A Bayesian life-course linear structural equations model (BLSEM) to explore the development of body mass index (BMI)...

We have developed a novel Bayesian Linear Structural Equations Model (BLSEM) with variable selection priors (available as an R package) to build directed acyclic graphs to delineate complex variable associations and pathways to BMI development. Conditional on standard assumptions used in causal inference, the model provides interpretable estimates with uncertainty for natural direct, indirect (mediated) and total effects.

We showcase our method using data on 4119 offspring followed from the pre-pregnancy period to age 46 years (y) in a Finnish population-based birth cohort. The BLSEM enabled efficiently to analyse all available data over the long-time span, identifying factors to distil potential causal pathways contributing to adult BMI development. All of the associations between early childhood and adolescence variables with adult BMI at 46 y (BMI46) were indirect via multiple paths. For example, maternal prepregnancy BMI, smoking and socioeconomic position are associated with BMI46 through 35, 31 and 26 paths. Another notable feature was that the contribution of very early life factors, particularly prenatal, was captured by growth patterns along childhood, which were the strongest early predictors of middle age BMI46 (the age at adiposity rebound (AgeAR), early growth parameters between the AgeAR to 11 y). BMI and blood pressure measured 15 y earlier also predicted BMI46, all other factors held constant. Genetic predisposition by the polygenic risk score for BMI showed an indirect effect that became apparent at AgeAR and thereafter.

The Bayesian approach we present and the BLSEM software developed advances methodologies for the analysis of complex, multifaceted life-course data prior to the estimation of potential causal pathways. Our results, although exploratory in nature, suggest that the effective interventions to tackle adverse BMI development could be designed throughout childhood, though the period by AgeAR may be paramount. We feature the importance of integrated life-course analyses that help to understand the contribution of life-stage factors of development.