调解
缺少数据
计算机科学
因果推理
估计员
贝叶斯概率
环境流行病学
环境数据
计量经济学
数据挖掘
统计
医学
机器学习
环境卫生
人工智能
数学
生物
法学
生态学
政治学
作者
A. Wang,Katrina L. Devick,J.F. Bobbs,Ana Navas‐Acién,Brent A. Coull,Linda Valeri
出处
期刊:Environmental health perspectives
[Environmental Health Perspectives]
日期:2020-10-26
卷期号:2020 (1)
被引量:2
标识
DOI:10.1289/isee.2020.virtual.p-0480
摘要
Background: Bayesian kernel machine regression (BKMR) is becoming a popular approach for studying the joint effect of environmental mixtures on health outcomes allowing for variable selection, which is particularly useful when continuous exposures display moderate correlations. An R package, bkmr, has been developed to implement this method. Recently, BKMR-causal mediation analysis (BKMR-CMA) has been proposed to estimate direct and indirect effects of mixtures through a hypothesized mediator. However, the current bkmr R package only allows the output of total effects of the mixtures. Method: We introduce two new commands within the bkmr package 1) BKMR-CMA: a command that allows the estimation of direct and indirect health effects of multiple environmental exposures through a single mediator; 2) BKMR-MI: a command that is used for valid estimation of environmental mixture effects and evaluation of uncertainty in the presence of missing data, which are imputed using multiple imputation techniques. The new commands also produce effective visualizations of the estimated causal effects and dose-response relationships. Results: The new computational approaches complement the existing bkmr package by allowing the study of direct and indirect effects and accounting for missing data in the estimation of the exposure effects. We illustrate the use of these novel commands in an environmental epidemiology study investigating the effects of metal mixtures on cardiovascular outcomes through a cardiovascular biomarker. Conclusion: Both BKMR-CMA and BKMR-MI are available on Github, and are expected to facilitate the conduct of reproducibility of mediation analysis in environmental epidemiology research.
科研通智能强力驱动
Strongly Powered by AbleSci AI