Structural Brain Correlates of Loneliness among Older Adults

Abstract

Ample evidence indicates that loneliness in old age is associated with poor bodily and mental health. However, little is known about structural cerebral correlates of loneliness in healthy older adults. We examined such correlates in a magnetic resonance imaging (MRI) subsample of 319 older adults aged 61 to 82 years drawn from the Berlin Aging Study II. Using voxel-based morphometry (VBM) and structural equation modeling (SEM), latent hierarchical regression analyses were performed to examine associations of (i) loneliness, (ii) a range of covariates, and (iii) loneliness by covariate interactions with latent brain volume estimates of brain structures known to be involved in processing, expressing, and regulating emotions. Results from whole-brain VBM analyses showed that individuals with higher loneliness scores tended to have smaller gray matter volumes in three clusters comprising (i) the left amygdala/anterior hippocampus, (ii) the left posterior parahippocampus and (iii) the left cerebellum. Significant associations and interactions between loneliness and latent factors for the amygdala and the hippocampus were confirmed with a region-of-interest (ROI)-based approach. These findings suggest that individual differences in loneliness among older adults are correlated with individual differences in the volumes of brain regions that are central to cognitive processing and emotional regulation, also after correcting for confounders such as social network size. We discuss possible mechanisms underlying these associations and their implications.

Figure 1

Left: Brain regions showing a negative association (red clusters) between loneliness and gray matter volume (p < 0.001; corrected for multiple comparisons; cluster extent threshold k > 97). Upper right panel: Correlation between the individual GM volumes in the first cluster and loneliness score. Lower right panel: Correlation between the individual GM volumes in the second cluster and loneliness score (p < 0.05; all scores are residualized for covariates).

Figure 2

A simplified illustration of the CFA of the latent brain factor model. Model fit is good (χ2 56 = 174.3, RMSEA = 0.058 (CI: 0.044–0.073); CFI = 0.979; SRMR = 0.02). Latent brain factors are drawn in circles. Squares represent observed variables. Single-headed arrows represent significant factor loadings, and double-headed arrows represent covariances. Double-headed arrows with both heads pointing on a manifest variable represent the variance of a variable.

Figure 3

Depiction of a simplified SEM (χ2 159 = 249.8, RMSEA = 0.047; CFI = 0.965) showing only all significant main effects of loneliness (lone) and number of confidants (social) as well as the interaction of age (lone*age), sex (lone*sex), education (lone*sex), morbidity (lo*orb), depression (lone*dpr) and optimism (lone*op) with loneliness, which are regressed on the intercorrelated latent brain factors (circles, intercorrelations not shown). Squares represent observed variables. Double-headed arrows with both heads pointing on a manifest variable represent the variance of a variable. Loneliness, depressive affect, openness and educations are represented as mean scores (squares). dlPFC = dorso-lateral-prefrontal cortex, mOFC = medio-orbito-prefrontal cortex, ACC = anterior cingulate cortex, AMY = amygdala, HC = hippocampus, I = insula, and NAcc = nucleus accumbens. Regression coefficients represent standardized estimates.

Figure 4

Exemplary plot of the interaction between loneliness, age, and left amygdala volume. ROI are extracted from CAT12 preprocessed gray matter maps. For illustration, the participants were divided into two age groups by applying a 69.8 split for age and loneliness across the sample (median age = 69.8 years; younger older adults  < 69.8 years, older adults = >69.8 years; low loneliness = −1 SD, high loneliness = +1 SD).