An application of hierarchical regression in the investigation of multiplepaternal occupational exposures and neuroblastoma in offspring

Background We used hierarchical regression to study the effects of 46 pater nal occupational exposures on the incidence of neuroblastoma in offspring. Methods The study population included 405 cases and 302 controls. The effec t of each exposure was estimated using both conventional maximum likelihood and hierarchical regression. Results Using hierarchical regression, overall precision was greatly enhanc ed compared to the conventional analysis. In addition, adjustment of effect estimates based on prespecified prior distributions of the true effect par ameters allowed a more consistent interpretation across the entire panel of exposures. Estimates for several metals and solvents were shrunk close to the null value, whereas estimates for several thinner solvents, diesel fuel , solders, wood dust, and grain dust remained moderately elevated. Conclusions Hierarchical regression may mitigate some of the problems of th e conventional approach by controlling for correlated exposures, enhancing the precision of estimates, and providing some adjustments of estimates bas ed on prior knowledge. Am. J. Ind. Med. 39:477-486, 2001. (C) 2001 Wiley-Li ss, Inc.