Age-period-cohort analysis of cancer incidence not related to screening or smoking: Estimating potentially avoidable cancer burden

The assessment of time trends in cancer can provide useful indications of future demands for clinical care and also can signal changes in underlying causal factors. Using delay-adjusted incidence data from the SEER 9 registry, we fitted age-period-cohort regression models to estimate site specific trends separately in white and black men and white and black women diagnosed between the ages of 21-84 during the years 1976-2003 for the following groupings of cancers: cancers associated with tobacco smoking, cancers associated with screening technologies, and all cancers outside of these two categories. If one removes cancers related to smoking or screening, increases in the incidence of all other cancers combined have persisted, with average annual changes of 1.12%, 0.98%, 0.80%, and 0.88% for white and black men, white and black women, respectively. These linear secular trends in cancer incidence not related to smoking or screening imply a 22% to 32% increase in risk over 25 years. Simple linear time trends alone do not adequately capture the temporal patterns in cancer incidence not related to screening or smoking. Only black women appear to have strictly linear period effects and all four race-by-gender subgroups exhibit deviations from linearity due to cohort effects. Estimated cohort deviations suggested that incidence rates accelerated from one birth cohort to the next, particularly among women. In conclusion, age-period-cohort models of cancer not related to smoking or screening show increasing secular incidence trends. Environmental exposures may explain these patterns.