Bayesian structural equation modeling for data from multiple cohorts

TL;DR

This paper introduces a Bayesian multi-group multi-level structural equation modeling approach to analyze the effects of prenatal alcohol exposure on cognition across multiple cohorts with different measures.

Contribution

It develops a Bayesian SEM framework capable of handling multiple cohorts with different outcomes, allowing for dose-response and change point analysis in PAE effects.

Findings

Bayesian SEM successfully models cognition across cohorts.

Identifies PAE levels linked to increased adverse effects.

Detects potential dose-response change points.

Abstract

While it is well known that high levels of prenatal alcohol exposure (PAE) result in significant cognitive deficits in children, the exact nature of the dose response is less well understood. In particular, there is a pressing need to identify the levels of PAE associated with an increased risk of clinically significant adverse effects. To address this issue, data have been combined from six longitudinal birth cohort studies in the United States that assessed the effects of PAE on cognitive outcomes measured from early school age through adolescence. Structural equation models (SEMs) are commonly used to capture the association among multiple observed outcomes in order to characterise the underlying variable of interest (in this case, cognition) and then relate it to PAE. However, it was not possible to apply classic SEM software in our context because different outcomes were measured in the six studies. In this paper we show how a Bayesian approach can be used to fit a multi-group multi-level structural model that maps cognition to a broad range of observed variables measured at multiple ages. These variables map to several different cognitive subdomains and are examined in relation to PAE after adjusting for confounding using propensity scores. The model also tests the possibility of a change point in the dose-response function.