Page 205 - Modern Spatiotemporal Geostatistics

186 Modern Spatiotemporal Geostatistics — Chapter 9

(multivariable) BME analysis using both the exposure and the health-
effect data are superior to the effect predictions obtained from scalar
(single-variable) BME analysis using only health-effect data.
The PEP criterion is well suited for exposure and health distributions
with pronounced spatial and temporal characteristics and provides a meaningful
representation of the exposure-effect association in spatiotemporal domains by
way of graphs and maps. Before the PEP criterion is applied in practice, certain
conditions must be satisfied (see Christakos and Hristopulos, 1998; Christakos
and Serre, 2000a):
(i.) Exposure precedes health effect (e.g., there may exist a history of reg-
ularity in such a precedence, or there is a biological possibility of the
precedence in light of existing knowledge about disease etiology).
(ii.) Exposure and health effect are contiguous in the spatiotemporal do-
main (i.e., there is a clear link in time and place of the exposure and
health effect that we are connecting causally). Condition (M.) requires
the existence of some spatiotemporal connection between exposure and
effect (e.g., when we say that a pollutant caused a group of receptors
to become ill, we imply that the pollutant and the receptors both are
located in the same geographical area). In many cases this contiguity is
not a trivial aspect, for biological or organic systems are in a constant
state of exchange with their surrounding environmental conditions.
(Hi.} The necessary adjustments for confounding variables (i.e., variables that
may be closely associated with both exposure and effect) have been
made, so that their effects can be clearly distinguished from those of the
exposure under investigation. Nevertheless, several studies have shown
that strong associations are highly unlikely to be due entirely to a hidden
confounding variable, unless this variable is closely associated with the
health effect and the risk factor (e.g., Flanders and Khoury, 1990; Khoury
and Yang, 1998). Also, Rothman and Greenland (1998) have suggested
that, given one's ignorance regarding the hidden causal components, the
best possible approach to health-risk assessment is to classify people
according to measured causal risk indicators and then assign the average
risk observed within a class to persons within the class.
While these three conditions are commonsense in epidemiologic investi-
gations (e.g., Hill, 1965), no one of them is an all-sufficient basis for judg-
ment. A novel condition introduced by Postulate 9.1 is that the existence and
strength of an exposure-effect association is judged on the basis of the suc-
cessful space/time predictions to which the combined physico-epidemiologic
analysis leads. Thus, a central feature of the scientific status of the PEP cri-
terion is its testability, i.e., the predictions made by the PEP criterion are
testable. The better the vector health-effect predictions (i.e., BME predictions
made on the basis of physical exposure and epidemiologic data) compare to
the scalar health-effect predictions (i.e., BME predictions made on the basis
of epidemiologic data only), the stronger is the exposure-effect association,