Interpreting age, period and cohort effects in plasma lipids and serum insulin using repeated measures regression...

Observed changes in health-related behaviours and disease risk factors may arise from physiological or environmental changes, or from biases due to sampling or measurement errors. We illustrate problems in the interpretation of such changes with longitudinal data from the Coronary Artery Risk Development in Young Adults (CARDIA) study. Mean plasma cholesterol was 14 mg/dl higher in 27- than in 20-year-old black men cross-sectionally, but longitudinally it declined by 4 mg/dl during the 7 years. To sort out these contradictory assessments of the effect of age/passage of time, we estimated age and period effects under the assumptions that age effects are a smooth function of age independent of period, and that period effects are changes common to persons across all ages. Simple estimates the age effect, such as the cross-sectional age slopes, may be confounded by cohort effects, by interactions of time and age after baseline, or by the occurrence of non-linearities in response after baseline. We note examples of each potential type of bias. The data and background literature support the assumption that cohort effects do not seriously compromise interpretation for these variables in the CARDIA study. Strong secular decreases in plasma cholesterol, apparently due to population-wide dietary change, mask increases with ageing. Age increases in triglycerides are largely explained by increases in body fatness. For these data, we cautiously accept the cross-sectional age slope as an estimate of ageing and the age-matched time trend as an estimate of secular trend.