In the workshop’s final session, its organizers shared their reflections from the workshop presentations and discussions. Each individual provided brief remarks on a variety of issues focused on encouraging and helping to support the use of systems science thinking and systems science approaches to address the complexity of obesity drivers and solutions. Following the organizers’ remarks and a brief panel question-and-answer session, a closing speaker delivered final reflections on the workshop’s discussions. Nicolaas (Nico) Pronk, president of HealthPartners Institute, chief science officer at HealthPartners, Inc., and adjunct professor of social and behavioral sciences at the Harvard T.H. Chan School of Public Health, and Christina Economos, co-founder and director of ChildObesity180 and professor and New Balance Chair in Childhood Nutrition at the Tufts University Friedman School of Nutrition Science and Policy, moderated the session.
REFLECTIONS AND DISCUSSION OF NEXT STEPS
Douglas Luke, Irving Louis Horowitz Professor in Social Policy and director of the Center for Public Health Systems Science at Washington University in St. Louis, commented on strategies for translating systems science modeling results and disseminating resources and tools to community partners. He outlined three points, attributing the first to his friend and colleague Ross Brownson. One of Brownson’s key tenets, he noted, is to encourage the conduct of dissemination planning at the outset of a project instead of waiting until its completion. Luke highlighted the importance of
ongoing dialogue with community partners about what type of resources they need to support their decision making.
Second, Luke pointed to several translational challenges or steps entailed in systems science modeling efforts. Modelers first translate for a scientific audience, he explained, as they consider how to convince epidemiologists or health economists, for example, about the reliability, validity, and utility of a model. The second translational step, he observed, occurs when modelers translate their learnings into data and wisdom that community planners and stakeholders can use, and he stressed that the tools and resources that are most helpful in this step are different from the typical outputs of a National Institutes of Health (NIH)-funded study. Luke appealed for highly visual resources that are tailored to the extent possible, commenting that policy makers are more likely to be engaged when they see a model’s projections for their specific area of concern.
Third, Luke emphasized tools that are interactive, dynamic, and explorable. As an example of such a tool, he highlighted Lee’s map of Baltimore (see Chapter 3), where the agents could be seen moving around. He also highlighted interactive dashboards that visually present a variety of information, maps, or other interfaces that show a physical area of interest; before-and-after presentations that display how an environment might look postintervention; and resources that allow people to explore what-if scenarios.
Sara Czaja, professor of gerontology in medicine at Weill Cornell Medicine, spoke about the importance of leveraging partnerships and multidisciplinary teams to address complex challenges. Given that systems science approaches involve the interaction of multisector stakeholders, she urged early planning for how the layers of participants will communicate effectively despite coming from different arenas. In terms of training and education, she continued, it is important for participants in systems science approaches to understand the roles and contributions of the various players, as well as their goals for the targeted effort.
Czaja also encouraged understanding and managing stakeholder expectations. For example, she stressed that it is important for systems science modelers to determine how community stakeholders’ input will be used and to understand how this determination aligns with those stakeholders’ expectations regarding the implementation of their inputs. She noted that, while it is not always feasible or even possible to integrate every input into a model, it helps if people understand that from the start.
Lastly, Czaja urged consideration of the importance of keeping people informed about progress and motivated, especially when efforts are complex and lengthy. She emphasized that people want to understand how they are making an important contribution to the process. Communication is key, she insisted.
David Mendez, associate professor in the Department of Health Management and Policy at the University of Michigan School of Public Health, assessed the utility of systems science modeling. The utility of these models, he postulated, depends on one’s expectations for them. Systems science modeling helps solve a specific problem or answer a question, he elaborated, and modelers can incorporate various levels of complexity into a model to answer a specific question. The model becomes a representation of an idea, he continued, a hypothesis about the processes that govern the topic of interest. If one has a relatively narrow problem to address, he observed, the corresponding mental map and systems science model may not be as complex as one that shows all of the interconnections among the greater systems at play.
Mendez shared a challenge of his work in evaluating tobacco policy—that the impacts of policies of interest take a long time to materialize. Realizing that a model can help evaluate the range of time in which such policies will be successful, his team developed a model to gauge the overall mortality caused by smoking over time. By better understanding the magnitude of the problem, Mendez explained, the team was able to derive a set of robust policies for addressing that problem. He argued that even though unintended consequences are likely, it is important to use the best information available to make policy decisions now. He suggested the use of comparative models—different models for the same topic—to assess how they differ in their results, explaining that this approach can help inform a level of confidence in the potential outcomes derived from the modeling.
Jamy Ard, professor in the Department of Epidemiology and Prevention and the Department of Medicine at the Wake Forest University Baptist Medical Center and a member of the 2020–2025 Dietary Guidelines Advisory Committee, shared his perspective on the inclusion of systems thinking and modeling in the development of the Dietary Guidelines for Americans and on how using systems science approaches at the micro level affects his daily work as a clinician.
Food, nutrition, and the consequences of dietary intake for health are interrelated, Ard remarked, and manipulating one variable leads to a series of consequences that affect nutritional status and health trajectory in the context of disparities in obesity and other chronic diseases. Therefore, he maintained, the process of developing the Dietary Guidelines could naturally be all about systems. He observed, however, that the process of developing the Dietary Guidelines has historically taken a reductionist approach, whereby until recently, the recommendations have been based more on nutrients than on foods or dietary patterns. The challenge going forward, Ard suggested, is to integrate systems science approaches into the development of dietary guidance at both the population and individual levels. The 2020–2025 Dietary Guidelines Advisory Committee took a
life-stage approach, he recounted, as it considered the science informing dietary guidance. Much more can be done, he argued, with regard to integrating a systems science approach into the nutrition science evidence base and the process of developing dietary recommendations to reduce the risk of chronic disease.
With regard to his role as a medical provider, Ard proposed that health care providers would benefit from training in systems thinking to help shape their perspectives on the context in which their prescribed interventions occur. He suggested that, upon learning about the web of systems influences that affect patients, clinicians might feel frustrated that their treatments or counseling will accomplish little because they cannot control the host of environmental and social influences that affect their patients’ health behaviors. Nonetheless, he appealed for helping providers understand how even at the individual level they can play an influencer role within systems, such as by helping to link patients to community supports.
Stella Yi discussed the formative work involved in engaging stakeholders and community members in participatory group model building. According to Yi, the participatory systems science process is a tool for providing structure around collaboration that is aimed at addressing shared priorities. Speaking from her perspective as a health disparities researcher who has worked in partnership with Asian American communities and community-based organizations for more than 15 years, she noted that her group’s long-term relationships with communities have enabled much dialogue about issues that are important to residents. These conversations served as a precursor, she said, to her team’s current efforts to engage communities in group model-building exercises to learn about healthy diet and lifestyle behaviors among older Chinese and Mexican immigrant communities in Brooklyn.
Yi pointed out to participants that although it takes time and effort to foster the connections that lead to long-term, sustainable change in communities, they likely already have relationships with the communities that they seek to help. If not, she suggested that relationship building can start with an initial conversation and progress organically as the parties discuss mutual interests in a particular population or topic area. She acknowledged that this may sound simplistic, but argued that it is truly how relationships begin, and she noted that this model has worked well to enable her group to engage with communities that are underrepresented in the health literature and policy space. Different stakeholders have unique contributions to offer, Yi stressed, sharing an example in which her university provided data collection tools to community-based organizations that wanted to collect data but lacked the capacity to do so. At the same time, however, the organizations brought the essential assets of the trust of the community members, cultural and linguistic knowledge, and an understanding of how programs can be designed to be maximally successful for those they are designed to
help. To end her remarks, Yi appealed for solutions that incorporate voices from diverse age, racial, and ethnic groups that mirror the makeup of the target population.
Daniel Rivera, professor of chemical engineering and program director of the Control Systems Engineering Laboratory at Arizona State University, offered suggestions for initiating community and stakeholder engagement in the systems science modeling process. These suggestions, he noted, were based on his viewpoint as a chemical engineer who works as a modeler in partnership with intervention and behavioral scientists to deliver optimized, personalized interventions related to behavioral medicine. Rivera maintained that, regardless of an intervention’s outcomes of interest, issues of language (jargon) and domain expertise persist. Some degree of tension also typically exists, he added, and he shared a personal example in which his scientific understanding of the definition of an experiment had been broadened by working with behavioral scientists.
Rivera highlighted the usefulness of guiding groups to draw path diagrams, which he described as an iterative process that ultimately results in the development of systems science dynamical models that are used for intervention and optimization. Some theories of behavior are amenable to a dynamical systems interpretation, he continued, giving the example of an effort that developed a dynamical systems science model for social cognitive theory. Good path diagrams for this theory did not exist in the literature, he said, so his team developed a system dynamics model with a fluid analogy that has similarities to the stock-and-flow models used in system dynamics. The fluid model can be correspondingly expressed as a path diagram, he reported, which, thanks to his team’s efforts, is now available for social cognitive theory.
Rivera echoed Jack Homer’s comments from earlier in the workshop (see Chapter 2) about the balance between testable models and models that are “good enough.” He pointed out that the goal is often to get to the simplest model that helps answer basic questions, and suggested that this goal is better than integrating every possible construct and contextual variable that could influence the outcome of interest.
Bruce Y. Lee, professor of health policy and management at the City University of New York Graduate School of Public Health and Health Policy and executive director of Public Health Informatics, Computational, and Operations Research, reviewed the purposes and capabilities of models and addressed the misconception that all models are the same, so only one model is needed for a given situation. He referenced the public conception that models for predicting COVID-19 mortality counts were “wrong” because they did not predict exactly what occurred, arguing that predictions represent only a sliver of a model’s capability. Predicting the future is difficult, he pointed out, and he described the process of building a budget
spreadsheet as an analogy for a model’s inputs and purposes. One would not use a budget spreadsheet to predict how much money would be earned in 1 year, he stated, because too many variables are involved, whereas a budget spreadsheet would be useful to help conceptualize one’s financial context and better understand the relative contribution of each line item to total spending. This latter use might involve checking receipts, he continued, which is analogous to data collection, and as more data are added, the budget model is revised. Moreover, he observed, an accountant or family members might be asked to weigh in, which he described as an example of modeling bringing people together to collect their inputs or inform them about what the model describes, ultimately refining the shared understanding of the situation and integrating other research methods and data.
To address a common misconception of stakeholders new to the modeling process that all systems science models are the same, Lee recalled cases in which people claimed that a model was not needed to address a particular question because they had already developed a model on that topic. He pointed out, however, that this would not be the way one would respond about a clinical trial; rather, more clinical trials would be conducted to examine similar outcomes in different populations and contexts. He emphasized that wide diversity exists within systems science approaches, and modelers sometimes use the same technique in different ways. Lee suggested that a problem or question would benefit from assessment by several modeling teams, with the results then being compared. Similarly, he observed, people’s conceptualizations of a given problem or issue are diverse. He argued that this diversity can be explored by perspectives of a range of stakeholders, which can help modelers understand differences in the understanding of or access to information about a problem.
Lee concluded his remarks by proposing a change in the paradigm of how systems science modeling is used, asserting that most stakeholders harness only a small percentage of modeling capabilities.
PANEL AND AUDIENCE DISCUSSION
Following the reflections summarized above, the workshop organizers discussed early-life opportunities for introducing systems thinking, systems science models of the effects of taxing sugar-sweetened beverages, strategies for enabling novices to get started with modeling, and potential unintended consequences of defining obesity in systems science models.
Training in Systems Thinking: How Early?
Ard suggested that it is never too early to introduce systems thinking concepts, clarifying that such conversations do not necessarily have to use
the term “systems thinking.” Rather, he argued, they can simply convey the interconnectedness of the world, the effects of one’s actions on others, and the effects on systems of actions at the population level. Yi mentioned the children’s book When a Butterfly Sneezes, which helps learners recognize the systems nature of such common objects as a tree or the human body and describes the downstream effects of something as simple as the flap of a butterfly’s wings. She noted that although the jargon of systems scientists can be intimidating, it describes a concept that is innately understood by the human brain. Luke suggested encouraging students to take a systems perspective on the world as they progress through school, and Pronk added that it may be helpful to call out systems thinking explicitly so that students recognize it in the context of their training and education. According to Lee, it is important for training to include an explanation of methods and their purpose so that people understand appropriate applications. Yi pointed out that much systems science modeling occurs in a vacuum, but that actual implementation of modeled interventions involves additional, on-the-ground factors and participation that result in decisions regarding the differences between efficacy and effectiveness. She stressed that the participation of multisector stakeholders can help make models more applicable to the real world with potentially enhanced sustainability and effectiveness. Czaja added that exposing people to other disciplines and ways of thinking can help prevent the formation of disciplinary siloes.
Mendez distinguished between systems thinking and systems science modeling. He agreed that early integration of systems thinking in education is important so that people understand how inputs flow into outcomes, as well as the concept of interconnectivity. With regard to the latter, he stressed the importance of recognizing that a problem is likely affected by a host of environmental determinants that may not all appear to be directly related to it. He suggested that moving from systems thinking to systems science modeling requires more formal, specialized training.
Systems Science Models of Taxing Sugar-Sweetened Beverages
Lee responded to a participant’s question about the existence of systems science models examining the effect of taxing sugar-sweetened beverages on their consumption and ultimately on the incidence of obesity. He confirmed that such models exist, noting that the Global Obesity Prevention Center funded a pilot project at the University of California, Berkeley, to collect data on these taxes in various Bay Area municipalities. The goal was to use the data to help calibrate and also validate different models for predicting the outcomes of implementing these taxes, he explained, acknowledging that the issue is complicated because of the heavy media attention and public interest that these policy proposals tend to generate.
Getting Started with Systems Science Modeling
In response to a participant’s question about how a novice can begin applying systems science modeling, Luke highlighted the value of social networking for identifying local experts who can help. Lee suggested a “happy hour approach,” which involves trying to identify someone with crossover interests or experience working in the same discipline who can help bridge the gap between one’s interests and systems science modeling.
Potential Unintended Consequences of Defining Obesity in Systems Science Models
A participant commented that some stakeholders have suggested that the use of weight status to assess health results in stigma and disordered eating, and asked whether assigning definitions for obesity in systems science models could have such unintended consequences. Lee replied that systems science models incorporate a host of factors and processes that affect the outcomes of interest, so there is rarely one factor, such as weight, that drives everything. He stated that the challenge is to ensure that the translation of a model’s results emphasizes this multifactorial complexity to communicate the full story.
FINAL REFLECTIONS
Patty Mabry, interdisciplinary scientist at HealthPartners Institute, delivered final reflections on the workshop’s discussions about applying systems science approaches to obesity solutions. She also included in her remarks additional content not covered by the speakers.
Mabry commended Homer’s chart comparing characteristics of three simulation approaches for systems science modeling (see Figure 2-1 in Chapter 2), and added that an entire field of network science and its sub-fields exists within the column for microsimulation or agent-based models. She also referenced Homer’s blueprint for an ideal, well-funded project, and suggested that another important component is providing tailored education to different stakeholders about appropriate uses and expectations of a model. She reminded participants that models are not reality; they are simplified versions of reality that provide a better understanding of complex problems. When models are developed with stakeholder input, she argued, they reflect the perspectives and experiences of those stakeholders.
Mabry next recalled Lee’s presentation about the Virtual Population Obesity Prevention Labs and the synthetic population it built using census data. She referenced secondary sources of data for populating systems science models, including FIGSHARE; Research Triangle Institute’s U.S.
Synthetic Household Population; Social Science One (data from private industry sources, such as Facebook); SyntheticMass (1 million synthetic electronic health records); and All of Us Research Hub, a new NIH research program that is collecting longitudinal data from participants of all backgrounds to support efforts to reduce health disparities and improve health equity, among other purposes.
Mabry touched on sources for data on social determinants of health, highlighting the NIH PhenX Toolkit and a number of additional sources on the Centers for Disease Control and Prevention’s website (RTI International, 2020a,b). She pointed out that most data on social determinants of health are available at the aggregate level, and that individual-level measures are often not available.
Mabry then called participants’ attention to work from Kevin Hall and colleagues that involved developing and validating a quantitative mathematical model of the dynamics of childhood growth and obesity (Hall et al., 2013). She supported reusing systems science models when appropriate, pointing to the Hall team’s model as one that could be applied to other modeling efforts.
Mabry next emphasized the importance of spatial models, referencing the map of healthy food priority areas in Baltimore presented by Buzogany and SALURBAL’s spatially explicit model of urban transport and mobility policy (see Chapter 3). Such models are important for understanding inequalities in health, she maintained, because the spatial environments for food and physical activity are critical factors in these inequalities.
Systems science methods are motivated by complexity in behavioral and social science data, Mabry stated, and she detailed aspects of this complexity: temporal properties, spatial properties, network structures, hierarchical and nested structures, feedback loops, variation at the individual and group levels, mediating and moderating variables, and nonlinear and nonparametric properties. She shared a figure depicting the complex context of health disparities and obesity (see Figure 4-1), in which a continuum of biological and social factors is portrayed as extending across the lifespan.
Mabry then reiterated the value of self-reflection and self-correction for enabling progress in modeling. Humans do not always solve complex problems, she suggested, because doing so takes hard work and complex solutions. She quoted a statement by advice columnist Ann Landers: “Opportunities are usually disguised as hard work, so most people don’t recognize them.” She maintained further that people tend to pursue low-hanging fruit—easy opportunities waiting to be seized and requiring little expenditure of effort. But that is not how most complex problems work, she insisted, urging participants to “go for the hard part.” Mabry also highlighted Thinking in Systems: A Primer, a book by scientist Donella
SOURCES: Presented by Patty Mabry, September 16, 2020; adapted from Glass and McAtee, 2006.
Meadows, as a good starting point for learning about systems thinking. According to Mabry, a key portion of this book is Chapter 6, in which the author explains that the easy places to intervene in a system will offer the lowest return on investment, while the most difficult places will net the greatest return.
Next, Mabry enumerated challenges in systems modeling that make it a difficult endeavor: finding appropriate sources of data, particularly longitudinal data; maintaining stakeholder engagement (i.e., with policy makers, health care systems, or businesses); dealing with impatience to produce actionable results, particularly for policy makers who have limited time in office; building interdisciplinary teams and communicating across disciplinary boundaries; doing what one knows instead of learning what one should do (i.e., lack of self-efficacy); relying too much on randomized controlled trials as the gold standard for evidence, because this is not the best method for every research problem or question; following through on the course of action that a model indicates, particularly when doing so runs counter to one’s ideology; and dealing with the “prevention conundrum,” which Mabry described as the difficulty of proving that an intervention indicated by a model avoided an impending fate. To illustrate this latter challenge, she recounted Lee’s earlier example (see Chapter 3) of a model indicating that increasing physical activity to a certain level for at least half
of the U.S. youth population would avert billions of dollars in medical care costs and lost productivity. It is difficult to convince people to take action based on these modeling results because they did not feel the pain of paying those billions of dollars, she explained, and they will not receive billions of dollars back in cash or some other tangible form by taking such action.
Mabry moved on to the topic of recruiting systems modeling teams, detailing various types of stakeholders and experts to consider including, such as health geographers or other experts who know how to integrate spatial information into models, modelers themselves, community members, and policy makers. She drew an analogy comparing these players to bricks and another player, interdisciplinary scientists, to the mortar between the bricks, emphasizing the importance of including team members who can facilitate communication across disciplinary boundaries.
Mabry closed her presentation with a list of additional resources for those who wish to delve further into systems modeling (see Box 4-1). She also outlined potential future directions for systems modeling efforts (see Box 4-2), emphasizing the value of building a community of systems scientists. She suggested that this latter good would be enabled by more coordination and perhaps a new society dedicated to the topic, and would benefit from increased sharing and repurposing of research assets. She also highlighted the need for funding to form community-based research relationships and to educate practitioners in the use of systems science to inform research priorities.
