Try Free

Past Research

Geostatistical Software for the Space-Time Analysis of Health Disparities

NCI SBIR Phase I and II Completed research 2004-2008
Principal Investigator: Pierre Goovaerts, PhD

Summary

This SBIR project is developing the first GIS-based software to offer tools that are specifically designed for the space-time analysis of health disparities, providing descriptions of spatial patterns of cancer mortality rates and identification of scales of variability, spatial filtering to correct for statistical instability caused by the smaller size of minority populations, statistical tests to detect significant differences in cancer risks among sub-populations, detection of clusters and outliers of significantly high or low health disparities, exploration of local relationships with covariates (i.e. demography, socio-economic variables) using geographically-weighted regression and multilevel analysis, and visualization of changes in disparities through time.

Phase I of this project:

  • Conducted a requirements analysis to identify the optimal spatial methods and functionality to incorporate into the software: TerraSeer Space-Time Intelligence System™ (STIS™).
  • Developed and tested innovative geostatistical techniques for spatial filtering of cancer rates and statistical tests to detect significant differences in cancer rates among sub-populations.

Phase II of the project will:

  • Expand the statistical methodology developed in Phase I in order to:
    • Estimate filtered rates over the same spatial support as potential covariates such as demography, behavioural, environmental and socio-economic variables, enabling the analysis of neighborhood variations in health disparities at the most informative levels of census geography.
    • Quantify the uncertainty attached to filtered rates and propagate it using Monte Carlo simulation through the statistical analysis, including the detection of health disparities, multilevel analysis and geographically-weighted regression.
    • Integrate space into the multilevel modeling of health data, through the construction of neighbor-hoods that are no longer arbitrary administrative units but rather account for the complex interaction of multiple socio-economic characteristics through spatially constrained clustering algorithms.
    • Test assumptions regarding the socio economic and demographic factors responsible for different aspects of the geography of disparities through the incorporation of neutral spatial models of disparity in univariate and bivariate LISA (Local Indicator of Spatial Autocorrelation) statistics.
  • Build and test a complete set of functionalities based on the research results, and incorporate them into BioMedware’s space-time visualization and analysis technology, providing a comprehensive suite for disease cluster detection, geographic boundary analysis, measurement and statistical analysis of health disparities, and detection of their changes in both space and time. These tools will be suited for the analysis of both aggregated and individual-level data.
  • Apply the software and methods to demonstrate the approach and its unique benefit for the measurement, mapping, detection and explanation of health disparities, including evaluation by members of the expert panel who will convene for a workshop during the second year of this project.
  • Create instructional materials, including a short course, to foster the adoption and use of this integrated approach in the health sciences.

These technologic, scientific and commercial innovations will revolutionize our ability to interpret geographic variation in cancer disparities, detect changes in space (e.g. cancer cluster) and through time (e.g. change in health disparities following strategies to improve cancer prevention and early detection), and to better understand the causes underlying observed racial disparities in cancer incidence, mortality and morbidity. In particular, the methods in this proposal make possible, for the first time ever, evaluation of the sensitivity of the results of ecological regression and local cluster analyses to the uncertainty attached to estimated rates. Within a study, this will provide detailed quantification of the reliability of the results, and should result in a spatially explicit analysis of potential false positives and false negatives.