Statistical Analysis of Occupational Exposure Data

项目总结/摘要 我是环境和职业健康科学（EOHS）部门的助理教授， 伊利诺伊大学芝加哥分校（UIC）公共卫生学院（SPH）。我的职业目标是成为 独立调查员，通过接触评估对职业健康作出贡献而得到承认。 我的研究有两个主题：（1）数学模型的发展，评估和应用 接触，和（2）接触和风险评估的传染性病原体。虽然我的研究 在定量方面，我侧重于模型和模拟的使用。自从来到UIC，我成长了 对职业接触数据的统计分析和解释越来越感兴趣， 风险评估和职业流行病学框架。很少有暴露科学家有适当的 统计培训，在这一领域进行研究;很少有统计学家有时间和兴趣， 研究职业暴露评估。我认识到这些技能是我所接受的培训和 职业虽然我已经自学了新的统计技术，这个奖项的正式培训是必需的 为了独立这些技能将补充我正在进行的研究主题，并使我有一个 在暴露评估方面的全面、独立的研究生涯。 拟议的研究职业发展计划包括三个目标， 课程和独立学习。目标1是培养高级统计概念的能力， 职业暴露评估。目标2将是增加对职业教育方法的了解， 流行病学目标3是获得概念和实践计算机科学技能。导师已确定 他们在生物统计学、职业流行病学和接触评估方面的专业知识。主要导师 获奖者是UIC生物统计学教授Donald Hedeker博士。UIC的共同导师包括Hakan博士 德米尔塔斯和莱斯利·斯泰纳分别在生物统计学和职业流行病学方面有专长。 博士明尼苏达大学的Gurumurthy Ramachandran将担任共同导师， 暴露评估。 拟议的研究项目包括两项研究。研究1将描述 以及钢铁工人在桥梁油漆准备过程中铅暴露的决定因素。我会 使用混合效应探索暴露的程度、可变性和决定因素以及暴露可变性 模型;并确定时变区域监测数据是否是个人呼吸的更好预测器 区域暴露的数据比油漆的金属含量的数据要多，而油漆的金属含量不是随时间变化的。本研究中的一些数据 丢失了--要么丢失了，要么没有被收集。传统上，仅对观察到的数据进行暴露分析 评估，但这种方法可能会导致信息丢失和不准确的标准误差。多 在职业暴露评估之外， 缺失值。使用这些数据，我将确定多重插补相对于完整病例的影响 分析暴露表征，并评估模型选择方法的等效性， 多重插补研究2将开发和测试一个流行病学的预防-反应模型， 个体工人的平均暴露由概率分布描述。我将使用 汉福德队列死亡率研究，1989年，其中包括年辐射剂量和死因特异性死亡率， 华盛顿州汉福德核电站的3万多名工人。这项研究的动机是 观察到使用个体平均暴露评估产生衰减的暴露-应答 相对于线性背景下组平均暴露评估所得结果的关系 回归分析这种衰减是由于暴露估计的不精确造成的，这表明衰减 如果暴露的不精确性在确认-响应中明确说明， 模型这项研究代表了一种管理暴露错误分类的方法，这种方法无处不在 职业流行病学的问题。

PROJECT SUMMARY/ABSTRACT I am an Assistant Professor in the Environmental and Occupational Health Sciences (EOHS) Division, at the School of Public Health (SPH) at the University of Illinois at Chicago (UIC). My career goal is to be an independent investigator, recognized for contributions to occupational health through exposure assessment. My research has had two themes: (1) the development, evaluation and application of mathematical models of exposure, and (2) exposure and risk assessment for infectious agents. While my research has been highly quantitative, I have focused on the use of models and simulation. Since coming to UIC I have grown increasingly interested in the statistical analysis and interpretation of occupational exposure data, within the frameworks of risk assessment and occupational epidemiology. Few exposure scientists have appropriate statistical training to conduct research in this area; and it is rare to find statisticians with the time and interest to study occupational exposure assessment. I recognize these skills as a gap in my training and in the profession. While I have taught myself new statistical techniques, the formal training of this award is required for independence. These skills will complement my ongoing research themes, and enable me to have a comprehensive, independent research career in exposure assessment. The proposed research career development plan includes three aims to be attained through coursework and independent study. Aim 1 is to develop competency in advanced statistical concepts for occupational exposure assessment. Aim 2 will be to increase knowledge of methods in occupational epidemiology. Aim 3 is to gain conceptual and practical computer science skills. Mentors have been identified for their expertise in biostatistics, occupational epidemiology, and exposure assessment. The primary mentor for this award is Dr. Donald Hedeker, Professor of Biostatistics at UIC. Co-mentors at UIC include Drs. Hakan Demirtas and Leslie Stayner who have expertise in biostatistics and occupational epidemiology, respectively. Dr. Gurumurthy Ramachandran of the University of Minnesota will serve as co-mentor, contributing expertise in exposure assessment. The proposed research project includes two studies. Study 1 will characterize the magnitude, variability and determinants of lead exposures among ironworkers during the preparation of bridges for painting. I will explore the magnitude, variability and determinants of exposure and exposure variability using mixed-effects models; and determine whether time-varying area monitoring data is a better predictor of personal breathing zone exposures than data about the metal content of paint, which is not time-varying. Some data in this study is missing - either lost or not collected. Traditionally, only observed data are analyzed for exposure assessment, but this approach can lead to a loss of information and inaccurate standard errors. Multiple imputation, is an increasingly poplar technique outside of occupational exposure assessment that fills in missing values. Using these data, I will determine the impact of multiple imputation relative to complete cases analysis on the exposure characterization, and evaluate the equivalence of model selection methods for multiple imputation. Study 2 will develop and test an epidemiologic exposure-response model in which the mean exposure of individual workers is described by a probability distribution. I will test the method using the Hanford Cohort Mortality Study, 1989, which includes annual radiation doses and cause-specific mortality for more than 30,000 workers at the Hanford nuclear site in Washington State. This study is motivated by the observation that using individual mean exposure assessments yields attenuated exposure-response relationships relative to those obtained with group mean exposure assessments in the context of linear regression. This attenuation arises because of imprecision in exposure estimates, suggesting that attenuation may be minimized or eliminated if imprecision in exposure is explicitly addressed in the exposure-response model. This study represents one approach to management of exposure misclassification, a ubiquitous problem in occupational epidemiology.