Flexible Bayesian Hierarchical Models for Estimating Inhalation Exposures

项目总结/摘要 我们建议开发创新的统计工具，用于融合暴露模型和观测数据， 在控制室条件下测量浓度。作为第一步，我们将构建 一个关于实验室暴露室和真实的工作场所环境中暴露情景的丰富数据集，包括- 关于暴露决定因素的数据，如污染物的产生和通风率以及暴露措施， 保证我们将开发一个全面的和计算可行的贝叶斯统计框架， 将物理暴露模型与工作场所的实验数据相结合， 不确定性的来源，并产生可靠的统计推断（估计和预测）。我们将雇用一名 根据监测数据验证物理模型的贝叶斯框架。我们的框架还将包括正式的 用观测数据验证模型的统计方法。我们这样做是通过评估 模型捕捉监测数据中的特征和模式，将灵敏度分析应用于先验的选择， 以及在一组竞争模型中选择或挑选模型。此外，我们亦会制订及派发 用户友好的统计软件包，使研究人员能够实施拟议的方法， 各种各样的物理模型，以无缝和方便的方式分析其数据。一旦成功 项目完成后，我们的开发将使研究人员和风险管理人员能够系统地 评估回顾性暴露，以预测当前和未来的暴露在没有工作过程 或操作，并仅用少量可能具有高度可变性的空气样本来估计接触。 只有几个监测数据点，我们的贝叶斯融合框架将提供更精确的估计， 曝光比监控随着计算方法的进步和廉价的软件实现， 我们的目的是提升正式建模在暴露评估师的军械库中不可或缺的位置。

Project Summary/Abstract We propose to develop innovative statistical tools for melding exposure models and observational data aris- ing from measurements of concentrations in controlled chamber conditions. As a first step, we will construct a rich dataset of exposure scenarios in laboratory exposure chambers and real workplace settings, contain- ing data on exposure determinants such as contaminant generation and ventilation rates and exposure mea- surements. We will develop a comprehensive and computationally feasible Bayesian statistical framework for melding the physical exposure models with experimental data from the workplace to effectively account for the sources of uncertainty and produce reliable statistical inference (estimation and predictions). We will employ a Bayesian framework to validate physical models from monitoring data. Our framework will also include formal statistical measures for validating models with observed field data. We do so by assessing how adequately the models capture features and patterns in the monitoring data, applying sensitivity analysis to the choice of priors, and choosing or selecting a model among a set of competing models. We will also develop and disseminate a user-friendly statistical software package that will enable researchers to implement the proposed methods for a wide variety of physical models to analyze their data in a seamless and convenient manner. Upon successful completion of the project, our developments will allow researchers and exposure managers to systematically evaluate retrospective exposure, to predict current and future exposure in the absence of the working process or operation, and to estimate exposure with only a small number of air samples with possibly high variability. With only a few monitoring data points, our Bayesian melding framework will provide more precise estimates of exposure than monitoring. With advances in computational methods and inexpensive software implementation, we purport to exalt formal modeling to an indispensable position in the exposure assessors' armory.