Indoor Allergens And Asthma

我们的研究项目侧重于环境在哮喘和过敏性疾病的发展和恶化中的作用。我们与CDC/NCHS的研究人员合作，为国家健康和营养检查调查（NHANES）开发并实施了过敏/哮喘重点组件。该部分纳入了NHANES 2005-2006，询问了过敏和哮喘的患病率和发病率，测量了卧室灰尘中常见室内过敏原和内毒素的水平，并量化了9000多名参与者血清中总过敏原和过敏原特异性IgE水平。对这一大型数据集的分析使我们能够1)估计全国范围内室内过敏原和内毒素暴露的流行程度，2)估计全国范围内对室内、室外和食物过敏原的过敏性致敏程度，3)估计全国范围内包括哮喘在内的过敏性疾病的流行程度，以及4)调查过敏原和内毒素暴露、过敏性致敏和过敏性疾病之间的复杂关系。与之前的研究相比，该成分不仅在更广泛的年龄范围内测试了更多数量的过敏原，而且还提供了有关过敏致敏程度和室内过敏原和内毒素暴露程度的定量信息。它建立了评估美国家庭中过敏原和内毒素暴露趋势的第二个时间点估计，第一个是在全国住房铅和过敏原调查中建立的，我们与住房和城市发展部合作完成了这项调查。这些数据使得对过敏原/内毒素暴露和ige介导的致敏在过敏性疾病中的作用的研究比以前可能的更有力和更广泛。我们在了解室内过敏原/内毒素暴露的患病率和决定因素及其与过敏性疾病的关系方面取得了重大进展。我们的研究表明，暴露于室内过敏原和内毒素是常见的，但在美国家庭中变化很大。我们的研究结果强调了环境因素对人类健康和疾病的影响和重要性，包括哮喘。虽然我们的重点仍然是哮喘/过敏相关的结果，但我们已经将兴趣扩展到其他健康结果以及流行病学方法学研究领域。尽管机器学习作为一种研究工具越来越受欢迎，但很少有研究调查了在复杂调查数据中实施机器学习方法的含义。我们使用来自NHANES的数据和不同设计场景的模拟来评估在梯度提升中考虑采样权的影响，梯度提升是一种强大的集成分类算法，可以单独使用，也可以作为流行的R软件中的超级学习者的一部分。为了评估超参数调优的作用，我们使用默认超参数以及从大规模并行计算环境（NIH Biowulf）中随机搜索超参数空间的每个模拟中选择的超参数进行了所有分析。这个高性能集群在世界上最强大的500台计算机中名列前茅，使我们能够使用10,000个并发cpu来拟合大约22亿个树模型，并在模型交叉验证和自举置信区间中获得高水平的置信度。我们证明了使用没有抽样权重的复杂调查数据配置的模型可能无法准确反映目标人群的预测，这取决于样本量和其他分析性质。我们还证明，在没有软件配置加权算法的情况下，使用加权观察结果对模型性能进行事后重新计算，可能比完全忽略权重更有效地代表目标人群的预测。我们的发现强调了进一步研究的必要性；随着机器学习的普及程度不断提高，未能更多地关注适当分析复杂的调查数据可能会错过利用这些新颖而强大的方法的机会。在另一个合作项目中，我们调查了美国普通人群中抗核抗体（最常见的自身免疫生物标志物）的流行情况。NHANES的研究结果表明，在25年的时间跨度内，抗体流行率呈上升趋势，近年来增幅更大。尽管这种增加的速度和时间因亚组而异，但年龄和性别始终与抗核抗体的流行有关。随着时间的推移，青少年、男性和非西班牙裔白人参与者的增长最为明显。这些结果为自身免疫疾病提供了有价值的流行病学见解，并将有助于设计进一步的研究，以更好地了解自身免疫患病率增加的潜在因素和原因。由于NHANES数据允许调查许多有趣的关系，我们将继续更详细地研究过敏原暴露、过敏致敏和疾病之间的复杂关系。我们的研究将有助于更好地了解环境暴露的特征，如室内过敏原和内毒素暴露，以及它们在过敏性疾病中的作用，从而为开发有效的环境干预方法来管理过敏性疾病，如哮喘，提供见解。

Our research program focuses on the role of the environment in the development and exacerbation of asthma and allergic diseases. In collaboration with investigators at the CDC/NCHS, we developed and implemented an allergy/asthma focused component for the National Health and Nutrition Examination Survey (NHANES). This component, included in NHANES 2005-2006, queried on allergy and asthma prevalence and morbidity, measured levels of common indoor allergens and endotoxin in bedroom dust, and quantified total and allergen-specific IgE levels in serum of more than 9000 participants. Analysis of this large data set has allowed us to 1) estimate nationwide prevalence of indoor allergen and endotoxin exposures, 2) estimate nationwide prevalence of allergic sensitization to indoor, outdoor and food allergens, 3) estimate nationwide prevalence of allergic diseases, including asthma, and 4) investigate the complex relationships between allergen and endotoxin exposures, allergic sensitization and allergic diseases. This component not only tested a greater number of allergens across a wider age range than prior studies, but also provided quantitative information on the extent of allergic sensitization and exposures to indoor allergens and endotoxin. It established a second point-in-time estimate for evaluating allergen and endotoxin exposure trends in U.S. homes, the first being established in the National Survey of Lead and Allergens in Housing, which we completed in collaboration with the Department of Housing and Urban Development. The data have enabled more robust and generalizable investigations of the role of allergen/endotoxin exposures and IgE-mediated sensitization in allergic diseases than previously possible. We have made significant advances in our understanding of the prevalence and determinants of indoor allergen/endotoxin exposures, and their relationships with allergic disease. Our research has demonstrated that exposure to indoor allergens and endotoxin is common but highly variable in U.S. homes. Our findings highlight the impacts and importance of environmental factors in human health and disease, including asthma. Although our focus continues to be on asthma/allergy-related outcomes, we have extended interest into other health outcomes as well as areas of methodological research in epidemiology. Despite the growing popularity of machine learning as a research tool, few studies have investigated implications of implementing machine learning approaches with complex survey data. We used data from NHANES and simulations of different design scenarios to assess the impact of accounting for sampling weights in gradient boosting, a powerful ensemble classification algorithm used on its own or as a component of the popular SuperLearner in R software. To evaluate the role of hyper-parameter tuning, we performed all analyses using the default hyper-parameters as well as hyper-parameters selected for each simulation from a randomized search of the hyper-parameter space in a massively parallelized computing environment (NIH Biowulf). This high-performance cluster, which ranks within the top 500 most powerful computers in the world, enabled us to use 10,000 simultaneous CPUs to fit approximately 2.2 billion tree models and obtain a high level of confidence in model cross-validation and bootstrapped confidence intervals. We demonstrated that models configured using complex survey data without sampling weights may not accurately reflect prediction in target populations, dependent on sample size and other analytic properties. We also demonstrated that, in the absence of software for configuring weighted algorithms, a post-hoc re-calculation of model performance with weighted observed outcomes might more validly represent prediction in target populations than ignoring weights entirely. Our findings underscore the need for further research; as the popularity of machine learning keeps increasing, failure to give more attention to appropriately analyzing complex survey data may represent a missed opportunity to leverage these novel and powerful approaches. In another collaborative project, we investigated prevalence of antinuclear antibodies, the most common biomarker of autoimmunity, in the general U.S. population. The findings from NHANES demonstrated an increasing trend in the antibody prevalence over a 25year time span, with a greater increase in recent years. Although the rate and timing of this increase varied by subgroups, both age and sex were consistently associated with prevalence of antinuclear antibodies. Increases over time were most marked in adolescents, men, and non-Hispanic White participants. The results provide valuable epidemiolocal insights on autoimmune disorders and will help design further studies to better understand factors underlying the increased prevalence and causes of autoimmunity. We continue to study the complex relationships between allergen exposures, allergic sensitization, and disease in more detail, as the NHANES data allow for the investigation of many interesting relationships. Our research will lead to a better understanding of the characteristics of environmental exposures, such as indoor allergen and endotoxin exposures, and their role in allergic disorders, which in turn provides insights into development of effective environmental intervention approaches for the management of allergic diseases such as asthma.