Du Ruofei, Luo Li, Hudson Laurie G, Nozadi Sara, Lewis Johnnye
Biostatistics Shared Resource, University of New Mexico Comprehensive Cancer Center, Albuquerque, NM, USA.
Department of Internal Medicine, University of New Mexico, Albuquerque, NM, USA.
J Appl Stat. 2022 Mar 5;50(8):1790-1811. doi: 10.1080/02664763.2022.2043254. eCollection 2023.
In a large-scale environmental health population study that is composed of subprojects, often different fractions of participants out of the total enrolled have measures of specific outcomes. It's conceptually reasonable to assume the association study would benefit from utilizing additional exposure information from those with a specific outcome not measured. Partial least squares regression is a practical approach to determine the exposure-outcome associations for mixture data. Like a typical regression approach, however, the partial least squares regression requires that each data observation must have both complete covariate and outcome for model fitting. In this paper, we propose novel adjustments to the general partial least squares regression to estimate and examine the association effects of individual environmental exposure to an outcome within a more complete context of the study population's environmental mixture exposures. The proposed framework takes advantage of the bilinear model structure. It allows information from all participants, with or without the outcome values, to contribute to the model fitting and the assessment of association effects. Using this proposed framework, incorporation of additional information will lead to smaller root mean square errors in the estimation of association effects, and improve the ability to assess the significance of the effects.
在一项由多个子项目组成的大规模环境卫生人群研究中，通常在总招募参与者中，有特定结局测量值的参与者比例各不相同。从概念上讲，假设关联研究能够利用来自那些未测量特定结局者的额外暴露信息，将会从中受益。偏最小二乘回归是一种用于确定混合数据暴露-结局关联的实用方法。然而，与典型的回归方法一样，偏最小二乘回归要求每个数据观测值在模型拟合时必须同时具有完整的协变量和结局。在本文中，我们对一般的偏最小二乘回归提出了新的调整方法，以便在研究人群环境混合物暴露的更完整背景下，估计和检验个体环境暴露与结局之间的关联效应。所提出的框架利用了双线性模型结构。它允许所有参与者的信息，无论有无结局值，都能为模型拟合和关联效应评估做出贡献。使用这个提出的框架，纳入额外信息将导致在关联效应估计中产生更小的均方根误差，并提高评估效应显著性的能力。
