The Biostatistics Epidemiology Core (BsEpC) will serve as a vital intellectual forum for the development of a structural framework for our specific hypotheses and guidance for efficient analysis strategies and procedures to implement them. The services (specific aims) of the BsEpC cover four areas: (1) Data management, quality assurance and oversite of quality control;(2) Development of structural framework for epidemiologic analysis;(3) Guidance and implementation of statistical analysis;and (4) provide user-friendly softare to UCB-Stanford data analysts and make software available to other children's environmental health centers. We illustrate these core services in the following description of the BsEpC by providing projectspecific examples ofthe structural framework and data analysis plans for testing specific hypotheses. By laying out our approach to handling intermediate variables in the estimation of direct causal effects of ambient air pollution, we emphasize the innovative aspects of one ofthe analytical approaches common to all projects within this Center. The central features of our distinctive approach are 1) construction of DAGs, and 2) implementation ofthe """"""""road map"""""""" approach to targeted maximum likelihood estimation (TMLE), and 3) analysis of the piecewise natural history study design (for 2 of the 3 projects). For each hypothesized mechanism, our approach partitions the total (crude) effect of air pollution on specific children's health outcomes into natural direct and indirect effects, taking a distributional view of both the exposure and the intermediates hypothesized. From this we derive a distribution ofthe strength of a particular intermediate in the pathway, relative to all pathways that do not include the specific intermediate. We will focus on a exposure distributions observed in our data or relevant for policy considerations along with a set of distributions for the hypothesized intermediates that reflect the heterogeneity expected in different populations. We will carry out the hypothesis tests in the context of a lifecourse view of both critical timing and duration of specific exposures. Moreover, we will do so in a time-efficient manner so that many of the lifecourse questions can be addressed in a single grant cycle.
We provide marginal-population-level estimates in contrast to the conditional estimates in more widely used traditional methods. We employ the population intervention model developed by our Biostatistics Core PI to provide a range of quantative estimates critically important for the translation of epidemiologic findings into formulation of public health and regulatory policy.
Showing the most recent 10 out of 11 publications
