Transforming ConditionsCausal Mapping: A Dynamic Hypothesis on the Problem of Outside Assistance43
Whereas the preceding section recounted an effort to understand better the dynamics of a single disease, the following example uses a hypothetical situation to illustrate how causal mapping and simulation modeling may support much broader, syndemic thinking. Imagine a neighborhood that, like too many in the United States and around the world, is struggling against adverse living conditions (e.g., poverty, crime, gang violence, substandard housing, joblessness, proximity to a toxic waste site). Not surprisingly, many residents report a high number of unhealthy days per month, citing a long list of intertwined afflictions that are poorly managed and rarely prevented (e.g., asthma, cancer, diabetes, HIV/AIDS, substance abuse, depression, violence, and more). Leaders in the neighborhood are trying valiantly to keep things from getting worse, but their people and resources are badly organized and collectively they lack the necessary power or public strength to effect change. A local philanthropic organization with ties to the neighborhood supports the residents’ struggle and wants to help. Their assistance can take one of three broad forms: (1) enhancing efforts to respond one by one to the most prevalent or burdensome afflictions; (2) improving the adverse living conditions that leave people vulnerable to one or more of those afflictions; or (3) building greater public strength so that the residents have a greater capacity to work across their differences in pursuit of better health for all. These three forms of assistance can be provided individually, in combination, or in sequence over a period of 12 years and decisions about the combination and/or sequence of assistance can be revised every four years. Everyone agrees on the goal of increasing the number of healthy days (i.e., reducing the overall burden of affliction), recognizing that adverse living conditions and public strength are closely linked to that goal and may undermine it if left unaddressed. But they are puzzled about how best to configure and allocate the philanthropy’s assistance. What strategy makes the most sense in this situation? After several unproductive meetings, the neighborhood residents and their allies jointly decide to embark on an SD modeling project. The dynamic nature of their problem is clear enough: a veritable perfect storm involving the convergence of entrenched adverse living conditions, multiple afflictions, and low levels of public strength. Many of the factors that make the situation resistant to change are also familiar: insidious cross-impacts among the afflictions; a high fraction of people at risk; widening social disparities; divided rather than united efforts, where only doctors, social workers, and health professionals are thought to have a meaningful role to play. With subtle but steady facilitation from an experienced SD modeler, the group gathers their insights, along with information gleaned from interviews and available literature, and eventually articulates a dynamic hypothesis about how the neighborhood health system functions and where each type of outside assistance fits in. After developing several iterations of their hypothesis, the group settles on a set of 13 feedback loops (Figure 20), defined as follows.44 Figure 20: Dynamic Hypothesis for the Problem of Outside Assistance
Satisfied that these hypotheses capture the essence of their dilemma, the team presses on to put these causal relationships into the testable form of a simulation model. This entails using the best information available to assign numerical values and functional forms that indicate how these forces relate to each other. Reference data were unavailable or inadequate to inform many parts of the model (such as past and current levels of public strength), however, the group was able to make reasoned assumptions and later conduct sensitivity tests to examine the significance of uncertain parameters. In this way, the model was initialized to depict a relatively weak neighborhood vulnerable to the high affliction prevalence typical of a syndemic. The initial “basic setting” assumed that the particular cluster of afflictions includes strong cross-impacts, the baseline prevalence of adverse living conditions is relatively high, and the baseline public strength is relatively low.45 Figure 21 presents simulation results, depicting the growth of affliction burden (i.e., unhealthy days per person per year) over a 20-year period for each of four scenarios. In each scenario, affliction prevalence was set at time 0 to a value of 20%, which corresponds to an affliction burden of 6.0: the nationwide average in 2001. This initial condition represents the health status of the population prior to the development of a syndemic, or perhaps describes that portion of the population that is new to the neighborhood in question. Figure 21: Simulating the Development of a Syndemic–Four Scenarios
Over a period of 20 years, affliction burden under the model’s basic setting (the blue line) grows and finally settles at an affliction burden of 10, which is quite high for an entire neighborhood.46 During these two decades, both the reinforcing and the balancing loops described above are active, but in these scenarios no outside assistance is provided. The result is a pattern of growth that is most rapid initially but then decelerates and converges to a steady-state value. With the increase in affliction comes greater social disparity and, consequently, some erosion in public strength (not shown). Although this erosion does weaken the problem-fighting loops somewhat, the effect is gradual and does not result in explosive growth in affliction. In the three other scenarios, one or another of the “pessimistic” assumptions of the basic setting is relaxed, and the result is less growth in affliction burden. These are the assumptions described above regarding affliction cross-impacts, baseline adverse living conditions prevalence, and baseline public strength. The results give some indication of how important each assumption is to determining the steady-state level of affliction in the model. The impact of adverse living conditions on vulnerability is perhaps the most important (see the green line). Also of great importance is the affliction cross-impact effect (see the red line). Of somewhat less importance in the model, though still significant, is the effect of public strength on problem fighting (see the gray line). One reason that public strength is not quite as important as the other factors is that some professional efforts to fight individual afflictions can be undertaken, with limited citizen involvement, even in a weaker neighborhood with fewer organizational resources. Although the model is exploratory and imperfect in many respects (as all useful models must be), the team is convinced that it behaves sensibly and, therefore, can support their thinking about how to devise an optimal assistance scheme. The model includes the three available types of outside assistance—support for fighting afflictions (AF), improving adverse living conditions (LC), and building public strength (PS). In the real world, resources for assistance may take many forms (i.e., money, information, personnel, material) but they are nevertheless limited in amount and duration, so priorities are needed. The model does not specify the size of the budget in dollars, but instead describes outside assistance as a total pie of 100% that must be divided among the AF, LC, and PS types. Model parameters specify the cost-effectiveness (broadly speaking) of each type of assistance in terms of its per-unit ability to boost effective response efforts in the neighborhood. The model is set up so that assistance may be provided for a total of 12 years (T0 to T12), and the decision about how to allocate assistance may be made and revised at three specific times: at T0, at T4, and at T8, to reflect typical grantmaking cycles. The group’s ultimate goal when experimenting with the model was to minimize the average affliction burden, both during the 12 years of assistance and for some time following its conclusion. An evaluation period starting at T4 and ending at T20, with outside assistance ending at T12, allowed team members to look symmetrically at eight years during which assistance is active as well as eight years of the post-assistance period. As with other aspects of the modeling, the choice of this evaluation period was guided by common sense and remained open to change, for it is possible that moving the evaluation start time or end time could affect the results, as discussed below. Tests under the basic setting revealed an optimal assistance scheme to be one that starts with 100% public strength (PS) assistance at T0, then switches to 100% affliction-fighting (AF) assistance thereafter (at T4 and at T8), with no assistance to improve living conditions (LC).47 This scheme is labeled “PS1,AF1,1”, and its results are presented in Figure 22. Figure 22: Results under Basic Setting with Optimal Assistance Scheme.
The initial PS assistance builds public strength for the first four years, thereby strengthening the citizens’ collective capacity for fighting afflictions and adverse living conditions, and ensuring that subsequent problem fighting will be more unified and do less to undermine public strength. The switch to AF assistance at T4 greatly boosts the affliction-fighting programs, and the affliction burden is reduced dramatically over the next eight years. However, after the assistance concludes at T12, the affliction burden rebounds significantly. The magnitude of this rebound is related to the fact that public strength gradually erodes after the end of PS assistance at T4, so that by T12 the neighborhood’s internal capacity to organize effective affliction fighting efforts is not as great as it would have been had the PS assistance continued. Figure 23 compares the optimal PS1,AF1,1 scheme under the basic setting with three other assistance schemes, in terms of their impacts on affliction burden.
The finding that PS111 is superior to LC111 after T4 may at first seem to contradict the previous finding from Figure 21 that better living conditions do more to reduce the growth of affliction than greater public strength does. But in those alternative growth scenarios we assumed better living conditions or public strength from the start, reflecting enduring qualities of the neighborhood. In contrast, in Figure 23 the improvement is caused by outside assistance, a process that has some negative side effects.48 These side effects undermine public strength to some extent, and therefore hinder local problem-fighting efforts. As a result, LC assistance fails to make as much improvement in living conditions and affliction burden as one might expect based on Figure 21 or the first four years of Figure 23. Figure 23: Comparison of Affliction Burden under Basic Setting for Four Different Assistance Schemes
The fact there are allocation schemes superior to PS1,AF1,1 early and late in the simulation (AF111 is superior prior to T6, and PS111 is superior after T14) suggests that perhaps the optimal allocation scheme could be different for evaluation periods other than T4 to T20. Further model testing indicates that changing the evaluation start time to T0 does not affect the choice of optimal scheme, but extending the evaluation end time can change the optimal scheme to one that puts more emphasis on PS assistance. If the evaluation end time is extended to something in the range of T21 to T26, the optimal scheme becomes “PS11,AF1”, with 100% PS assistance at T0 and T4, then AF assistance at T8. If the evaluation end time is T27 or later, the optimal scheme becomes PS111, with 100% PS assistance throughout the 12 year period of assistance. Given fundamentally weak democratic institutions in the neighborhood, the longer the period of post-assistance evaluation is, the more priority one must give to public strength during the period of assistance, so as to minimize the post-assistance rebound in affliction. Even with an exploratory model, not yet verified and refined through case study application, it was possible for the team members to better appreciate the dynamic impacts of various types of outside assistance on population health. One insight, familiar to seasoned leaders in the neighborhood, is that the first priority of philanthropies or government wanting to help neighborhoods that are weak and struggling against multiple afflictions should be to assist in building public strength (enabling a greater degree of citizen-led public work), perhaps even before substantial assistance is provided for direct fighting of prevalent diseases. Another insight suggested by the modeling is that outside assistance aimed directly at improving living conditions may often be insufficiently cost-effective, due to time lags and unintended side effects, to warrant making such assistance a high priority in the absence of widespread citizen participation—this despite the fact that adverse living conditions are a powerful determinant of vulnerability to affliction. The power of a useful simulation model lies not only in the identification of hypotheses for optimal decision making, but also in the ability it provides to explain how those hypotheses emerge logically from a feedback structure that integrates the best available knowledge on the subject. The insights described above, for example, reflect the presence in the model of relationships depicting the perverse effect that problem-fighting programs may have on public strength when the democratic institutions are weak and people are divided to begin with—exactly the opposite of what the philanthropists intended.
43. Material for this section comes from an ongoing collaboration with system dynamics expert Dr. Jack Homer. The text was drawn primarily from a conference presentation written jointly by Dr. Homer and myself (Homer and Milstein, 2004). Additional details about this line of inquiry may be found in (Homer and Milstein, 2002a, 2002b, 2003b, 2004; Milstein, 2006). 44. The relationships hypothesized in Figure 20 are based on a synthesis of public health literature as well as participant observations from myself and colleagues regarding the dynamics of neighborhood-based health improvement ventures. It is a general theory, meant to fit any neighborhood and any cluster of afflictions. The model has not yet been applied to any particular circumstance, and for this reason should be considered exploratory and suggestive, not a model that is fully tested and determined to be reliable for decision making in specific situations. 45. For precise figures describing how these and other parameters were set, see (Homer and Milstein, 2004). 46. The CDC’s Healthy Days survey (Centers for Disease Control and Prevention, 2003a; Zahran, Kobau, Moriarty, et.al., 2005) asks individuals to describe their overall health as excellent, good, fair, or poor, and then to estimate their number of unhealthy days per month. In the 2001 survey, 15% of the 200,000 surveyed described their health as fair or poor, with an average of 15.7 unhealthy days per month, while 85% described their health as excellent or good, with an average of 4.3 unhealthy days per month. The overall average of 6.0 thus disguises a very skewed distribution of unhealthy days. For a neighborhood to have an overall average of 10—still assuming that 15.7 represents fair or poor, while 4.3 represents excellent or good—the fraction reporting fair or poor would have to be 50%, much greater than the national average of 15%. 47. It must be emphasized again that even with thorough sensitivity testing, an exploratory model such as this is prone to some misspecification, which could affect the results. Any results discussed here must therefore be taken as suggestive rather than prescriptive. 48. The reinforcing R3 loops of Figure 20 can undermine public strength when LC assistance is overlaid onto a situation in which public strength is relatively low to begin with. Greater outside assistance to improve living conditions means more effort to improve living conditions, thus greater magnitude of ameliorative efforts, and thus—in the context of low public strength and low public work fraction—more divided efforts, which are a further drag on public strength.
Page last reviewed: January 30, 2008 Content source: Division of Adult and Community Health, National Center for Chronic Disease Prevention and Health Promotion |
