'Needles in a haystack' is often quipped before studies of community-level estimation of the prevalence of disease and behavioral risk factors. In a presentation aptly regarded as "haystack epidemiology," we will present analyses at multiple levels, used with a progressive logic of stochastic curtailment for adjacent and nested geographic areas. In such circumstances, simple accruing probability may be applied as a means for identifying surrogate or relative measures of the variation in local prevalence estimates. Our statistical solutions use prior distributions to form population-based probability-weighted geographic surfaces. This approach relies on established disease occurrence or risk behavior patterns, as well as detailed population characteristics.
The age-, race-, gender-, space-specific prior distributions provide a Bayesian 'expected' probability surface for mapping small area prevalence distributions, building on 'observed' large area simulations. One may also select sentinel events for their biologic value from suspected environmental attributes, or high-risk configurations for under-served groups (e.g., rural populations, children) to provide a hierarchical weighting to findings. Then, sequential solutions may be programmed with recognition thresholds for identifying evidence of emerging patterns (e.g., behavior change).
We have simulated these applications to illustrate these 'haystack' applications of sequential observation for declining spatial variability with studies of cancer. Illustration of the progressive logic of the spatially declining randomness will be described. This strategy with public health surveillance can involve multiple-level analyses, e.g., statewide, regions within a state, counties, and for some metrics even ZIP codes. The 2000 census also offers a profound opportunity for census tract-based spatial analyses.
Learning Objectives: At the conclusion of this talk, the attendee should be able to describe how established disease occurrence or risk behavior patterns may be combined with detailed population characteristics, large area simulations, and a stochastic curtailment strategy, to build a Bayesian 'expected' probability surface for mapping small area prevalence distributions.
Presenting author's disclosure statement:
Organization/institution whose products or services will be discussed: None
I do not have any significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.