There are several basic prerequisites for the risk assessment of combined exposures to pesticides and dioxins using human health effects as the endpoint. First, all the target chemical substances exert the same toxicity to humans through the same mechanisms. Second, there is a linear dose–response relationship between the toxicity and effects of individual chemicals. With these two prerequisites, the effects of combined exposures are estimated as the sum of the toxicities of individual chemicals. For example, the toxicities of dioxins are calculated using their toxic equivalent quantities (TEQ) by considering the assigned toxic equivalent factor (TEF) of 2,3,7,8-tetrachlorodibenzo-p-dioxin (2,3,7,8-TCDD) set individually from their isomers and homologs. In conventional epidemiological studies, when the impact of each of multiple chemical substances is examined, methods such as multiple regression analysis or using a generalized linear model (GLM) have been used on the basis of the same prerequisites. However, in practice, some of the chemicals exhibit collinearity in their effects or do not show a linear dose–response relationship. In recent years, there have been several methods developed in the field of machine learning being applied to epidemiological research. Typical examples were methods using Bayesian kernel machine regression (BKMR) and weighted quantile sum (WQS), and the shrinkage method, i.e., using the least absolute shrinkage and selection operator (Lasso) and elastic network model (ENM). In the future, while taking into account the findings of experimental studies in biology, epidemiology, and other fields, it is expected that various methods will be applied and selected.
