[Application of multivariate statistical methods to the descriptive epidemiology of malignant tumors]

A study utilizing cluster analysis was undertaken to evaluate the association between standardized cancer mortality rates in different Italian provinces during the period 1975-77. The cancer sites examined were: stomach, lung, colon-rectum, liver and biliary ducts, pancreas, urinary bladder, oral cavity and pharynx, oesophagus. In addition larynx and prostate were examined in males while and uterus were considered in females. The cluster analysis was carried out using the "Euclidean distance" as dissimilarity index and the UPGMA (Unweighted pair group method using arithmetic average) as a clustering method, and considering as OTUs (Operational Taxonomic Units) both cancer sites and Italian provinces. Cluster analysis applied to cancer sites showed a strong relationship, in both sexes, between the alcohol-tobacco-related neoplasms. The same association was found by other Authors in several countries of the world. A similar analysis, carried out in USA during the period 1950-67 and consistent with our results, inserted lung cancer in the same cluster. In the present study this site demonstrated different characteristics in the two sexes: in females, together with the pancreas, is clustered, at higher dissimilarity level, with oral cavity, while it remains unclustered in males. This fact is probably due to different smoking habits between the two sexes and to the importance of the alcohol-tobacco interaction. Another association, also in agreement with the results obtained by other Authors, regarded female breast and colorectal cancer on one side, and female liver and uterus cancers on the other. Gastric cancer, particularly in males, is clearly distinguished from other sites, agreeing with the concept that mortality from this site is often inversely related with that of other cancers. Cluster analysis applied to Italian provinces discriminated, in both sexes, two clusters: Northern and Southern provinces. Into these two, some clusters group provinces at high mortality from alcohol-tobacco related cancers, gastric and lung cancer. Moreover, several clusters are shown and the interpretation of the causes for those association can lead to different causal hypotheses. In conclusions, the Authors confirm that the application of this multivariate statistical method can be an efficient tool in descriptive epidemiology. The results obtained in this study, while partially agreeing with the international literature, constitute a starting point for formulating new hypotheses which take into account the relationships and interactions between several risk factors.