The problems which concern epidemiologists today, and
those anticipated to emerge in the future, encompass a wide spectrum of
issues from genomics to the health impacts of global environment
changes. However, amongst the many factors discussed, a recent paper on
the ?Future of Epidemiology? (Ness em, 2009) does not identify the role
of statistics as important in underpinning the ability of epidemiology
to respond to the increasing complex modelling challenges facing the
field. Arguably, there is a growing need to extend traditional
epidemiological paradigms and adopt a fully integrated systems approach
to address the interactions between disease processes and the physical,
ecological, environmental and social systems that may contribute to
future disease burden. This emerging interdisciplinary research
challenge requires understanding the behaviour of systems of complex,
dynamic and uncertain interactions involving hierarchies of information
at different spatial and temporal resolution. The modelling of such
complex hierarchical systems and the quantification of uncertainty and
risk within them requires exploitation of new advances in statistics.
However, the gap between the ongoing developments in statistical
methodology and modelling and the capacity of epidemiologists to
recognise, understand and exploit the growing power of such
technologies has increased very quickly in recent years. At the same
time, there is always a clear need to appreciate the limitations of
seductively ?bright new? statistical methodology for its own sake and
for epidemiology and public health to remain true to its roots in the
substantive practical objective of improving the health of populations.
In this paper, we illustrate some of these issues and the complexity
faced by contemporary epidemiological studies in the context of the
epidemiological surveillance of dengue fever mosquitoes in one area of
Brazil. In order to detect hot spots of mosquitoes two different traps
were tested in three neighbourhoods in Rio de Janeiro, one designed to
capture adult females and the other one the eggs. Both were evenly
distributed over the area, in different dwellings. Adults and eggs were
counted weekly, for 78 weeks. Firstly, the time trends over each area
were analysed, with differing results among neighbourhoods, no time
structure in one of the areas and the expected increase in mosquito
density during summer in other places, with similar patterns for both
types of traps. Observations were then aggregated over time and
investigated using a generalised additive model (GAM): the spatial
patterns of each trap were entirely different in all areas, in spite of
the significance of the spatial term in some of the models.
Subsequently a latent Gaussian Markov random field model was applied,
but no spatial, temporal or spatio-temporal structured effects were
detected. Only unstructured random terms both for space and for time
were apparent for both types of traps. These analyses present public
health professionals with some problems:
? Is the absence of any spatial effects due to: too large a distance
between traps to detect spatial dependence (each neighbourhood was only
0.25km2 with 80 traps evenly distributed), or lack of power of the
model, or the wrong model?
? Considering that in the GAM model the adult trap presented some spatial structure, how to interpret
it? Should we do that?
? Which type of trap should the Brazilian Health Ministry indicate to do the entomological surveillance?
(the cost of the egg trap is one tenth of that for the adult females).
This simple illustration emphasises that incorporation of increasingly sophisticated statistical methodology
into epidemiology should be seriously addressed in order to avoid on the one hand, a severe scientific
limitation in the complexity of contemporary public health studies and on the other, over (or mistaken?)
interpretation of the increased complexity of statistical models.
References
Ness, R.B.; Andrews, E.B.; Gaudino, J.A. et al (2009), ?The Future of Epidemiology?, Academic Medicine,
84(11):1631-1637. |