Biopsychosocial Correlates of Edge-community Entropy and Modularity Across the Adult Lifespan

1171 

Abstract Submission 

Anita Shankar¹, Madhura Phansikar¹, Richard Betzel², Ruchika Prakash¹

¹The Ohio State University, Columbus, OH, ²Indiana University, Bloomington, IN

Anita Shankar  
The Ohio State University
Columbus, OH

Madhura Phansikar, PhD  
The Ohio State University
Columbus, OH

Richard Betzel  
Indiana University
Bloomington, IN

Ruchika Prakash, PhD  
The Ohio State University
Columbus, OH

Contemporary models of cognitive aging emphasize the complex relationship between biological, environmental, and lifestyle factors as determinants of cognitive and brain health1-3. However, few studies have examined demographic and biopsychosocial determinants of functional brain networks across the lifespan, and none, to our knowledge, have attempted to comprehensively account for collinearity among potential predictors. In our previous work, we used an edge-centric approach to demonstrate that community entropy, a measure of nodal despecialization, increases significantly across the lifespan and negatively impacts fluid cognitive performance4,5. In the current study, we used elastic-net regression to identify candidate biopsychosocial features that correlate with entropy cross-sectionally and potentially mediate these age-related entropy changes.

The current study used a subset of data from the Human Connectome Project 2.0 Lifespan Release from 528 individuals between the ages of 35 and 100 years with task fMRI, physical and emotional health, and demographic data for 33 literature-informed predictors. Entropy was calculated at the level of individual nodes as described in previous work4,5 and averaged across the whole brain. Elastic net regression was used to identify features associated with entropy from four conceptual groups: demographic factors, stress, physical health, and emotional well-being. We used a 70/30 training-test set split, using the training set to tune the model's hyperparameters and select predictors, and using the test set to determine the model's generalizability to unseen data.

The final model identified 11 biopsychosocial factors, in addition to demographic factors of age, sex, and education, that predicted entropy across the lifespan. Application of the final model to the test set yielded a significant correlation between predicted and observed entropy values (R = 0.58, p = 1.33e-15; Figure 1a), which includes variance accounted for by age and variance accounted for by mediators of the relationship between age and entropy. Significant predictors above and beyond age consisted of, in order of importance: body mass index, sex, instrumental social support, average hours of sleep, visual acuity, hs-C reactive protein, perceived stress, life meaning and purpose, diastolic blood pressure, education, triglyceride levels, hemoglobin A1C, and social isolation. Removing the variance contributed by age completely, biopsychosocial factors still accounted for a significant portion of the variance in entropy (R=0.24, p = 0.002), suggesting certain factors act on entropy independent of the process of aging. We also probed how these features interact with age by running a second model that included the original features and their age-interaction terms. In the interaction model (R = 0.59, p =6.17e-16), sex exhibited the most significant moderation effect, with males demonstrating higher entropy across all ages and greater lifespan increases in entropy than females, starting at age 54 (Figure 1b). Finally, we performed these analyses using the network property whole-brain modularity and found similar model performance (R = 0.57, p = 6.26e-15) and feature selection differed by only a single predictor (systolic blood pressure instead of diastolic blood pressure), suggesting these biopsychosocial factors reliably influence multiple measures of whole-brain network organization.

In the current study, we examined biopsychosocial correlates of whole-brain entropy and modularity across the adult lifespan. We present evidence that certain demographic, metabolic, cardiovascular, and psychosocial factors have system-wide influences on the function of the human brain and identify candidate age-dependent and independent predictors of network measures across the lifespan. Given previous research implicating modularity and entropy's relationship with cognition, these results may inform treatment targets for cognitive health in aging.

Aging ¹

fMRI Connectivity and Network Modeling ²

Aging

Machine Learning