Space-time visualization and detection of health disparities using geostatistics and a space-Time information system: The case of prostate and cervix cancer mortality in the United States, 1970-1994

B6

Geographic Information Systems (GIS) facilitate digital cartography, and Atlases of health outcomes for both infectious and chronic diseases are now published by national and state health agencies. These have proven useful for quantifying patterns in health outcomes such as incidence and mortality, for documenting access to health care, providing tools for risk communication and for assessing disparities in cancer burdens in underserved populations. Notwithstanding these benefits, there are substantial limitations that arise from using conventional GIS technology, in particular for the visualization and detection of health disparities in cancer mortality. First, the smaller size of minority populations leads to rates that can be very unreliable and need to be stabilized prior to any analysis. Second, the interpretation of choropleth maps suffers from the common biased visual perception that larger rural and sparsely populated areas are of greater importance. Third, the temporal nature of the data is not properly accounted for in most GIS, which hampers the visualization of temporal trends that can pinpoint locations where disparities greatly increased over time and health policies need to be changed. Fourth, the analysis of health disparities over small geographic areas suffers from the multiple testing problem caused by the repeated use of statistical tests and leading to misclassify disparities as present when they actually are not.
 This paper presents recent developments in the field of space-time information system and geostatistical analysis of cancer rate data. A methodology is presented to create isopleth maps of cancer mortality risk from observed rates and to propagate the uncertainty attached to the risk estimates through the detection of significant health disparities. Spatial patterns are characterized using population-weighted semivariograms, while Poisson kriging incorporates the size and shape of administrative units, as well as the population density, into the filtering of noisy mortality rates. Temporal changes in health disparities are investigated using both absolute (rate difference) and relative (rate ratio) measures, while the False Discovery Rate approach is used for multiple testing correction. The approach is applied to the detection of disparities in cervix and prostate cancer mortality between black and white populations, using data recorded over all US State Economic Areas for five time periods of 5 years each.
 The two cancers display opposite temporal trends: prostate cancer mortality rates have steadily increased since 1970, while prevention has reduced the mortality rate for cervix cancer during the same period. The absolute measure of disparity indicates that the gap between white and black mortality rates widens for prostate cancer while it shrinks for cervix cancer. This decline is however no longer observed when factoring in the temporal trend through the computation of rate ratios. The assessment of significant racial disparities across geographic areas is an important tool in guiding cancer control practices, and public health officials must consider the problems of small population size and multiple comparisons, and should conduct disparity analyses using both absolute and relative measures.