An overview of methods to address distinct research questions on environmental mixtures: an application to persistent organic pollutants and leukocyte telomere length

Abstract

Background Numerous methods exist to analyze complex environmental mixtures in health studies. As an illustration of the different uses of mixture methods, we employed methods geared toward distinct research questions concerning persistent organic chemicals (POPs) as a mixture and leukocyte telomere length (LTL) as an outcome. Methods With information on 18 POPs and LTL among 1,003 U.S. adults (NHANES, 2001–2002), we used unsupervised methods including clustering to identify profiles of similarly exposed participants, and Principal Component Analysis (PCA) and Exploratory Factor Analysis (EFA) to identify common exposure patterns. We also employed supervised learning techniques, including penalized, weighted quantile sum (WQS), and Bayesian kernel machine (BKMR) regressions, to identify potentially toxic agents, and characterize nonlinear associations, interactions, and the overall mixture effect. Results Clustering separated participants into high, medium, and low POP exposure groups; longer log-LTL was found among those with high exposure. The first PCA component represented overall POP exposure and was positively associated with log-LTL. Two EFA factors, one representing furans and the other PCBs 126 and 118, were positively associated with log-LTL. Penalized regression methods selected three congeners in common (PCB 126, PCB 118, and furan 2,3,4,7,8-pncdf) as potentially toxic agents. WQS found a positive overall effect of the POP mixture and identified six POPs as potentially toxic agents (furans 1,2,3,4,6,7,8-hxcdf, 2,3,4,7,8-pncdf, and 1,2,3,6,7,8-hxcdf, and PCBs 99, 126, 169). BKMR found a positive linear association with furan 2,3,4,7,8-pncdf, suggestive evidence of linear associations with PCBs 126 and 169, and a positive overall effect of the mixture, but no interactions among congeners. Conclusions Using different methods, we identified patterns of POP exposure, potentially toxic agents, the absence of interaction, and estimated the overall mixture effect. These applications and results may serve as a guide for mixture method selection based on specific research questions. Electronic supplementary material The online version of this article (10.1186/s12940-019-0515-1) contains supplementary material, which is available to authorized users.

Outdoor air pollution and terminal duct lobular involution of the normal breast

“…We used K -means clustering to identify groups of individuals with similar patterns of PM 2.5 component exposure. K -means is an unsupervised clustering method to identify a pre-specified number ( k ) of representative joint exposure values (centroids) that can be used to classify individuals into k groups according to which centroid is closest to an individual’s exposure [ 27 ]. We examined a plot of the proportion of within over total sums of squares by number of clusters and used cluster stability analysis to determine the smallest number of clusters with the highest Rand index [ 28 ].…”

Prenatal exposure to mixtures of persistent endocrine disrupting chemicals and early menarche in a population-based cohort of British girls

“…Bayesian kernel machine regression (BKMR) was used as a complementary mixture method to WQS regression. BKMR is a flexible semi-parametric technique that models the combined effects of different chemicals, while also allowing for nonlinearity and interactions among chemicals (55). This approach allows for the examination of independent effects of mixture members, interactions among them, and the overall mixture effect.…”

A family of partial-linear single-index models for analyzing complex environmental exposures with continuous, categorical, time-to-event, and longitudinal health outcomes

“…In addition, WQS and BKMR have been generalized to study environmental mixtures with several types of outcomes, such as WQS for longitudinal outcomes [15] and BKMR for time-to-event outcomes [16]. However, a general modeling framework that can alleviate the above limitations in environmental health research is still desired [17].…”