Akademik Veri Yönetim Sistemi
26th International Conference on Computational Statistics, Giessen, Almanya, 27 - 30 Ağustos 2024, ss.1
The Benchmark Dose has become a popular tool for identifying an exposure level that is associated with a specified increased risk of an adverse health outcome. However, the classic approach does not apply in settings where the exposure of interest can be measured in several different dimensions. In the application that motivates our work, for example, it is thought that cognitive deficits associated with prenatal alcohol exposure depend not simply on average alcohol consumption during pregnancy, but on both the frequency of drinking days as well as the amount of alcohol consumed on each drinking day during pregnancy. In this work, we propose a flexible framework for benchmark analysis that allows a more nuanced assessment of risks associated with an exposure that can be measured in several dimensions. The method entails fitting a joint model for the effect of the exposures while adjusting for potential confounders via propensity scores. From this model we obtain a benchmark dose contour that relates the two continuous exposure variables to an outcome. Additionally, we use generalized additive models which yield a flexible dose-response surface without requiring specification of a parametric form for non-linear effects. We illustrate our method using data assembled from six U.S. cohort studies that measured maternal reports of alcohol use during pregnancy, along with longitudinal measurements of cognitive function in their offspring. While our results provide important scientific insights regarding adverse effects associated with various pre-natal drinking profiles, more generally the method may be useful in a broad range of settings involving exposures or mixtures of exposures that can be measured in several dimensions.