, maintaining, and potentially addressing disparities.

The emphasis on upstream factors—in CLDs and descriptions of system function—adds to the large literature suggesting that the most effective means of addressing disparities in a range of health-related behaviors and outcomes (including diet) is to target fundamental social causes of disease [31], including poverty, housing security, neighborhood stability, and community power [11, 31]. Many variables identified by participants in this study sit at the nexus between the community and societal levels of influence and the built and sociocultural environment domains.

A second insight is that participants were able to identify a high number of reinforcing loops that underpin and perpetuate disadvantage among minoritized populations and communities, as well as balancing feedbacks that may prevent changes to the system. Across the three workshops, participants identified 43 unique feedback loops related to the production and persistence of disparities. These loops align with research suggesting that health disparities, and, by extension, diet disparities, are the product of a complex system [10]. Systems thinking and participatory systems methods like group model building can complement existing public health frameworks by elucidating dynamics that are sometimes overlooked but that are critical for understanding the production, persistence, and durability of health disparities.

A third insight is that both contemporary and historical examples of several of the causal connections and feedback loops identified by participants can be identified in Philadelphia. Several feedback loops described mechanisms via which specific forms of community power and inequities in power are interconnected with access to affordable, healthy, and culturally meaningful food. For example, this includes loops related to community land use and control (i.e., R5 in the synthesis model), racism (R4), and community investment and empowerment (R5, R7). Recent work by Gripper et al. (2022) illustrates a similar dynamic, in which many Philadelphia neighborhoods experienced economic devastation in the 1950s and 1990s because of population decline, White families fleeing to suburbs, and businesses leaving the area [32]. Black and immigrant communities reclaimed vacant spaces to grow their own food and, as a result, community-based organizations emerged that were focused on urban agriculture, ultimately increasing access to fresh produce [32].

Contemporary examples can also be identified for causal connections in feedback loop R5 linking real estate development, community land use, food access, and community investment and empowerment. Research that informed the city’s recent urban agriculture plan revealed that over 140 community gardens in Philadelphia, most of which were developed organically by local urban gardeners on abandoned or blighted property, were lost due to demolition by land owners, redevelopment, or other factors [33]. Similarly, an estimated one-third of the remaining active community gardens are threatened by gentrification, particularly those on tax-delinquent land [33]. Communities’ repurposing of vacant and blighted properties for community gardens exemplifies a link in R5, positing that community land use and control can improve access to affordable, healthy, and culturally meaningful foods. A contemporary example of the next relationship in R5—that improving food access can increase community investment and empowerment—is the role that local restaurants and food markets have played in opposing the development of a proposed professional basketball arena in Philadelphia’s Chinatown neighborhood [34, 35]. Community coalitions used a survey of 100 Chinatown businesses—90 of which oppose the new construction—in lobbying and political organizing efforts to oppose the proposal [35]. This political organizing also exemplifies the final posited relationship in R5: that empowered communities can act as a check on real estate development.

Though the current study is grounded in the Philadelphia context, many of the study’s insights are relevant to other urban contexts. For example, there are several overlaps between our findings and those of Freedman et al. (2022), who explored nutrition equity in Cleveland. Both studies identified variables and feedback structures related to the role that community land use plays in ensuring access to affordable and appropriate food, as well as social and political dynamics that impact communities’ fair access to land. Similarly, both studies emphasized the importance of community investment and empowerment to advance racial equity, including by creating well-paying jobs in communities and ensuring access to healthy food retail. Both studies also included feedback loops via which increases in communities’ political power and policy engagement can lead to food policies that advance equity. These similarities may represent common underlying structures that can be leveraged in efforts to build and advance food system equity across varying urban contexts.

This study has several important strengths and some limitations. First, we used an established method—group model building—to engage a broad range of policy, research, and community stakeholders to define the systems that drive disparities in food access and diet between Philadelphia neighborhoods, as well as to identify systems-informed action ideas to address disparities. The synthesis CLD includes variables that are generally consistent with those found in other disparities frameworks; the main innovation, however, is that the CLD emphasizes the interrelationships between these variables and the feedback loops that perpetuate disparities. Another strength of the study is our approach to developing the synthesis CLD, particularly via the combination of thematic cluster analysis, degree centrality computation, and RIQ methods [28, 29]. This triangulation approach helped us to identify and confirm elements (i.e., variables, relationships, feedback loops) across the multiple CLDs produced in the three workshops and to ensure consistency with participants’ perspectives.

A potential limitation of the study is that we did not attempt to recruit a probabilistic sample of stakeholders in the Philadelphia food system, and as with any community-engaged research, findings reflect the unique perspectives and positionality of participants. Furthermore, we did not attempt to “empirically verify” causal links posited by participants (e.g., by identifying studies to support and quantify the causal influence of X on Y). This general limitation notwithstanding, many of the upstream variables that were most central to the synthesis CLD (e.g., those related to racism, community empowerment, income) are clearly highly relevant to Philadelphia and other cities.

Another potential limitation is that the synthesis CLD was developed by the research team and not directly verified by participants post-synthesis (though it was shared with them). This may have resulted in deviations from participants’ perspectives. However, these risks were mitigated through our use of content analysis of notes and transcripts, degree centrality, and RIQ methods in the development of the CLD. An alternative approach would be to design a follow-up workshop with return participants to directly develop the synthesis CLD or to review, consider, and refine the synthesis CLD. Such an approach, however, would increase participant burden and could also omit the perspectives of participants unable to return for a synthesis activity.

Stakeholder-engaged systems thinking methods like group model building are useful for creating a shared understanding of the systems that produce and reinforce inequities in health. Findings from this study, driven by local experts, suggest that effectively addressing disparities in food access and diet will require not only targeting upstream social determinants, but also recognizing and disrupting the relationships and feedback loops that connect upstream factors, reinforce disparities and place minoritized neighborhoods and communities at ongoing disadvantage, and limit the effectiveness of policies and interventions targeting a single factor. Promising policies include those that empower communities, provide communities with mechanisms to retain and use land and other assets for their own benefit, and disrupt the multiple, reinforcing mechanisms via which racism creates and sustains health and social disparities.