Large-Scale Open Science Simulated Data to Study the Development of Brain and Behavior Relationships: Epidemiologic, Clinical, and Neuroimaging Characteristics

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Both brain and behavior undergo considerable changes during adolescence, a time synonymous with the emergence several psychiatric disorders, including mood and psychotic disorders. Multiple studies have explored the interaction between brain and behavior during childhood and adolescence, with both mixed and convergent findings between the studies. One challenge is that different longitudinal models can be applied to both the same or similar dataset and yet yield different results. Without knowing the ground truth, the best modeling approach to understanding the brain/behavior interplay during development remains unclear. Within this context, simulated datasets can provide a mechanism to test many different neurodevelopmental features, not only regarding the interplay between brain and behavior, but also how specific epidemiologic factors can influence study findings. Simulated datasets have the ‘ground truth’ relationships embedded within the models used to create these datasets. Thus, the goal of this study was to assess the epidemiologic and clinical characteristics of 15 large simulated longitudinal neuroimaging datasets, independently created from five research groups from three different continents. We found that while most of the neuroimaging and behavioral trajectories in the simulated datasets reflected growth patterns identified in the literature, there was also considerable variability between how the groups modeled the epidemiologic, clinical, and neurodevelopmental growth patterns. The conclusion is that while simulation studies are an underused strategy to test assumptions related to neurodevelopmental disorders, simulation studies require iterative approaches, testing simulated against real-world data and challenging the different assumptions associated with the neurodevelopmental models, until there is convergence between actual and simulated datasets.