Effects of measurement errors on relationships between resting-state functional connectivity and psychological phenotypes

Abstract

Abstract Recent neuroscientific studies have focused on interindividual relationships between resting-state functional connectivity (RSFC) and psychological phenotypes using large datasets with repeated measurements, including the Human Connectome Project (HCP). However, previous studies on RSFC-phenotype relationships have failed to differentiate trait, state, and error effects of RSFC. Latent functional connectivity, which can be estimated in structural equation model (SEM), can be useful in finding RSFC-phenotype relationships controlling state and error effects. We also accounted for measurement errors in psychological phenotypes at the test-, subscale-, or item-level. This study investigates: (i) how measurement errors, including state effects, weaken the associations between RSFC and psychological phenotypes, including cognition, mental health, and personality, and influence sample size planning and (ii) predictive accuracy on the phenotypes from RSFC, using SEM. We found that the extent of the weakening of RSFC-phenotype associations ranged from 15.3 to 33.8% across the phenotypes, and they were higher in sensorimotor networks than in higher order cognitive networks. Importantly, measurement errors can lead to requirement of about double sample size to find RSFC-phenotype associations in general. Factor scores of RSFC enhanced the coefficients of determination under some conditions. Future studies should explore more effective predictive methods by accounting for measurement errors.

Keywords

• Resting-state functional connectivity
• Structural equation modeling
• Measurement error
• Human connectome project
• Prediction