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

Information on exposure in cohort studies of occupational groups is often very incomplete. If assessments of exposure level are attempted, they are inevitably subject to random error. Even when unbiased, such measurement error can give rise to bias in an estimate of an exposure-response relationship. Since practical techniques exist to account for this effect only in certain situations where response may be measured on a continuous scale, a technique suitable for mortality studies needs developing. In the present report a procedure is described for reducing the bias due to random error in the estimated exposure-response relationship from a cohort mortality study. This procedure is based on a mechanism for adjusting exposure assessments, taking into account the distribution of exposures, and the independently estimated distribution of measurement errors. The initial bias and its reduction through analysis is shown for some simulated data sets. The results of applying this technique to a cohort mortality study of asbestos workers are presented.