Abstract
Trends in mortality rates are usually presented per tumour site or per country without an overall analysis of the complete data encompassing all three aspects (tumour sites, countries, trends). This paper presents a methodology for such an overall analysis using three-way methods applied to a data set on female mortality rates for 17 tumour sites of 43 countries for the years 1968-1985. Multivariate techniques like biplots and three-mode principal component analysis within an overall three-way analysis-of-variance framework were used. We confirmed the known patterns of comparatively high mortality for women due to cancer of the bladder, intestines, pancreas, rectum, breast, ovary, skin and leukaemia and the relatively low mortality rates for liver cancer in Western and Northern Europe, the USA, Australia and New Zealand. Also, the reverse pattern was observed for Middle and Southern Europe, Hong Kong, Singapore, and in Japan, and in some but not all Latin American countries. The relatively mortality due to cancer was high in the lungs, mouth, larynx and oesophagus in the British Isles, but was much less in other European countries. Mortality due to cancer of the thyroid, uterus, gall bladder and stomach was high in Middle European countries, as was the case in Japan, Chile and Costa Rica. Rates were low for Southern European countries, North America, Australia and New Zealand. Specific deviating patterns in the data were the more rapidly decreasing mortality rates for stomach cancer in Chile and Japan and the more rapidly increasing mortality rates for lung cancer in the USA, Scotland and Denmark. In conclusion, using three-way methods, it was feasible to analyse the cancer mortality data in their entirety. This enabled the simultaneous comparison of trends in relative mortality rates between all countries due to all tumour sites, as well as the identification of specific deviating trends for specific tumour sites in specific countries.