Epidemiologic investigations of associations between protracted low level occupational exposures and cancer mortality routinely encounter the following problems: 1) potential latency effects between exposure and disease;2) potential bias resulting from exposure measurement error;and, 3) potential bias resulting from health-related selection out of employment. The identified problems are of direct relevance to worker protection, as each is a source of bias that may lead to spurious conclusions about the adverse effects of occupational hazards. The goal of this research is to accelerate the development and dissemination of innovative analytical tools to reduce bias and increase precision of risk estimates derived from cohort studies. Leveraging work in our initial project, we draw upon the insights into each problem developed and demonstrate how solutions can be found via analogies to state-of-the-art methods applied in other research areas. We can quickly translate these methods for application to our purposes and apply them in cohort analyses to illustrate their utility. We will demonstrate how the standard approach to exposure time-window analysis may be coupled with a second stage parametric latency model to reduce bias and improve precision of estimates of exposure-time-response associations. During the initial project we developed an innovative method to directly fit a parametric latency model in a simple (single stage) regression model. Hierarchical regression methods have been applied in other research areas for smoothing of parametric functions and shown to yield notable gains in the accuracy of effect estimates. Next, we will develop an approach to correctly account for uncertainty in exposure estimates derived via a job-exposure-matrix (JEM). Recently, investigators have used exposure simulation approaches to generate 'realizations'of exposure scores sampled from underlying distributions. We will demonstrate that the exposure simulation approach may induce attenuation bias in estimates of exposure-disease associations and how Bayesian methods may be coupled with JEMs to provide an intuitive framework for handling uncertainty in exposure estimates without introducing attenuation bias. Finally, we will develop a readily-implementable approach for fitting structural nested models that provide estimates of occupational exposure-cancer associations that are not biased by the healthy worker survivor effect. During the initial project, we assessed bias in exposure-mortality associations in order to explore conditions under which standard regression methods are inadequate. This work showed that there are many settings in which standard regression analysis yielded strongly biased effect measures. Drawing upon methods for structural nested models recently applied in infectious disease epidemiology, we will demonstrate how regression models can be fitted to produce estimates of association that are unbiased by the HSWE. The approach that we will develop produces standard effect measures and overcomes many limitations of prior applications of G-methods. This research will improve the methods used in occupational cancer studies.
The goal of the proposed research, which is a competing continuation of R01-CA117841 Cohort Analysis Methods for Occupational Cancer Studies, is to accelerate the development and dissemination of innovative analytical tools to improve the analysis of occupational cohort studies of cancer.
Showing the most recent 10 out of 22 publications
