Psychological profiles associated with mental, cognitive and brain health in middle-aged and older adults

0
Psychological profiles associated with mental, cognitive and brain health in middle-aged and older adults

Participant characteristics

A total of 823 cognitively unimpaired participants from the Barcelona Brain Health Initiative (BBHI; including n = 750 with baseline mental, cognitive and/or brain health data and n = 533 with both magnetic resonance imaging (MRI) and cognitive data at follow-up) and 282 cognitively unimpaired participants from the Medit-Ageing study were included. Baseline demographic and clinical characteristics for each cohort are provided in Table 1. The BBHI cohort were younger (51.4 ± 7.0 versus 71.1. ± 5.9 years (mean ± s.d.); W = 10,331, P < 0.001, Wilcoxon rank-sum test) and included a larger proportion of men (48.5% versus 36.9%; χ2 = 11.60, P < 0.001, chi-squared test) than the Medit-Ageing cohort.

Table 1 Baseline demographic characteristics for the total sample and for each of the three latent profiles in BBHI and Medit-Ageing

Latent profile identification

The model fit statistics for the LPA, conducted separately in BBHI and Medit-Ageing, are presented in Supplementary Table 1. The Vuong–Lo–Medell–Rubin likelihood ratio test (VLMR-LRT) became non-significant for the four-profile solution in both samples, indicating that a three-profile solution was the best fitting. Increasing the number of profiles resulted in small class sizes (<5% of the total sample), alongside poorer fit in many of the other metrics, supporting three profiles as the most robust solution.

As shown in Fig. 1, we found three distinct profiles in both cohorts that demonstrated comparable aggregations of psychological characteristics. Compared to the other profiles, profile 1 was characterized by lower levels of positive or protective psychological characteristics, profile 2 by higher negative or psychological risk characteristics, and profile 3 by moderately high protective and moderately low risk characteristics. We labeled the three profiles as follows: ‘low protective’ (profile 1: BBHI, n = 196, 26.1%; Medit-Ageing, n = 59, 20.9%), ‘high risk’ (profile 2: BBHI, n = 149, 19.9%; Medit-Ageing, n = 64, 22.7%) and ‘well-balanced’ (profile 3: BBHI, n = 405, 54.0%; Medit-Ageing, n = 159, 56.4%). Descriptive statistics for each profile, in both cohorts, are presented in Table 1.

Fig. 1: Z-score distribution of psychological characteristics in the three profiles defined in the latent profile analysis for BBHI and Medit-Ageing.
figure 1

Psychological factors that have been associated with dementia risk in the existing literature are in shades of red, and psychological factors that have been associated with protection against dementia are in shades of green.

BBHI and Medit-Ageing cross-sectional analyses

The results from linear regression models examining the cross-sectional associations between psychological profile membership and mental health, cognition, lifestyle and cortical thickness after adjusting for covariates (age, sex, education and (in Medit-Ageing) study) are described below, presented in Supplementary Table 2 and displayed in Figs. 2–4.

Fig. 2: Association between psychological profiles and mental health in the BBHI and Medit-Ageing cohorts.
figure 2

Raw data distributions of depression and anxiety by profile, with the white circles representing the estimated marginal means following adjustment for covariates (age, sex and years of education, as well as study group (for Medit-Ageing data)). The 95% confidence intervals are displayed as vertical black lines. Higher scores across all measures represent greater levels of depression and anxiety. Two-tailed linear regressions were performed to test for the effect of psychological profile group membership on depression (BBHI, N = 749, F2,746 = 63.6, P < 0.001; Medit-Ageing, N = 282, F2,275 = 24.6, P < 0.001) and anxiety (BBHI, N = 749, F2,746 = 131.8, P < 0.001; Medit-Ageing, N = 281, F2,274 = 71.9, P < 0.001). A significant main effect of psychological profile is represented by a bold horizontal line at the top of the graph, with pairwise differences displayed by thinner horizontal lines below. Precise P values for pairwise comparisons are reported in Supplementary Table 2. There were no corrections for multiple comparisons. DASS-21, Depression, Anxiety and Stress Scale–21 items; GDS, Geriatric Depression Scale; STAI-B, State and Trait Anxiety Inventory–Scale B; ***P < 0.001; **P < 0.01; *P < 0.05.

Fig. 3: Association between psychological profiles and cognition in the BBHI and Medit-Ageing cohorts.
figure 3

Raw data distributions of objective cognition and subjective cognitive complaints by profile, with the white circles representing the estimated marginal means following adjustment for covariates (age, sex and years of education, and study group (for Medit-Ageing data)). The 95% confidence intervals are displayed as vertical black lines. Higher scores across all measures represent better objective cognition and greater levels of subjective cognitive complaints (the scores for the latter are inversed for visualization purposes). Two-tailed linear regressions were performed to test for the effect of psychological profile group membership on objective cognition (BBHI, N = 729, F2,723 = 7.2, P < 0.00; Medit-Ageing, N = 280, F2,273 = 8.0, P < 0.001) and subjective cognitive complaints (BBHI, N = 738, F2,732 = 60.1, P < 0.001; Medit-Ageing, N = 276, F2,269 = 21.0, P < 0.001). A significant main effect of psychological profile is represented by a bold horizontal line at the top of the graph, with pairwise differences displayed by thinner horizontal lines below. Precise P values for pairwise comparisons are reported in Supplementary Table 2. There were no corrections for multiple comparisons. For visualization purposes, scores of subjective cognitive complaints for the BBHI sample were inverted from those utilized in the statistical analyses, so that higher scores reflect more subjective cognitive complaints. McNair CDS, McNair Cognitive Difficulties Scale; Neuro-QoL, Quality of Life in Neuroradiological Disorders; PACC5Abridged, Preclinical Alzheimer’s Cognitive Composite 5 Abridged; ***P < 0.001; **P < 0.01; *P < 0.05.

Fig. 4: Association between psychological profiles and health and lifestyle factors in the BBHI and Medit-Ageing cohorts.
figure 4

Raw data distributions of subjective sleep problems, loneliness, social network engagement and LIBRA scores by profile, with the white circles representing the estimated marginal means following adjustment for covariates (age, sex and years of education, and study group (for Medit-Ageing data)). The 95% confidence intervals are displayed as vertical black lines. Higher scores across all measures represent greater levels of subjective sleep problems and higher levels of loneliness; and higher social network and LIBRA scores indicate a larger social network and a greater dementia risk, respectively. Two-tailed linear regressions were performed to test for the effect of psychological profile group membership on subjective sleep problems (BBHI, N = 735, F2,697 = 42.1, P < 0.001; Medit-Ageing, N = 277, F2,270 = 15.4, P < 0.001), loneliness (BBHI, N = 703, F2,697 = 76.0, P < 0.001; Medit-Ageing, N = 277, F2,270 = 20.6, P < 0.001), social network engagement (BBHI, N = 738, F2,697 = 41.1, P < 0.001) and LIBRA scores (BBHI, N = 704, F2,698 = 15.7, P < 0.001). A significant main effect of psychological profile is represented by a bold horizontal line at the top of the graph, with pairwise differences displayed as thinner horizontal lines below. Precise P values for pairwise comparisons are reported in Supplementary Table 2. There were no corrections for multiple comparisons. Jenkins, Jenkins Sleep Evaluation Questionnaire; LIBRA, Lifestyle for BRAin health; LSNS, Lubben Social Network Scale; PSQI, Pittsburgh Sleep Quality Index; UCLA, University of California Loneliness Scale; ***P < 0.001; **P < 0.01; *P < 0.05.

Mental health

In both BBHI and Medit-Ageing, psychological profile membership was associated with anxiety (BBHI, F2,746 = 63.6, P < 0.001; Medit-Ageing, F2,274 = 71.9, P < 0.001) and depressive (BBHI, F2,746 = 131.8, P < 0.001; Medit-Ageing, F2,275 = 24.6, P < 0.001) symptoms. Planned pairwise comparisons revealed that individuals in profile 2 exhibited significantly elevated levels of depressive and anxiety symptoms in comparison to those in profiles 1 and 3, in both cohorts. Furthermore, participants in Profile 1 had higher anxiety and depressive symptoms compared to those in profile 3 in both cohorts.

Cognition

Psychological profile membership was associated with a global cognitive composite sensitive to detecting and tracking preclinical Alzheimer’s disease-related decline (that is, the four-item ‘abridged’ Preclinical Alzheimer’s Cognitive Composite 5 (PACC5abridged)) in both BBHI (F2,723 = 8.0, P < 0.001) and Medit-Ageing (F2,273 = 12.7, P < 0.001). Although no statistically significant differences emerged in planned pairwise comparisons in BBHI, the pattern of findings closely resembled that of Medit-Ageing. In Medit-Ageing, individuals in profile 1 had significantly worse global cognitive function (that is, lower PACC5abridged scores) compared to those in profiles 2 and 3.

In addition to associations with objective cognition, psychological profile membership was related to subjective cognition in both BBHI (F2,732 = 60.1, P < 0.001) and Medit-Ageing (F2,269 = 21.0, P < 0.001). Individuals in profile 2 reported greater perceived subjective memory concerns compared to profile 3 individuals in both cohorts. Additionally, in BBHI, individuals in profile 2 reported more memory concerns than individuals in profile 1, while profile 1 individuals reported greater concerns than those in profile 3.

Following additional adjustment for anxiety and depressive symptoms in sensitivity analyses, objective and subjective cognition results remained largely unchanged (Supplementary Table 3).

Health and lifestyle

In BBHI, psychological profile membership was associated with health and lifestyle factors related to dementia risk, as captured by the late-life ‘Lifestyle for Brain Health’ composite (LIBRA; F2,698 = 15.7, P < 0.001). Planned pairwise comparisons revealed that individuals in profiles 1 and 2 had poorer health and lifestyles (that is, higher LIBRA scores) compared to those in profile 3.

Exploratory analyses revealed associations between psychological profile membership and the LIBRA constituent measures of cognitive activity, hypercholesterolemia, adherence to the Mediterranean diet and smoking. No pairwise differences were observed in relation to hypercholesterolemia and smoking; however, individuals in profiles 1 and 2 reported lower levels of cognitive activity and less adherence to the Mediterranean diet than individuals in profile 3 (Supplementary Table 4).

It was not possible to compute the LIBRA in Medit-Ageing. Instead, exploratory analyses were performed to examine the association between psychological profile membership and LIBRA components. Specifically, we focused on the components that were associated with psychological profile membership in BBHI and were also available in Medit-Ageing (that is, cognitive activity, adherence to the Mediterranean diet and smoking). Partially aligning with the findings from BBHI, psychological profile membership was associated with cognitive activity and adherence to the Mediterranean diet, but not smoking. No pairwise differences were observed in relation to cognitive activity; however, individuals in profile 1 reported greater adherence to the Mediterranean diet than individuals in profiles 2 and 3.

The relationship between psychological profile membership and subjective sleep quality, loneliness and social network engagement—health and lifestyle factors associated with dementia risk but not captured by the LIBRA—were also examined. In both cohorts, analyses revealed associations between psychological profile membership and subjective sleep quality (BBHI, F2,697 = 42.1, P < 0.001; Medit-Ageing, F2,270 = 15.4, P < 0.001) and loneliness (BBHI, F2,697 = 76.0, P < 0.001; Medit-Ageing, F2,270 = 20.6, P < 0.001). Planned pairwise comparisons revealed that individuals in profile 2 had worse perceived sleep quality and higher levels of loneliness compared to those in profiles 1 and 3, in both cohorts. Furthermore, in BBHI, individuals in profile 1 also reported worse sleep quality and greater loneliness than those in profile 3. In BBHI, where social network engagement was also assessed, an association was observed with psychological profile membership (F2,697 = 41.1, P < 0.001), with planned pairwise comparisons indicating that participants in profiles 1 and 2 had smaller social network engagement in comparison to those in profile 3.

Following additional adjustment for anxiety and depressive symptoms in sensitivity analyses, all results remained largely unchanged (Supplementary Table 3).

Cortical thickness

In the BBHI sample, where MRI baseline data were available for 716 participants, psychological profile membership was not associated with differences in cortical thickness (as revealed by a vertex-wise general linear model conducted on FreeSurfer).

BBHI longitudinal analyses subsample

Both cognition and MRI data were obtained for 533 BBHI participants at a follow-up assessment (profile 1, n = 139, 26.1%; profile 2, n = 101, 18.9%; profile 3, n = 293, 55.0%). Over an average follow-up period of 2.3 years (range, 0.7 to 3.4 years), there were no differences in attrition rates across psychological profiles (χ2 = 1.2, P = 0.540). Compared to the total BBHI sample, individuals with longitudinal data were on average older (T381 = 2.2, P = 0.029, t-test), but did not differ in relation to the proportion of women (χ2 = 0.1, P = 0.765) or education level (T342 = 0.4, P = 0.726).

As a sensitivity check, all baseline cross-sectional analyses were re-conducted in the BBHI subsample with longitudinal data (Supplementary Tables 5 and 6). The results mirrored those for the entire sample, showing baseline associations between psychological profile membership and mental health, cognition and lifestyle factors. Planned pairwise comparisons within the BBHI subsample remained largely consistent with the full sample, revealing that profile 1 membership was associated with the lowest levels of objective cognition, poorest health and lifestyle as measured by the LIBRA, and the smallest social network. Also, individuals in profile 2 exhibited the highest levels of depression, anxiety, loneliness, subjective cognitive concerns and the worst perceived sleep quality (Supplementary Tables 5 and 6). In addition, we did not observe differences in cortical thickness.

Longitudinal findings

Cognition

During the 2.3-year follow-up period, no change in global cognitive function (that is, PACC5abridged scores) was observed from baseline to follow-up (β = −0.0, P = 0.787) when analyzing longitudinal data in an adjusted linear mixed-effects model. In a separate mixed-effects model, a group-by-time interaction revealed no differences in PACC5abridged score changes according to psychological profile (F2,519 = 0.7, P = 0.519). However, although psychological profile membership was not associated with changes in PACC5abridged scores, analyses revealed stability in the association between psychological profile membership and global cognitive function. Specifically, psychological profile membership was associated with PACC5abridged scores at follow-up (F2,512 = 6.5, P = 0.002) when fitting a linear regression model. Planned pairwise comparisons revealed that individuals in profile 1 demonstrated worse global cognitive function compared to those in profile 3 (β = −0.2, P = 0.011). Following additional adjustment for anxiety and depressive symptoms in sensitivity analyses, the results remained largely unchanged.

Cortical thickness

Vertex-wise general linear models were fitted using FreeSurfer to investigate longitudinal changes in cortical thickness. During the follow-up period, cortical thinning was observed, spanning the lateral and medial parts of the frontal cortex (that is, dorsolateral prefrontal cortex, anterior cingulate and orbital cortices), the inferior parietal lobule and the precuneus/posterior cingulate region, as well as the lateral, middle and anterior parts of the temporal lobe. Other regions, such as the primary visual and motor cortices, were less affected (Supplementary Fig. 1).

Psychological profile membership was associated with change in cortical thickness from baseline to an average of 2.3 years of follow-up. Specifically, planned pairwise comparisons revealed that, compared to profile 3, individuals in profile 1 exhibited the greatest cortical thinning in the inferior and middle temporal regions and the fusiform gyri bilaterally, as well as in the lateral occipital and pericalcarine area (spanning the lingual gyrus and cuneus) in the left hemisphere. These differences were maintained after adjusting for anxiety and depressive symptoms, and in this adjusted model, differences were also observed between profiles 1 and 2 in the thinning of the inferior temporal lobe region where the former group exhibited accelerated brain atrophy (Fig. 5). In sensitivity analyses that included additional adjustments for cognitive change, differences between profiles 1 and 3 were still observed in the inferior temporal and lateral occipital regions in the primary model. In the adjusted analyses, significant differences were only observed in the inferior temporal region.

Fig. 5: Longitudinal profile comparisons in cortical thickness atrophy in BBHI.
figure 5

a,b, Vertex-wise symmetrized percent change maps of significant clusters surviving family-wise error multiple comparison correction in 533 participants from BBHI, for the primary model (adjusted for age, sex and education; a) and the adjusted model (further adjusted for depression and anxiety symptoms; b). Blue to light blue reflects higher cortical thickness loss for profile 1 in comparison to profile 3 (in a and upper row of b) and profile 2 (lower row of b). In a, final cluster-wise P values are <0.001 for the cluster around the right lateral occipital area, 0.034 for the cluster around the right fusiform area and <0.001 around the left inferior temporal area. In b, when comparing profile 1 versus profile 3, final cluster-wise P values are <0.001 for the cluster around the right lateral occipital area and <0.001 for the cluster around the left inferior temporal area. When comparing profile 1 versus profile 2, the final cluster-wise P value is 0.004 for the cluster around the left inferior temporal area. All analyses were performed with vertex-wise one-factor/two-level general linear models, as provided by FreeSurfer.

link

Leave a Reply

Your email address will not be published. Required fields are marked *