A methodology to assess possible effects of enhanced surveillance on the risk estimate from ecologic studies of thyroid cancer after the Chernobyl accident

Summary For two reasons quantitative post-Chernobyl assessment of the thyroid cancer risk from 131 I exposure in children under age 18 at the time of accident has been carried out with spatially aggregated data. Firstly, individual data has not been available up to now. Secondly, a large number of individuals can be included in the analysis for a better statistical power. But aggregation can destroy important individual information. Therefore, the risk estimate from an ecologic study may be biased, especially in the presence of confounding factors. In 1998 Lubin has investigated such an ecologic bias for lung cancer caused by radon exposure confounded by smoking. We generalise his approach for the Chernobyl case where enhanced medical surveillance (or screening) has been identified as the most important confounder of cancer incidence data. Our investigations were performed with both Monte-Carlo simulations and analytical calculations, respectively. Based on realistic dose estimates for 743 Ukrainian settlements we simulated individual data sets on exposure, screening and health status for 366397 children. To generate the cancer cases, an individual risk model with linear dose response has been combined with three different screening models. Poisson regression on a linear model for a mean settlement risk produced an ecologic bias in most cases, when screening information was neglected. The estimated risk parameters deviate from the true mean parameters which describe the true risk in the population. The bias is caused by those correlations between screening and risk factors, which have not been included in the definition of the true population-based risk. Analytical relations have been derived to calculate the bias numerically exact from the population data.