Steven Stellman - Imputation method for lifetime exposure assessment in air pollution epidemiologic studies

Version 1

      Publication Details (including relevant citation   information):

      Beyea, J., Stellman, S. D., Teitelbaum, S., Mordukhovich, I.,   Gammon, M. D. 12 62-

      Abstract: BACKGROUND: Environmental   epidemiology, when focused on the life course of exposure to a   specific pollutant, requires historical exposure estimates that   are difficult to obtain for the full time period due to gaps in   the historical record, especially in earlier years. We show that   these gaps can be filled by applying multiple imputation methods   to a formal risk equation that incorporates lifetime exposure. We   also address challenges that arise, including choice of   imputation method, potential bias in regression coefficients, and   uncertainty in age-at-exposure sensitivities. METHODS: During   time periods when parameters needed in the risk equation are   missing for an individual, the parameters are filled by an   imputation model using group level information or interpolation.   A random component is added to match the variance found in the   estimates for study subjects not needing imputation. The process   is repeated to obtain multiple data sets, whose regressions   against health data can be combined statistically to develop   confidence limits using Rubin's rules to account for the   uncertainty introduced by the imputations. To test for possible   recall bias between cases and controls, which can occur when   historical residence location is obtained by interview, and which   can lead to misclassification of imputed exposure by disease   status, we introduce an "incompleteness index," equal to the   percentage of dose imputed (PDI) for a subject. "Effective doses"   can be computed using different functional dependencies of   relative risk on age of exposure, allowing intercomparison of   different risk models. To illustrate our approach, we quantify   lifetime exposure (dose) from traffic air pollution in an   established case-control study on Long Island, New York, where   considerable in-migration occurred over a period of many decades.   RESULTS: The major result is the described approach to   imputation. The illustrative example revealed potential recall   bias, suggesting that regressions against health data should be   done as a function of PDI to check for consistency of results.   The 1% of study subjects who lived for long durations near   heavily trafficked intersections, had very high cumulative   exposures. Thus, imputation methods must be designed to reproduce   non-standard distributions. CONCLUSIONS: Our approach meets a   number of methodological challenges to extending historical   exposure reconstruction over a lifetime and shows promise for   environmental epidemiology. Application to assessment of breast   cancer risks will be reported in a subsequent manuscript.

      Address (URL): http://www.ncbi.nlm.nih.gov/pubmed/23919666