The application of age, period and cohort models to predict Swiss cancer mortality.

In order to project trends in mortality from 11 major cancer sites in Switzerland to the end of the current century, a log-linear Poisson age/period/cohort model with arbitrary constraints on the parameters was used, fitted to the observed rates for the period 1950-84. One projection was based on the assumption of a total absence of change in the effect of period, the second was based on a linear extrapolation of the logarithms of the seven known periods, and the third was related to a series of a priori external epidemiological hypotheses, whenever available. For instance, coefficients below unity were used for lung and other tobacco-related neoplasms in men, since some decline in exposure to tobacco carcinogens was observed among Swiss men, and above unity for women since the prevalence of smoking has risen among successive generations of women. Although the method has limitations and uncertainties, several qualitative indications could be derived from this exercise. For instance, the various models suggest that the age-standardized mortality from oral cancer in men will probably increase up to the end of the century, even under the optimistic assumption of an appreciable decline in smoking, while cancer of the oesophagus is likely to level-off around current values, as other tobacco-related neoplasms, prostate cancer in men, and breast cancer in women will probably do. Some steady decline is predicted by various models fitted to the incidence of stomach and intestinal cancer in both sexes, and to ovarian cancer. Lung cancer will continue to rise in women but will stop rising in men, and it will possibly fall if the hypothesis of a decline in exposure to tobacco carcinogens proves correct. Although any prediction has, by definition, substantial difficulties and uncertainties, projections of cancer mortality in the near future are based on a substantial amount of information already available, and may offer valuable information for epidemiological inferences and health planning purposes.