Statistical techniques applied to environmental health sciences

我们最近指导了几个项目的数据分析，主要涉及来自农业健康研究（AHS）的数据，这是一组来自爱荷华州和北卡罗来纳州的持牌农药施用者（主要是农民）。 在第一个项目中，我们试图评估伤害死亡率是否与AHS中男性农民施用者的农药使用实践有关。 从1993年到2008年，我们在727，543人年的随访中观察到338例伤害死亡。 我们使用考克斯比例风险模型来研究致命伤害与使用特定杀虫剂、个人防护设备、特定类型的农场设备和其他因素的关系。 在入组时收集死亡前1 - 15年的暴露信息。 致命伤害随着农民每年使用杀虫剂的天数增加而增加，随着使用耐化学品手套而减少。 某些除草剂与致命伤害有关。 我们认为这些关联值得进一步研究，特别是在暴露与死亡的时间关系方面有更好的数据，我们强调必须谨慎解释这些结果。 有机磷杀虫剂（OP）是美国最常用的杀虫剂之一。 在第二个项目中，我们的特点是使用有机磷农药的农民农药施用者在AHS。 为此，我们评估了701名农民施用者中OP的终生使用情况，这些农民被招募进行神经系统测试研究，增加了OP使用的信息。 该样本是在复杂的抽样设计中随机选择的，该设计对高度暴露的个人进行了过采样（基于通常的AHS数据），并仅限于居住在四个临床地点150英里范围内的个人。 我们使用经验性抽样权重，以便我们的估计值反映所有3863名符合神经系统研究资格的男性施源者的OP经验。 马拉硫磷和毒死蜱是最常用的有机磷农药。 OP的使用在AHS的15年过程中下降。 另一个项目是探索AHS中男性私人施用者接触农药与抑郁之间可能的关系。 我们使用逆概率加权来解释由于缺失数据而导致的排除，并使用多分类logistic回归来估计比值比。 曾经使用熏蒸剂作为一类和几个单独的农药与抑郁症有关。 利用AHS的数据，我们还研究了农业暴露是否会导致哮喘恶化。 我们注意到哮喘与几种农业活动和两种除草剂的使用呈负相关，但没有正相关（除了哮喘和过敏的病例子集）。 这些观察结果与哮喘易于加重的病例倾向于避免引发症状的暴露的可能性一致。 最后一个使用AHS数据的项目检查了糖尿病与注册前使用特定农药之间的可能关系。 以年龄为时间尺度，以糖尿病诊断为关注事件，我们拟合考克斯比例风险模型来估计风险比。 曾经使用过五种杀虫剂（其中三种是有机磷）与农民妻子的糖尿病有关，他们报告曾经混合或使用过杀虫剂。

We have recently guided the data analysis for several projects, primarily involving data from the Agricultural Health Study (AHS), a cohort of licensed pesticide applicators (mostly farmers) from Iowa and North Carolina. In the first project, we sought to assess whether injury mortality was associated with pesticide-use practices among male farmer applicators in the AHS. From 1993 to 2008, we observed 338 injury fatalities over 727,543 person-years of follow-up. We used Cox proportional hazards modeling to study the association of fatal injury with use of specific pesticides, personal protective equipment, specific types of farm equipment, and other factors. Exposure information was collected at enrollment, from 1 to 15 years preceding death. Fatal injury increased with the number of days per year that a farmer applied pesticides and decreased with use of chemical-resistant gloves. Certain herbicides were associated with fatal injury. We regard these associations as deserving of further study, especially with better data on the temporal relationship of exposure and death, and we emphasize that these results must be interpreted with caution. Organophosphorous insecticides (OPs) are among the most commonly used insecticides in the US. In the second project, we characterized the use of OPs among farmer pesticide applicators in the AHS. For this effort, we assessed lifetime use of OPs among 701 farmer applicators who were recruited for a neurological testing study with augmented informaton about OP use. This sample was randomly selected in complex sampling design that oversampled highly exposed individuals (based on the usual AHS data) and restricted to individuals residing within 150 miles of four clinical sites. We used empirical sampling weights so that our estimates reflect the OP experience of all 3863 male applicators eligible for the neurological study. Malathion and chlorpyriphos were the most commonly used OPs. OP use declined over the 15 year course of the AHS. Another project is probing a possible relationship between pesticide exposure and depression among male private applicators in the AHS. We used inverse probability weighting to account for exclusions due to missing data and polytomous logistic regression to estimate odds ratios. Ever use of fumigants as a class and several individual pesticides were associated with depression. Using data from the AHS, we have also studied whether farming exposures contribute to exacerbation of asthma. We noted inverse association of asthma with several farming activities and with ever use of two herbicides but no positive associations (except in a subset of cases with asthma and allergy). These observations are consistent with the possibility that asthma cases prone to exacerbation tend to avoid exposures that trigger symptoms. A last project using data from the AHS examined the possible relationship between diabetes and ever use of specific pesticides before enrollment. Using age as the time scale and diagnosis with diabetes as the event of interest, we fit Cox proportional hazards models to estimate hazard ratios. Ever use of five pesticides (three of them organophosphates) were associated with diabetes among farmers wives who reported ever mixing or applying pesticides.