Enhancing causal inference in population-based neuroimaging data in children and adolescents

Recent years have seen the increasing availability of large, population-based, longitudinal neuroimaging datasets, providing unprecedented capacity to examine brain-behavior relationships in the neurodevelopmental context. However, the ability of these datasets to deliver causal insights into brain-behavior relationships relies on the application of purpose-built analysis methods to counter the biases that otherwise preclude causal inference from observational data. Here we introduce these approaches (i.e., propensity score-based methods, the 'G-methods', targeted maximum likelihood estimation, and causal mediation analysis) and conduct a review to determine the extent to which they have been applied thus far in the field of developmental cognitive neuroscience. We identify just eight relevant studies, most of which employ propensity score-based methods. Many approaches are entirely absent from the literature, particularly those that promote causal inference in settings with complex, multi-wave data and repeated neuroimaging assessments. Causality is central to an etiological understanding of the relationship between the brain and behavior, as well as for identifying targets for prevention and intervention. Careful application of methods for causal inference may help the field of developmental cognitive neuroscience approach these goals.