Background the exposome constitutes a promising framework for better understanding the effect of the totality of an individual’s exposure history, both in terms of the composition and timing of exposures. Although there has been much progress in the development of suitable approaches for screening exposures that possibly relate to an health outcome, a statistical framework that fully accounts for the inherent complexities of a comprehensive exposome analysis is only just emerging. Such a framework needs to be able to address a multitude of methodological issues, such as how to adjust inferences to account for the multiplicity of hypotheses being tested, how to deal with highly correlated exposures, how to include interactions and possible non-linear exposure-response relations, and how to account for differential exposure misclassification across compounds. In addition, when the dynamics of exposure and exposure-outcome relations over time are important, exposure lagging and time-window analyses add to the complexity of the problem.

Discussion although statistical methods are available that account for the multiplicity of hypotheses being tested, those based on simple univariate testing are unable to cope with the strong correlations that are likely to be present between different exposures or between different exposure windows for a single exposure, while multi-variable methods may suffer from low power due to the resulting high collinearity. More sophisticated variable selection methods developed for machine learning tend to fare better, but have so far been applied almost exclusively under the assumption of linear exposure-response relations, and without any concern for the dynamics of these relations.

Conclusion there is a clear need for modern multivariable statistical methods in exposome research, but it is unclear to what extent existing methods can accommodate the full breadth of an exposome analysis. In this presentation we will explore the analyses needs and current available tools.