Assessing the effect of time-varying covariates in cross-sectional studies.
In cross-sectional studies exposure and disease status of individuals are assessed at the same point in time, but sometimes information on prior exposure status is also gathered. Under such circumstances one approach to assessing the relationship between disease and exposure is using linear or logistic regression analysis, adjusting for exposure status at different points in time. It is shown that estimates for the effect of exposure at a certain point in time, adjusted for exposure at another point in time, are obtained from comparisons between groups with different patterns of exposure. Careful interpretation of the resulting estimates is necessary, taking into account a detailed consideration of possible exposure patterns. In addition, if changes in exposure are caused by the occurrence of the disease, then adjusting for multiple measurements of exposure can give misleading results. A regression analysis on dummy variables describing possible patterns of changes in exposure is proposed as an alternative approach. This approach facilitates interpretation of the resulting estimates. Furthermore, it can serve as a diagnostic tool to check for disease related changes in exposure. For this case transferring exposure change rates of healthy subjects to diseased subjects is suggested as an ad hoc method for assessing the hypothetical current relationship between exposure and disease.