SPATIAL AUTOCORRELATION OF CANCER MORTALITY

Studying the geographic variations in cancer mortality can be useful as a means for generating etiologic hypotheses, providing support or casting doubt on existing hypotheses, and suggesting places and scales for future epidemiological research. A descriptive statistical technique, spatial autocorrelation analysis, is introduced and suggested as a method which might enable the researcher to glean more useful information from spatial patterns of disease mortality. An empirical study of the county pattern of cancer in Pennsylvania is provided as an illustration of the potential of this technique. Considerable variation in the degree of spatial autocorrelation was found among the nine sex-specific cancer patterns studied. Three different pair-wise weighting or linkage schemes were used to further explore the spatial structure of the mortality rates. Finally, a spatial autocorrelation function was estimated for those patterns which displayed significant first-order autocorrelation (cancer of the lung, the stomach and the bladder). The shapes of these functions provided further insight into the mortality pattern and this information was evaluated in relation to current etiologic hypotheses. © 1979.