Do the predictions for cancer incidence come true? Experience from Finland

The observed occurrence of cancer in Finland in 1980 was compared with that predicted by extrapolation of the trends in cancer incidence by site in 1957 to 1968. The predictions had been constructed as 90% confidence intervals. Of the mutually exclusive 33 incidence predictions, 22 (67%) included the observed incidence rate. The incidence rate was successfully predicted for intestinal, pancreatic, and urinary cancers, as well as for lymphomas, whereas incidence forecasts concerning cancers of the lung, breast, and cervix uteri, and melanomas of the skin did not come true. The incidence of male lung cancer did not increase as was expected, breast cancer incidence increased even more than expected, and the decrease in the incidence of cervical cancer was steeper than expected. Because of failure in the population forecast, only 55% of the predictions (90% confidence intervals) for the numbers of new cancer cases included the observed number. For administrative purposes it would be important for the predictions to come true. On the other hand, the successes or failures of the predictions related to specific cancers may be used to study the effect of changes in etiologic factors, diagnostic methods, or the definition of cancer, as well as the effect of mass screening or other intervention measures upon the cancer incidence. Extrapolation as a method of forecasting cancer incidence is adequate as far as the control of random error is concerned when only short-term forecasts, ranging from 5 to 10 years, are needed.