Statistical software for analyzing the health effects of multiple concurrent exposures via Bayesian kernel machine regression.
Bobb JF, Claus Henn B, Valeri L, Coull BA. · 2018
Read Statistical software for analyzing the health effects of multiple concurrent exposures via Bayesian kernel machine regression. by Bobb JF, Claus Henn B, Valeri L, Coull BA. (2018) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc6102907.txt.
Introduction
Bobb and co-authors release software for studying health effects of exposure mixtures. Their article appeared in Environmental Health. People face many pollutants at once. Their joint effects can bend and interact. A method called Bayesian kernel machine regression handles that tangle.
Use of the method had stalled for lack of tools. The new bkmr package runs in R. It adds plots for high-dimensional response surfaces. A probit version handles yes-no outcomes. A faster approximation cuts runtime sharply. Worked examples come with full code. The method also picks which exposures matter most. Readers get a usable kit for mixture epidemiology.
Catalog record
| CREATOR | Bobb JF, Claus Henn B, Valeri L, Coull BA. |
|---|---|
| YEAR | 2018 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 45 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Statistical software for analyzing the health effects of multiple concurrent exposures via Bayesian kernel machine regression By Jennifer F Bobb, Birgit Claus Henn, Linda Valeri, Brent A Coull Published in: Environmental Health DOI: 10.1186/s12940-018-0413-y ABSTRACT Abstract Background Estimating the health effects of multi-pollutant mixtures is of increasing interest in environmental epidemiology. Recently, a new approach for estimating the health effects of mixtures, Bayesian kernel machine regression (BKMR), has been developed. This method estimates the multivariable exposure-response function in a flexible and parsimonious way, conducts variable selection on the (potentially high-dimensional) vector of exposures, and allows for a grouped variable selection approach that can accommodate highly correlated exposures. However, the application of this novel method has been limited by a lack of available software, the need to derive interpretable output in a computationally efficient manner, and the inability to apply the method to non-continuous outcome variables. Methods This paper addresses these limitations by (i) introducing an open-source software package in the R programming language, the bkmr R package, (ii) demonstrating methods for visualizing high-dimensional exposure-response functions, and for estimating scientifically relevant summaries, (iii) illustrating a probit regression implementation of BKMR for binary outcomes, and (iv) describing a fast version of BKMR that utilizes a Gaussian predictive process approach. All of the methods are illustrated using fully reproducible examples with the provided R code. Results Applying the methods to a continuous outcome example illustrated the ability of the BKMR implementation to estimate the health effects of multi-pollutant mixtures in the context of a highly nonlinear, biologically-based dose-response function, and to estimate overall, single-exposure, and interactive health effects. The Gaussian predictive process method led to a substantial reduction in the runtime, without a major decrease in accuracy. In the setting of a larger number of exposures and a dichotomous outcome, the probit BKMR implementation was able to correctly identify the variables included in the exposure-response function and yielded interpretable quantities on the scale of a latent continuous outcome or on the scale of the outcome probability. Conclusions This newly developed software, integrated suite of tools, and extended methodology makes BKMR accessible for use across a broad range of epidemiological applications in which multiple risk factors have complex effects on health. Keywords: Multiple exposures, Mixtures, Exposure-response, Variable selection, Health risk estimation BACKGROUND Estimating the health effects of several concurrent exposures is of increasing interest in epidemiology. For example, in environmental health interest lies in estimating the impacts of multi-pollutant mixtures, such as air pollution [1], toxic waste [2], and persistent organic chemicals [3]. Although studies have traditionally focused on estimating the health impacts of individual exposures, it is increasingly being recognized that populations are exposed to a wide range of factors across multiple domains, including environmental stressors and genetic and psychosocial determinants, and that these factors should be considered in conjunction [4]. A major barrier to studying the joint effects of many exposures concurrently is the lack of established statistical methods and corresponding software. Estimating the health effects of environmental mixtures is challenging because (i) exposures often have nonlinear and non-additive (eg, interactive) relationships with health outcomes, (ii) a high-dimensional vector of exposures may lead to poorly fitting regression models as the number of exposures increases relative to the number of observations in the dataset, and (iii) exposures are often highly correlated. Additionally, there are often several objectives of a multi-exposure health effect analysis, which may include estimating the overall effect of the mixture, identifying individual components that are responsible for the health effects of the mixture, visualizing the exposure-response function, and detecting interactions among pollutants [5]. Several statistical methods have been proposed for estimating the health effects of multiple exposures, including machine learning methods such as random forests [6]; clustering methods and other dimensional reduction methods such as principal components analysis, factor analysis, and structural equation models; and regression penalization methods such as the lasso [7]. However, these methods have typically addressed some but not all of the challenges and/or scientific objectives described above. Reviews of prior methods and their limitations, as well as systematic comparisons of the performance of selected methods, have been published previously [1, 8–12]. Recently, we developed a new approach for estimating the joint health effects of multivariate exposures, Bayesian kernel machine regression (BKMR), that simultaneously addresses the challenges and scientific objectives described above [13]. First, through use of a kernel function, this approach estimates the multivariable exposure-response function in a flexible way that allows for nonlinear and non-additive effects, while adjusting for covariates including potential confounding factors. Second, the approach simultaneously incorporates variable selection on the (potentially large number of) exposures in a way that controls for multiple testing [14]; this enables a parsimonious representation of the exposure-response function. Third, we developed a hierarchical variable selection approach that addresses the issue of multicollinearity by first classifying highly correlated exposures into groups, and then simultaneously conducting variable selection on the groups of correlated exposures as well as on the individual exposures within each group. In our prior methodological work describing BKMR [13, 15], we conducted a comprehensive evaluation of the performance of this approach. Through simulation studies based on real-world datasets, we found that (i) BKMR could well estimate exposure-response functions that included both nonlinear and non-additive effects, (ii) BKMR could identify important mixture components through variable selection, and (iii) the hierarchical variable selection approach could detect important groups of highly correlated exposures even in situations where individual components could not be identified. Additionally, BKMR has been applied previously in both toxicological and epidemiological studies leading to scientific insights that were not uncovered using standard regression approaches [15, 16]. Several important gaps limit the applicability of statistical methods for estimating the health effects of multi-pollutant mixtures in environmental health studies. These include a lack of software applying new methodology, data-generating scenarios with complex features (e.g., clustered outcome data), and the need for computationally efficient algorithms that yield correct results in a fraction of the time. A particular challenge of studies estimating the health effects of mixtures is the need to visualize the high-dimensional exposure-response function and to conduct inference in the presence of possible nonlinear and interactive associations of the exposures with the health outcome. Here we provide several contributions addressing these gaps. First, we introduce an open-source software package (bkmr) [17] that implements the new BKMR approach for studying mixtures within the R statistical program [18]. This software provides a general, user-friendly implementation of BKMR, along with a suite of functions for processing model output to enable investigators to address the multifold objectives of a multi-exposure heath effect analysis. Second, we demonstrate methods for characterizing high-dimensional exposure-response functions, including visualizing the exposure-response relationship, and estimating scientifically relevant summaries, such as overall, single-exposure, and interactive health effects. Third, we present extensions to BKMR to enable model fitting to a broader class of applications, including a probit regression implementation for binary outcomes, and the ability to include a random intercept to account for correlated outcome data. Finally, we illustrate how a fast version of BKMR that utilizes a Gaussian predictive process approach can substantially speed up the model fitting. All of the examples used to illustrate these methods are fully reproducible with the provided R code. METHODS Overview of BKMR