A new method for spatially moving correlation analysis in geomedicine

A method of correlation analysis within a spatially moving window was applied to two sets of epidemiological/geochemical data in Norway, (1) mortality/disability rates of multiple sclerosis (MS) versus atmospheric fallout of Mg and concentrations of Se in overbank sediment, and (2) incidence rates of malignant melanoma of the skin (MM) versus concentrations of Ca and K in overbank sediment. It appears that n = 17 observation sites within the moving window is a practical compromise between noise in the data at small values of n and a spatial resolution good enough to detect trends in the distribution patterns of the correlation coefficient. For MS versus Mg, MS versus Se and MM versus Ca the correlation coefficients are generally negative and depict systematic distribution patterns with anomalous clusters of sites with good correlation. For MS versus Se the correlation coefficients also form an additional cluster of positive coefficients. Tests with permutated data show that more than 70% of the negative correlation coefficients for MS versus Mg and for MM versus Ca are both significantly different from zero at p<0.05, while less than 15% of those for MS versus Se are significant at the same level. For MM versus K the correlation coefficients are randomly distributed and not significantly different from zero. The described correlations may be effects of confounders and do not per se indicate any causal relationships. However, further research based on these results may well lead to the identification of possible aetiological factors.