Heterogenous factors modulate social cognition across diverse settings in brain health and disease

Abstract

AbstractBackgroundSocial cognition dysfunction in neurodegenerative diseases such as the behavioral variant frontotemporal dementia (bvFTD) and Alzheimer’s Disease (AD) is typically ascribed to the underlying pathophysiological processes spanning specific atrophy and connectivity alterations. However, social cognition performance might be modulated by a set of more heterogenous, non‐specific factors, including demographics (sex, age, education, socioeconomic status), general cognitive and executive abilities, and behavior inside the scanner (i.e., movements), but their relative contribution is poorly understood. Here, we develop a machine learning approach to characterize multimodal factors that explain the inter‐individual heterogeneity of social cognition in a diverse multicentric sample.MethodA total of 998 individuals > 50 years from international research networks in high‐ and middle‐income countries participated in this study: 102 with bvFTD, 339 with AD, 96 with mild cognitive impairment (MCI), and 461 with preserved cognition (with and without cognitive complaints). Social cognition was assessed with the Mini Social Cognition and Emotional Assessment (MiniSEA) battery, which includes tests of facial emotion recognition and theory of mind. After data harmonization, we implemented support vector regressions combined with bootstrapping to predict social cognition outcomes from multimodal features, including diagnosis, demographics, cognitive and executive functions, grey matter volume, fMRI resting‐state network connectivity, and movement artifacts, and assessed their relative contribution to performance (Figure 1). Models were assessed with the coefficient of determination R², Fisher’s F test, Cohen’s F², 95% confidence intervals, and p‐values.ResultGeneral cognitive and executive functions and years of education consistently emerged among the top predictive features of social cognition performance and explained more variance than diagnosis. Other features making significant contributions were: sex, the gray matter volume of relevant regions, and in‐scanner movements. Age and resting‐state network connectivity did not make any significant contribution to performance heterogeneity (Figure 2).ConclusionSocial cognition performance was more strongly predicted by a set of heterogenous individual differences than by neurodegenerative diagnoses and associated brain signatures. Diagnoses might relate to complex profiles. Results highlight a critical modulator role of heterogeneity on social cognition, with implications for the neuropsychological assessment in brain health and disease across diverse, multi‐setting populations.