Interpretation of time trends in disease rates in the presence of generation effects.

A comparison is made of indices used for assessing trends in mortality rates over time, and results for four types of cancer are used as examples. The approaches contrasted as measures of change over calendar years of time are the use of crude death rates, standardized death rates (both direct and indirect approaches) and period effects estimated from an age-period-cohort model (including and excluding the cohort factor). The use of these indices is also compared with examination of the original age-specific rates. Similar analyses are made of the equivalent indices for examining changes between different birth cohorts over time. Submodels of the age-period-cohort model are shown to be algebraically and empirically related to age-standardized indices. The use of the full model, as contrasted to standardization methods, can result in important modifications to the interpretation in some cases. However, there are difficulties with this approach, as with all models, if applied uncritically. It is a wise precaution to examine the age-specific rates in each instance to ensure that any summary index is appropriate.