Some effects of within-person variability in epidemiological studies

Abstract When individuals with high levels of a biological variable are given a treatment to lower these levels, it is important to be able to assess how much of any subsequent decrease is due to the treatment. The fact that individuals selected for their high levels on one occasion will (on average) show lower levels on subsequent occasions even in the absence of treatment is now recognized, and has been termed the ‘regression’ effect. The reverse is true for low levels. This paper presents a method for predicting how much change can be expected from regression, and thus a means of estimating the treatment effect. The analysis is based on measures of within-person and between-person variability, the within-person source being responsible for the regression effect. The effect that within-person variability has on misclassification of people into high-risk groups, and a means of determining how many estimates of one individual's level are required for a given precision of classification, are also derived from the same analytical approach. Examples are given using data on cholesterol levels following treatment with Clofibrate, caloric intake levels in a diet survey, and the relationship between the incidence of ischemic heart disease and systolic blood pressure level.