Effects of approximation in exposure assessments on estimates of exposure- response relationships.