Biomedical Engineering Reference
In-Depth Information
4.
DISCUSSION AND CONCLUSIONS
The combined effect of structured social networks and disease-reactive con-
tact dynamics is to force smoothly growing epidemics into unstable kinetic re-
gimes where they can be easily extinguished. Sociospatial sparseness interacts
with temporal sparseness to generate decisive qualitative boundaries from the
mass action of individually quantitative behaviors. The emergent social immune
response need not involve any central coordination, but can easily arise from the
widespread distribution of comparatively simple rules regarding the behavior of
sick individuals. The computational simplicity of these rules and their aggre-
gated survival consequences provide considerable pressure for the evolution of
molecular crosstalk between the nervous and immune systems. Indeed, the pre-
sent analyses suggest that sickness behavior has coevolved in concert with the
biological immune response to provide a synergistic set of defenses against in-
fectious pathogens. The results of these analyses also suggest that our biological
immune system might have become considerably more menacing had it not de-
veloped a sociobehavioral ally.
The present studies examine "sickness behavior" in the context of agent-
based models, but illness-reactive behavior can also be analyzed in more tradi-
tional algebraic simulations by varying the contact rates that mix susceptible and
infected individuals (4). Such variations can damp epidemics that would other-
wise oscillate, shift the basal prevalence of disease, or kick stable epidemics into
truly chaotic behavior, all depending upon the exact specification of the feed-
back function (nonlinear? time-lagged?) and whether or not it affects other pa-
rameters aside from the mixing rate (e.g., whether falling contact rates also
reduce host resistance). However, agent-based analyses have several advantages
over algebraic models in forecasting epidemic trajectories, particularly in terms
of realistic confidence bounds. First, complex real-world social structures are
more easily encoded in the explicit interaction matrices shown in Figure 2 than
they are in analytically tractable continuous functions that modulate population-
wide mixing rates. This is especially helpful in assessing the impact of small
behavior changes generated in reaction to locally available information. Agent-
based models also provide an opportunity to analyze network-mediated distribu-
tion of recursive operators that reshape individual behavior, host-pathogen dy-
namics, or linkage matrices depending on the realized course of an epidemic
(e.g., dispersing and reconstituting groups). A third advantage is the natural dis-
creteness of agent-based models in regions of spatiotemporal sparseness. As
noted above, this is key to understanding the epidemic-extinguishing behavior of
dynamic host networks, and natural discretization reduces the likelihood that
minor, seemingly ignorable boundary conditions will propagate into large pre-
diction errors. Figure 7D shows a prototypic example—an epidemic simultane-
ously subject to all of the influences considered above, including a complex
clustered social structure with a small number of intergroup contacts, behavioral
