They discuss trying to separate period effects (which would include improvements in detection) from cohort effects. Here's what they say about it: > To evaluat

They discuss trying to separate period effects (which would include improvements in detection) from cohort effects. Here's what they say about it: > To evaluate birth cohort trends in cancer rates, adjusted for age and period effects, we fitted age-period-cohort models to the incidence or mortality rates of each cancer type using weighted least squares, assuming Poisson-distributed counts and including overdispersion parameters for potential extra-Poisson variation. A change in the birth cohort trend usually indicates changes in exposure prevalence, leading to varying risks of developing cancers for individuals of the same generation. By contrast, a period effect reflects systematic changes in cancer ascertainment or the influence of newly introduced or improved medical interventions, affecting all age groups simultaneously during the same period. I was somewhat surprised that cohort effects are tied to changes in exposure prevalence, when I would have assumed those would give rise to period effects. Presumably exposure during childhood carries lifelong consequences.