Time period compared to birth cohort in Connecticut incidence rates for twenty-five malignant neoplasms.

Out of necessity and convenience many reports on population-based rates for cancer are limited to analyses by time period of diagnosis, and just how often cohort effects are important in cancer data has not been fully explored. To address this question, Connecticut cancer incidence rates for the years 1940-79 were fitted to the model: Log (incidence rate) = constant + age effect + period effect + birth cohort effect + error term. Data for each cancer site and sex were categorized into 10-year intervals by time period and age group. Significance testing for the curvilinear effects (which are estimable functions) of age (A), period (P), and cohort (C) in the 44 data sets led to no clear choice of model for three data sets; an APC model for 20, an AP model for 7, and an AC model for 14. These choices were corroborated by the RA2 index. Limitations in the interpretation of the results were enumerated. Presentation of population-based cancer rates by implicitly assuming an AP model is valuable (e.g., for studying age distribution in different regions or for age-adjustment in examining international variation or time trends). However, the assumption of an AP model may often be incorrect, as was shown to be the case for most of these 44 data sets. The implications for monitoring trends and generating etiologic hypotheses were discussed in light of the results for cutaneous malignant melanoma and cancers of the cervix, breast, ovary, lung, and bladder.