Functional Methods for Radiation Exposure and Biomarker Data

辐射效应研究基金会（RERF）提供了关于原子弹幸存者的非常丰富的数据。虽然一些测量误差方法已被应用于调整RERF数据的辐射测量误差，这些现有的方法通常依赖于某些参数测量误差的假设。因此，重要的是进一步开发不需要这些难以检验的参数假设的半参数或非参数方法。剂量测定数据可被视为未观测到的潜在辐射照射的替代变量。生物标志物（如具有稳定染色体畸变的细胞百分比）可被视为未观察到的辐射剂量的一种工具变量。从约120，000名原子弹幸存者的确定队列（寿命研究，LSS，队列）中，约4，000名具有DS 02辐射剂量估计值、稳定染色体畸变数据和心血管疾病、胃癌、肺癌或乳腺癌等疾病结局数据的子队列将构成校准样本。通过使用校准样品的数据，我们可以估计整个LSS的辐射剂量响应，并对DS 02剂量估计的不确定性进行适当调整。这里的一个重要结果是，测量误差标准差将没有假设值，而是将从数据中估计，即使数据不包括重复测量或辐射剂量估计。在该提案中，感兴趣的回归问题将有一个主要队列，该队列对所有受试者进行辐射估计，但在子队列中提供多个辐射相关变量。这一建议的具体重点包括：（一）当暴露估计值受到经典加性误差的影响，并且某些受试者的生物标志物数据可用时，逻辑回归剂量反应模型的方法。(ii)当暴露估计值受到经典加性误差的影响且某些受试者的生物标志物数据可用时，生存剂量-反应模型的方法。(iii)当暴露估计值受到Berkson和经典误差的混合影响时，logistic和考克斯回归剂量-反应模型的方法以及某些受试者的生物标志物数据可用。新方法也将应用于妇女健康倡议营养生物标志物研究的双重标记水数据。所提出的方法将有一般的应用程序，任何分析的风险，疾病的关系，其中暴露测量误差和潜在的仪器变量可用于校准样品。 公共卫生相关性：我们建议使用半参数或非参数的方法来调整辐射剂量测量误差的影响，估计辐射剂量反应的健康影响的幸存者的广岛和长崎的原子弹爆炸。我们提出的方法将稳定的染色体畸变数据作为一个工具变量。所提出的方法也将被应用到双重标记的水数据从营养生物标志物研究的妇女健康倡议，并将有一般的应用程序，任何分析的风险与疾病的关系，其中暴露测量误差和潜在的工具变量可用于子样本。

DESCRIPTION (provided by applicant): The Radiation Effects Research Foundation (RERF) provides very rich data on the atomic bomb survivors. Although some measurement error methods have been applied to adjust for radiation measurement error for RERF data, these existing methods generally rely on certain parametric measurement error assumptions. Therefore, it is important to further develop semiparametric or nonparametric methods that do not need these parametric assumptions that are difficult to test. Dosimetry data may be considered as a surrogate variable for the unobserved underlying radiation exposure. A biomarker such as percentage of cells with stable chromosome aberrations can be treated as a type of instrumental variable for the un- observed radiation dose. From a defined cohort of about 120,000 A-bomb survivors (the Life Span Study, LSS, cohort), the subcohort of about 4,000 who have DS02 radiation dose estimates, stable chromosome aberration data, and outcome data for diseases such as cardiovascular disease, stomach cancer, lung cancer, or breast cancer, will comprise the calibration sample. By using data from the calibration sample, we can estimate radiation dose responses for the entire LSS, with an appropriate adjustment for the uncertainty in DS02 dose estimates. An important result here is that the measurement error standard deviation will not have an assumed value, but rather will be estimated from the data, even though the data do not include replicate measurements or estimates of radiation doses. In the proposal, the regression problem of interest will have a main cohort that has radiation estimation for all subjects, but multiple radiation-related variables available in a subcohort. Specific foci of this proposal include: (i) Methods for logistic regression dose-response models, when the exposure estimates are subject to classical additive errors and biomarker data are available for some subjects. (ii) Methods for survival dose-response models, when the exposure estimates are subject to classical additive errors and biomarker data are available for some subjects. (iii) Methods for logistic and Cox regression dose-response models, when the expo- sure estimates are subject to mixtures of Berkson and classical errors and biomarker data are available for some subjects. The new methods will also be applied to the doubly labeled water data from the Nutritional Biomarker Study of the Womens Health Initiative. The proposed methods will have general applications to any analysis of exposure-disease relationships in which exposures are measured with error and potential instrumental variables are available for a calibration sample. PUBLIC HEALTH RELEVANCE: We propose to use semiparametric or nonparametric methods to adjust for the effects of radiation dose measurement error on the estimation of radiation dose responses for health effects in survivors of the atomic bombings of Hiroshima and Nagasaki. Our proposed approaches will treat stable chromosome aberration data as an instrumental variable. The proposed methods will also be applied to the doubly labeled water data from the Nutritional Biomarker Study of the Womens Health Initiative, and will have general applications to any analysis of exposure-disease relationships in which exposures are measured with error and potential instrumental variables are available for a subsample.