摘要：Dynamic modeling and simulation are systems science tools that examine behaviors and outcomes resulting from interactions among multiple system components over time. Although there are excellent examples of their application, they have not been adopted as mainstream tools in population health planning and policymaking. Impediments to their use include the legacy and ease of use of statistical approaches that produce estimates with confidence intervals, the difficulty of multidisciplinary collaboration for modeling and simulation, systems scientists’ inability to communicate effectively the added value of the tools, and low funding for population health systems science. Proposed remedies include aggregation of diverse data sets, systems science training for public health and other health professionals, changing research incentives toward collaboration, and increased funding for population health systems science projects. A recent analysis reveals that the global burden of disease in terms of global disability adjusted life years has not improved significantly over the past 2 decades. 1 The same analysis shows that there has been a shift in the global burden of disease from communicable to noncommunicable diseases and from premature death to years lived with disability. The burden of disease and the factors contributing to it vary greatly among countries and within countries, regions, and cities. Local interventions and practices are required to address the needs and underlying determinants within particular communities. Although there is a large and ever-growing body of evidence on effective population health strategies, that literature often does not systematically take into account the dynamic socioecological and cultural context of a specific community or the policies that are already in place. Moreover, resources are always limited, which frequently causes strategies that have been effective in one environment to be unaffordable or unsustainable in another. Nor does it make sense to randomly pull strategies from a grab bag of evidence-based policies and interventions. From a scientific viewpoint, conducting experiments to see what would be most effective in a particular community, especially the combined effects of different policies and their impact over time, is preferable. Yet the real world presents practical barriers that impede an experimental approach. For example, it is impossible to examine the counterfactual. That is, we cannot implement a new policy, wait to see what the results were, then go back to a time before the policy was implemented and try a different policy to see how it would stack up against the first one. However, this exact type of experiment can be done by working in a virtual environment (i.e., a computer-simulated model of the real world). In this essay, we invite readers to explore the utility of dynamic modeling and simulation approaches to addressing population health challenges and we lay out a roadmap for facilitating the uptake of these methods. By “dynamic modeling and simulation” we are referring to a variety of mathematical and computational modeling methods including system dynamics modeling, agent-based modeling, Markov modeling, discrete event modeling, microsimulation, and more. These methods share the ability to quantitatively represent, in silico , the behavior of a system and its components over time. Major population health challenges, such as cardiovascular diseases, diabetes, and cancer, are rooted in complex interactions among multiple variables, including people, genes, and social, economic, and physical environments. 2,3 In fact, the variables are often subsystems containing a whole level of complexity themselves. For example, each of the following factors can be thought of as a system within a broader interdependent system of human health: food, agriculture, transportation, education, and health care delivery systems. Their interactions can result in outcomes that are difficult to understand or predict, and that often cannot be traced reliably to the behavior of any single component. Prevalent approaches to solving complex problems often involve decomposing them into component parts and analyzing some subset individually to identify candidate interventions. For example, many proposed solutions to noncommunicable diseases in high-income countries have focused primarily on the health care delivery system. In the United States, population health challenges resulting from noncommunicable diseases are largely being addressed by increasing access to health services through government-subsidized health insurance, delivery system expansion of primary care, improvements in the quality of clinical practice, and new performance-sensitive payment mechanisms. 4 Although health care delivery system reforms are essential and may appear poised to be cost-effective, it is simply not known whether the reforms will improve population health. For example, essential components of health maintenance, such as exercise, better nutrition and weight management, medication effectiveness, and stress management, can be thwarted by outdoor public safety issues in neighborhoods, prohibitive costs of fresh fruits and vegetables, housing with smokers in indoor environments, and time demands on patients from work or elder- or child-care responsibilities. Modeling and simulation-based insight into population health scenarios resulting from multisectoral changes could improve decision-making in policy development, planning, and implementation. One way of improving decision-making is by identifying unintended consequences of a program or policy (e.g., the impact on the health sector resulting from interventions in the transportation sector). A second way is by revealing the likely dynamics of the system and how impact on variables of interest plays out over time (e.g., in some cases an intervention produces a worsening of circumstances before improvement). A third way is by exposing the individual and the combined effects of programs or policies (e.g., clean indoor air laws might prompt more people to quit smoking, and simultaneously offering medication and other cessation services might help more people to successfully quit). Dynamic modeling and simulation methods are well suited to discovering potential outcomes from significant system disruptions. Standard econometric, epidemiological, and biostatistical methods have demonstrated value in assessing effects of social determinants of health, but these methods have inherent limitations that dynamic modeling and simulations can help address. For example, they often assume unrealistic constraints on relationships among variables (i.e., statistical independence), on samples (representativeness), and on distributions (i.e., Gaussian). In addition, such approaches are not well equipped to measure nonlinear and bidirectional relationships, or relationships with delayed effects. Modeling and simulation methods may be combined with more traditional and prevalent methods to gain insights when new laws and policies create real-world experiments. For example, these methods can help disclose interactions among many parties likely to be affected by changing payment schemes in different types of health care models, 5 such as health organizations that assume risk for cost and clinical outcomes (so-called “accountable care organizations”). Modeling and simulation could be used to identify optimum payment approaches for accountable care organizations and other care delivery systems by using models for patient, provider, supplier, and payer behavior, along with models for key medical conditions such as diabetes. 6,7 The need for improved intelligence in complex systems is also increasing in cities and states. Plagued by the impact of increasing health care costs on state and municipal budgets, decreasing family incomes, and increasing labor cost for employers, state and local governments are seeking to mitigate causes of poor health in schools, neighborhoods, and housing, and from food sources and pollution. 8–10 In addressing upstream determinants of health, city officials have learned to take account of system dynamics to understand potential outcomes of interventions. For example, the impact of changes to the nutritional content of school food sources is dependent on the price and availability of low-value food sources and snacks in surrounding food outlets. These outlets in turn are incentivized to stock packaged low-value foods because of their lower cost per unit of space compared with fresh fruits and vegetables. 11 Cities and states are interested in better understanding city systems dynamics affecting health to identify leverage points for improved efficiency and effectiveness. 12 Dynamic modeling and simulation are essential tools to achieve these objectives. Complex system problems in health will often require cross-sector solutions. Modeling and simulation can help decision-making for such problems by capturing salient aspects of a problem, using data of different types and from multiple domains, handling the computationally intensive task of calculating the impact of multiple interacting relationships, and playing out the potential outcomes in policy scenarios or potential interventions. Modeling and simulation can also help examine trade-offs between policy options and assess the time horizon over which policies may have an impact. 13 The current state of modeling and simulation is such that they are most useful for computationally intensive tasks that far exceed human capabilities. However, models are not good at nuance and judgment. Rather, models are best used to help inform decision-making, leaving the actual decisions to people.
