Novel Statistical Approach to Handle Limit of Detection in Environmental Mixture Analysis - Dr. Zhao
Biostatistics Branch Seminar Series
November 6, 2019 | 10:30 AM – 11:30 AM
NCI Shady Grove 3E032/034 Rockville, MD
Speaker
Shanshan Zhao, Ph.D.
Principal Investigator
Biostatistics & Computational Biology Branch
National Institute of Environmental Health Sciences
Abstract
In environmental research, it is important to study the impact of chemical mixtures on human health since human beings are exposed to a variety of chemicals, through food, water, air and consumer products. A major challenge in such analysis is the limit of detection (LOD) issue that chemicals below certain concentration cannot be detected. Conventional approaches to deal with LOD include complete-case analysis, which excludes subjects with chemicals below LOD, and substitution method, which replaces values below LOD with an arbitrary value such as LOD/√2. These two approaches could result in either efficiency loss or biases. In studies relating chemical mixture to health outcomes, the efficiency loss and bias can be quite dramatic when multiple chemical measures suffer from LOD.
In this project, we considered the chemicals with LOD issue as survival data with left-censoring. We modeled these correlated chemical measures using multiple accelerated failure times (AFT) models and estimated the joint distribution of the error terms nonparametrically. We further assumed a generalized linear model to relate the chemical mixture to the health outcome. The joint likelihood and is optimized to recover the effect of chemicals on health outcome, while accounting for LOD of multiple chemicals. We conducted extensive numerical studies to understand the performance of our proposed method. We further applied this approach to a subset of the LIFECODES cohort of 280 pregnant women to investigate the relationship between 17 urinary metals, which may subject to LOD, and their 8-isoprostane, a urinary oxidative stress marker. Our results suggest appropriate statistical methods to handle LOD in the mixture setting could result in very different results when compared with conventional approaches.
**The mission of the Biostatistics Branch (BB) is to be an outstanding biostatistics unit that can contribute to the understanding of cancer etiology and to improve public health by the development and application of quantitative methods. The BB Investigators develop statistical methods and data resources to strengthen observational studies, intervention trials, and laboratory investigations of cancer.**