Modelling disease progression in terms of exposure history.

We consider the relationship between accumulating exposure to a putative agent and the associated change in physiologic function. This type of problem is common to prospective studies of cognitive, pulmonary and cardiovascular function. A general model is proposed for data from prospective, observational studies with concurrent measures of exposures and continuous outcome measures. This model permits non-linearity in the relationship between exposure and outcome and is designed to describe outcome in terms of one's entire exposure history. As exposure data are often severely right-skewed, we use regression spline estimation methods which localize the influence of extreme points. We illustrate our methodology using data from a longitudinal epidemiologic investigation of the effects of amateur boxing on neuropsychologic function. Copyright 1999 John Wiley & Sons, Ltd.