DOSE RESPONSE MODELING IN EPIDEMIOLOGIC COHORT STUDIES

本提案通过探索两种更灵活的回归策略：广义加性模型（非参数回归）和非线性剂量度量来解决环境和职业队列研究中的非线性剂量反应估计问题。通常，剂量反应模型假定这种关系在一定尺度上是线性的。然而，许多疾病机制，如致敏或致癌，可能在剂量-反应曲线中产生非线性。此外，线性模型可能不适用于职业队列研究，因为健康工人效应可能导致暴露程度较高的人的风险出现明显的平台期甚至下降。一般加性模型将用于描述累积暴露和三个队列死亡率研究中选定结果之间的剂量-反应曲线的形状，这些研究已建立了暴露反应关联。可用于剂量反应建模的三个数据集是：46,400名汽车工人暴露于金属加工液，5,414名佛蒙特州花岗岩工人暴露于石英形式的二氧化硅，2,342名硅藻土矿工暴露于方英石形式的结晶二氧化硅。疾病结局将包括金属加工液组的胃癌、食道癌、胰腺癌和肝癌，以及两个二氧化硅组的肺癌。在所有队列中检查非恶性呼吸道疾病死亡率。此外，我们将应用金属加工液和结晶二氧化硅的灵活剂量模型，其中包括作为特殊情况的简单累积暴露。与仅限于与累积暴露的线性关系的标准分析不同，Seixas提出的这个模型具有足够的灵活性，可以研究非线性剂量率效应和可变的疾病诱发/潜伏期。次要目标包括直接比较四种金属加工液（矿物油、可溶物、合成物和半合成物）以及石英和结晶二氧化硅的方石多晶岩的致癌性。

This proposal addresses the problem of nonlinear dose response estimation in environmental and occupational cohort studies by exploring two more flexible regression strategies: Generalized Additive Models (non-parametric regression) and a nonlinear dose metric. Typically, dose response models assume that the relationship is linear on some scale. Many disease mechanisms, however, such as sensitization or carcinogenesis, may produce nonlinearities in the dose-response curve. Moreover, linear models may be inappropriate in occupational cohort studies where the healthy worker effect can lead to an apparent plateau or even downturn in risk among the more highly exposed. General additive models will be used to describe the shapes of the dose-response curve between cumulative exposures and selected outcomes in three cohort mortality studies with well established exposure response associations. The three data sets available for dose-response modeling are: 46,400 autoworkers exposed to metalworking fluids, 5,414 Vermont granite workers exposed to silica in quartz form and 2,342 diatomaceous earth miners exposed to crystalline silica in cristobalite form. Disease outcomes of interest will include cancers of the stomach, esophagus, pancreas, and liver in the metalworking fluid cohort, and cancer of the lung in the two silica cohorts. Nonmalignant respiratory disease mortality will be examined in all cohorts. In addition, we will apply a flexible dose model for metalworking fluids and crystalline silica that includes simple cumulative exposure as a special case. Unlike standard analyses that are limited to linear relations with cumulative exposure, this model, proposed by Seixas, is sufficiently flexible to enable investigation of nonlinear dose-rate effects and variable disease induction/latency intervals. Secondary objectives include the direct comparisons of the carcinogenicity of the four types of metalworking fluids (mineral oil, solubles, synthetic, and semi- synthetics) and of quartz and cristobalite polymorphs of crystalline silica.