Structural Equation Modelling in the exploration and analysis of intrauterine environmental exposures with infant health effects.

Introduction: In epidemiology, generalized linear models are the main statistical methods used to explore associations. However, the use of other methods such as Structural Equation Modelling (SEM) is gradually increasing.

Objective: The aim of the study was to illustrate the use of SEM in the assessment of salivary cortisol concentration in infants as a biomarker of perinatal exposure to inorganic arsenic.

Material And Methods: This was a cohort study of pregnant women recruited from public health care centres in Arica, Chile, in 2013. Socio-demographic information and urine samples to assess inorganic arsenic were collected during the second trimester of pregnancy. Saliva samples were collected to assess cortisol in infants between 18-24 months of age. Four linear regression models (LRMs) and two SEMs were run to estimate the effect of prenatal exposure to inorganic arsenic on cortisol concentration in infants.

Results: According to LRMs and SEMs, prenatal exposure to inorganic arsenic and salivary cortisol were not associated. However, the association between maternal cortisol and cortisol in infants was statistically significant in all models; for each increase in standard deviation of the covariate Ln(maternal cortisol), the outcome Ln(cortisol in infant) increased by 0.49 units of variance in both SEMs.

Conclusions: LRMs and SEMs were useful to assess the effect of prenatal exposure to inorganic arsenic on cortisol in infants. However, SEM allowed the adjustment of estimations by an estimated latent that obtained the information about income, occupation, education and ethnicity in a more comprehensive way than achieved by LRM.