A nonparametric approach for determining significance of county cancer rates compared to the overall state rate: illustrated with Minnesota data.

BACKGROUND

The study of the geographical distribution of disease has expanded greatly with GIS technology and its application to increasingly available public health data. The emergence of this technology has increased the challenges for public health practitioners to provide meaningful interpretations for county-based state cancer maps.

METHODS

One of these challenges-spurious inferences about the significance of differences between county and overall state cancer rates-can be addressed through a nonparametric statistical method. The Wilcoxon's signed rank test (WSRT) has a practical application for determining the significance of county cancer rates compared to the statewide rate. This extension of the WSRT, developed by John Tukey, forms the basis for constructing a single confidence interval for all differences in county and state directly age-adjusted cancer rates. Empirical evaluation of this WSRT application was conducted using Minnesota cancer incidence data.

RESULTS

The WSRT procedure reduced the impact of statistical artifacts that are frequently encountered with standard normal significance testing of the difference between directly age-adjusted county and the overall state cancer rates.

CONCLUSION

Although further assessment of its performance is required, the WSRT procedure appears to be a useful complement for mapping directly age-adjusted state cancer rates by county.