Time trends for population-based disease rates often are summarized by using direct adjustment by period of diagnosis or death. Similarly, the effect of age often is presented graphically as age-specific rates for a given period of diagnosis. These approaches may be necessary if there is an absence of long-term data, as they provide a natural way for annually updating information when monitoring trends, or they may be a convenient way of summarizing a large amount of data (7, 10, 11, 39, 45). However, these summaries only can adjust for the effect of age in a given period; they implicitly ignore the cohort effect. The effect of cohort is an important factor in understanding time trends for many diseases. Thus, it is not advisable to use data analytic strategies that routinely ignore it. Another alternative to modeling is to give a graphical presentation of the age-specific rates themselves. As I noted in the introduction, some of the first analyses to identify the effect of cohort on diseases, such as tuberculosis and lung cancer, relied entirely on a graphical analysis. Although graphs certainly are an important part of the interpretation of time trends, it would be a mistake to limit your analysis to impressions of points on a graph. For example, such a perusal would not give an objective indication of the statistical significance of a particular pattern. Regression analysis forces us to recognize a fundamental problem with interpreting time trends in disease rates--a problem that you should remember, even when trying to understand a graphical display of time trends in age-specific rates.
PIP: The article is concerned with the effects of age, period, and cohort on disease incidence and with mortality from that disease. The emphasis is on the interpretation of results from fitting regression models to data that include age, birth cohort, and period or year of diagnosis. Some examples are presented using U.S. data for lung cancer.