Causal modelling of multivariate outcomes from developmental toxicity experiments

Abstract

Exposure to developmental toxicants may cause foetal malformations, increase foetal death or resorption and reduce foetal weight. The correlations among these developmental endpoints have been reported, but their causal relationships have not been investigated. Structural equation models (path models) were applied to study the patterns of causation among the four developmental outcomes, number of viable foetuses, number of malformations, number of deaths/resorptions, and average foetal weight; these outcomes were modelled as response variables with the dose level and number of implants (litter size) modelled as independent variables. Three hypothesized path models were fitted to developmental toxicity data from a study of the herbicide 2,4,5-T exposure in mice. One model, which hypothesized that foetal weight, implicitly a mechanism such as cellular growth, affects both malformations and death/resorption (equivalently, viability) with no causation relationship between malformation and death/resorption, described data consistently well for three strains of mice. The most consistent relationship found was a strong direct effect of foetal weight on malformations, and the correlation between malformations and death/resorption (or reduced viability) was likely spurious.