Social Relationships, Wealth, and Cardiometabolic Risk: Evidence from a National Longitudinal Study of U.S. Older Adults

Abstract

Objectives:

To investigate multiple dimensions of social relationships related to biomarkers of cardiometabolic health and how their associations vary by wealth in older adults.

Methods:

Growth curve models were used to investigate the longitudinal associations between measures of both positive and negative social relationships and cardiometabolic risk (CMR) over a 10-year period from 2006 to 2016 and the moderation of this association by wealth in the Health and Retirement Study (HRS).

Results:

Older adults with better social relationships had lower CMR on average. The protective effects of positive social relationships, however, waned at older ages, particularly for low-wealth individuals.

Discussion:

Our results suggest that good social relationships promote healthy aging by buffering against harmful cardiometabolic consequences of psychosocial stress, particularly among relatively wealthy individuals. Efforts to improve old age health would be more effective when focusing simultaneously on fostering social connections and boosting financial resources.

INTRODUCTION

Social relationships are fundamental to health and longevity in humans and other animals (Snyder-Mackler et al., 2020). Previous literature has found that social relationships are associated with a wide range of health outcomes including cardiovascular disease (Brummett et al., 2001), cognitive decline (Evans et al., 2019; Seeman et al., 2001), depression (Beutel et al., 2017), disability (Crowe et al., 2021), biomarkers of physiological dysregulation (Ford et al., 2006; Seeman et al., 2002), and mortality (Holt-Lunstad et al., 2010). While social relationship-health associations have been found at all stages of the life course (Hämmig, 2019; Seeman et al., 2014; Yang et al., 2016), they may be particularly important for older adults. Older adults are more likely to experience social stress associated with late-life transitions, such as retirement and loss of job contacts and income, the loss of a spouse, and onsets of chronic illnesses and impairments that limit their social interactions and lead to additional health declines (Choi et al., 2018; Coyle et al., 2017). Yet we know relatively little about how social relationships operate to impact cardiometabolic risk (CMR), which is of particular importance to health in old age (Ford, 2004; Mottillo et al., 2010), or how the cardiometabolic health benefits of social relationships vary across SES groups. We identified three gaps in the literature related to the 1) measurement of social relationships; 2) differential cardiometabolic health benefits of social relationships across SES levels; and 3) use of longitudinal data.

First, although literature examining the association between social relationships and health features diverse measurement strategies of social relationships across disciplines (Valtorta et al., 2016), few empirical studies have assessed these measures comprehensively. Measures of social relationships consist of both structural and functional dimensions (Holt-Lunstad, 2018). Structural aspects are quantitative measures of social relationships such as social network size and frequency of contact with others, while functional aspects comprise qualitative measures such as actual or perceived aid and loneliness (Berkman et al., 2000; Holt-Lunstad, 2017; Holt-Lunstad et al., 2010; Valtorta et al., 2016). Despite social relationships being conceptualized as multidimensional, previous research often is confined to unidimensional or single indicators (Ford et al., 2006) that fails to completely capture the wide range of social relationship dimensions related to health. In fact, research shows that multidimensional measures of social relationships are more predictive of mortality than single indicators (Holt-Lunstad et al., 2010). Additionally, measuring social relationships as a multidimensional construct may better capture the multiple pathways by which social relationships shape individual health (Holt-Lunstad, 2018). One hypothesized pathway is through stress-induced physiological responses (Umberson et al., 2010). The stress process activates the sympathetic nervous system and hypothalamic-pituitary-adrenal axis (Brotman et al., 2007). Prolonged activation of these systems can lead to cardiometabolic dysregulation, which is indicated by markers of cardiometabolic risk (CMR; Björntorp & Rosmond, 1999). This biophysiological process shows how social connections can “get under the skin” to affect health. Close ties with friends and relatives, for instance, are associated with lower CMR across older adulthood (Shiovitz-Ezra & Parag, 2019; Yang et al., 2013).

Additional research has found that social network structure, social integration, strain, and support are linked with cardiometabolic biomarkers (Ford et al., 2000; Goldman, 2016; Seeman et al., 2014). While these studies show that social relationships are associated with individual cardiometabolic biomarkers, more research is needed examining composite measures of CMR to better capture the multidimensional and cumulative biophysiological impact of social relationships. This association is important as high CMR has been shown to be an important risk indicator of subsequent poor health, including increased risk of cardiovascular disease, accelerated cognitive decline, diabetes and higher mortality risk in the elderly (Alberti et al., 2009; Eckel et al., 2005; Mottillo et al., 2010).

Second, previous research has not adequately addressed whether the cardiometabolic health benefits of social relationships are equally experienced by individuals across population subgroups defined by social class and status (Sonderlund et al., 2019). Theories, such as the fundamental cause theory (Link & Phelan, 1995), suggests that social connections may function differently according to SES. According to this theory, high SES groups have better access to knowledge, resources, and beneficial social connections that generally leads to healthier lifestyles, and thus lower CMR, compared to low SES groups. Furthermore, healthy lifestyles are strongly supported by social norms and other forms of social support among high status groups (Phelan et al., 2010).

Similarly, social capital theory suggests ways in which types of social capital—resources accessed through social connections-- may confer differential benefits depending on SES. Bonding social capital refers to resources that are accessed within networks or groups in which members share similar background characteristics, whereas bridging social capital describes resources that are accessed across networks that “bridge” social characteristics (Villalonga-Olives et al., 2016). Bonding social capital, as measured by indicators of social cohesion and network support, may be particularly beneficial for higher SES groups and less so for lower SES groups. High bonding can have detrimental mental and behavioral effects through excessive demands, for example, which can be particularly detrimental to those with lower SES by adding to an already high stress burden (Rodgers et al., 2019; Szreter & Woolcock, 2004). There is evidence that the association between social integration and metabolic disorders was more salient for respondents with less than a high school education (Yang et al., 2013).

Wealth is a holistic indicator of financial well-being and becomes particularly relevant to health in old ages as individuals exit the labor market and turn increasingly to their accumulated assets for support. Past studies have found that in late adulthood, wealth is a better measure of SES than household income because it reflects available financial resources and takes into account the cumulative effects of a lifetime of deprivation or privilege (Duncan et al., 2002; Robert & House, 1996). Further research found that wealth contributes more than education or income to differences in biomarkers of metabolic disorders in late life (Yang et al., 2020). It is unknown, however, what role wealth may play in conditioning the association between social relationships and cardiometabolic risks.

Third, most research to date on the links between singular measures of social relationships, physical health, and SES is based on cross-sectional data. The question remains regarding how these links unfold over time as individuals age into late life. Longitudinal investigations are needed to test how the social relationship-cardiovascular risk association may change (strengthen or weaken) with age and whether it is further modified by SES. Additionally, cross-sectional analyses have the potential problem of reverse causality. For example, having higher CMR can lead to more chronic conditions, resulting in withdraw from social interactions (Cantarero-Prieto et al., 2018). Longitudinal designs are necessary to determine the temporal ordering of the association under question.

The present study fills these gaps by examining how multidimensional measures of social relationships are associated with age trajectories of CMR among a nationally representative longitudinal sample of older adults. Additionally, we focused on wealth as the measure of SES in this study of older adults and examined whether the health-protective dimensions of social relationships varied between high and low-wealth groups. Given the protective effects of social relationships and wealth for a broad range of health outcomes from previous research, we hypothesize that higher baseline levels of positive social relationships would be associated with lower baseline and slower rates of increase in CMR. We also expect that higher wealth at baseline would be associated with lower baseline and slower growth of CMR. Furthermore, we tested an exploratory hypothesis that wealth status at baseline would moderate the relationship between social relationships and CMR, but we did not have a priori hypotheses about the direction of the association by wealth status.

DATA AND METHODS

Study Sample

This study used data from the Health and Retirement Study (HRS), a nationally representative sample of Americans aged 50 and older that has been ongoing since 1992 with follow-ups every two years (Sonnega et al., 2014). Details of the HRS longitudinal panel design and sampling are available in further detail on the HRS website (http://hrsonline.isr.umich.edu). In 2006, HRS began Enhanced Face-to-Face Interviews which included biomarker collection and a Leave-Behind Questionnaire on psychosocial topics. A random half of households were selected for the enhanced face-to-face interview in 2006, with the other half receiving the enhanced interview in 2008, after which participants were given the enhanced face-to-face interview every four years. We pooled data from the 2006 and 2008 interviews to serve as a baseline. Interviews from 2010 and 2012 as well as 2014 and 2016 were also pooled to serve as two follow-ups. The analytic sample includes a total of 11,943 respondents with valid measures of CMR for at least one wave and social relationships, wealth, and covariate measures at baseline.

Measures

Dependent Variables

Our main outcome of interest is cardiometabolic risk (CMR). We measured cardiometabolic risk as the sum of dichotomous (0/1) values for seven biological markers (range 0–7, see Yang et al., 2020): blood pressure (systolic blood pressure >=140 mmHg, diastolic blood pressure >= 90 mmHg, or taking anti-hypertensive medications), glycated hemoglobin (HbA1c, >6.5%), high-density lipoprotein (HDL, <40 mg/dL for men or <50 mg/dL for women), total cholesterol (>240 mg/dL), waist circumference (>102 cm for men or >88 cm for women), cystatin-c (top quartile), and CRP (>3 mg/dL). Dichotomous values were used in accordance with clinical literature, to indicate high levels of risk for each measure and adjust for respondents taking medications (Alberti et al., 2009; Eckel et al., 2005). CMR was calculated for respondents with non-missing data on waist circumference, blood pressure, and two or more blood biomarkers. The scores were prorated to account for missing biomarker data and top-coded at five to reduce skewness.

Exposure Variables

We measured social relationships as a multidimensional construct comprising of five variables: social support, perceived neighborhood social cohesion, social integration, social strain, and loneliness. All measures were collected in the HRS psychosocial leave behind survey at baseline. Coding of measures for inclusion in the scale are described in Table 1. Social support comprises three Likert-scale questions asked on four relationship types: spouse/partner, children, family, and friends. Responses were coded with higher values representing higher levels of support, then averaged across questions and relationship types. Perceived neighborhood social cohesion was measured as an average of four variables representing neighborhood trust, friendliness, helpfulness, and belonging. Social integration was measured as a count of five items: marital status, parental contact, child contact, neighborhood contact, and volunteering (see Yang et al., 2016). Social strain was comprised of four Likert-scale questions asking whether their spouse/partner, children, family, or friends made too many demands, criticized them, let them down, or got on their nerves. Responses were reverse coded with higher values indicating less strain and averaged across questions and relationship types. Lastly, loneliness was measured using three-items of the UCLA Loneliness Scale (Russell, 1996). Items were reverse coded with higher values indicating less loneliness. All of the individual social relationship measures were standardized to a mean of zero and a standard deviation of one.

Table 1.

Social Relationships Measurement

Variable	Survey item	Coding
SOCIAL SUPPORT	How much do they really understand the way you feel about things? How much can you rely on them if you have a serious problem? How much can you open up to them if you need to talk about your worries? Questions above asked separately about one’s spouse/partner, children, family, and friends.	4 = A lot, 3 = Some, 2 = A little, 1 = Not at all Average across all relationship types
PERCEIVED NEIGHBORHOOD SOCIAL COHESION	I really feel part of this area/I feel that I don’t belong in this area. Most people in this area can be trusted/Most people in this area can’t be trusted. Most people in this area are friendly/Most people in this area are unfriendly. If you were in trouble, there are lots of people in this area who would help you/If you were in trouble, there is nobody in this area who would help you.	1–7 scale Average of items
SOCIAL INTEGRATION	Marital status In the past 12 month, how often have you had contact either in person or by phone, mail or email with your parents? In the past 12 month, how often have you had contact either in person or by phone, mail or email with your children? How often do you get together with people in or near the facility/any of your neighbors just to chat or for a social visit? Have you spent any time in the past 12 months doing volunteer work for religious, educational, health-related or other charitable organizations?	1 = married or cohabiting; 0 = otherwise 1 = once a week or more; 0 otherwise 1 = once a week or more; 0 otherwise 1 = once a week or more; 0 otherwise 1= ever volunteered in the past year; 0 = otherwise Sum of 5 items
SOCIAL STRAIN	How often do they make too many demands on you? How much do they criticize you? How much do they let you down when you are counting on them? How much do they get on your nerves? Questions above asked separately about one’s spouse/partner, children, family, and friends.	4 = Not at all, 3 = A little, 2 = Some, 1 = A lot Average across all relationship types
LONELINESS	How often do you feel you lack companionship? How often do you feel left out? How often do you feel isolated from others?	3=Hardly ever or never, 2=Some of the time, 1=Often Average
PCA SCORE	Weighted score based on principal component analysis	Mean = 0 Standard deviation = 1

We also created composite scores of social relationships using the individual measures. To construct the score for overall social relationships, we reversed coded social strain indicating negative social relationships (loneliness was already reverse coded as a singular indicator) so that all variables were coded in the same direction, with higher values indicating more positive social relationships. Next, we ran a principal component analysis (PCA) with all five factors. The PCA provided evidence for one component (1st factor with an Eigenvalue of 1.92). All factors had moderate factor loadings except for social integration, whose factor loading was comparatively lower (Table 2). Because of the substantial evidence of the health benefits of social integration as a key quantity indicator of social relationships and the PCA results for one component, we kept social integration in the PCA. We then created a weighted score based on the PCA, standardized in the full sample to a mean of zero and standard deviation of one.

Table 2.

Eigenvectors of the Principal Components of Social Relationships

Items	Overall Social Relationships	Positive Social Relationships	Negative Social Relationships
Social Support	0.50	0.60
Perceived Neighborhood Social Cohesion	0.42	0.61
Social Integration	0.26	0.51
Social Strain	0.46		0.70
Loneliness	0.54		0.70

We also created PCA scores for positive and negative aspects of social relationships. The positive social relationship PCA consisted of three measures, including social support, perceived neighborhood social cohesion, and the social integration scale. The negative social relationship PCA included social strain and loneliness scales. Both the positive and negative PCA scores had an Eigenvalue above one, suggesting evidence for one component. Results are further reported in Table 2.

Wealth was calculated using a measure of net value of total wealth constructed by RAND which sums all assets, including a second home, minus all debt at baseline. Wealth values were transformed to reduce skewness using an inverse hyperbolic sine (IHS) transformation (Burbidge et al., 1988). Additionally, wealth is transformed to have a mean of zero and standard deviation of one for ease of interpretation.

Covariates

Our analyses adjusted for socio-demographic characteristics and health behaviors at the baseline. Socio-demographic variables include age (centered at 50, top-coded at 95, and divided by ten), gender (reference = male), and race (Non-Hispanic White (reference), Non-Hispanic Black, Hispanics, and Other). Health-behavior variables come from the RAND longitudinal file and included smoking status (current-smoker/non-smoker), binge-drinking status (consuming four or more alcohol drinks on at least one occasion in the past three months/non-binge-drinker), and physical activity (engaged in moderate/vigorous physical activity more than once per week/sedentary). Moderate physical activity included sports and activities such as gardening, cleaning the car, walking at a moderate pace, dance, and floor or stretching exercises; while vigorous physical activity included running or jogging, swimming, cycling, aerobics or gym workouts, tennis, or digging with a spade or shovel.

Analytic Approach

We used hierarchical linear mixed effects or growth curve models to examine the association of social relationships and wealth with age trajectories of cardiometabolic risk (CMR). Growth curve models estimate person-specific intercepts (CMR at age 50) and slopes to describe interindividual differences in intraindividual change (Raudenbush & Bryk, 2002). Fixed effects were estimated for all exposure variables and covariates, while random effects were estimated for the intercept and linear slope of age. This approach allowed for data unbalanced in time and the inclusion of all respondents with one to three repeated measures of CMR.

To account for attrition related selection bias, we used inverse probability of attrition weighting (IPAW; Weuve et al., 2012). Weights were estimated using logistic regression models which predicted the probability of attrition due to either death or non-response as a function of a set of variables thought to influence attrition: age, gender, race, education, number of comorbidities, number of activities of daily living (ADLs), smoking, physical activity, and binge drinking. Weights were stabilized and applied to our mixed models.

We estimated models in a stepwise fashion. Model 1 estimated the gross age trajectories of CMR, adjusting for covariates, including gender, race, and health behaviors. Model 2 added baseline social relationship variables and wealth as well as their interactions with age or the slope of CMR. And Model 3 included an interaction term of wealth and social relationships. We estimated these models for each of the three social relationship measures or PCA composite scores (overall, positive, negative social relationships). For the individual social relationship indicators, we reported results from Model 3 in the Appendix Table B.

RESULTS

Sample Characteristics

Table 3 presents weighted baseline characteristics for the analytic sample. Overall, the sample was mostly white (78 percent), slightly more female (59 percent), and ranged in age from 50 to 95 with a mean of 68. The average cardiometabolic risk (CMR) index was 2.5 (SD=1.4), indicating that respondents are in the high-risk group for two or three of the seven biomarkers in the index. Sample members reported moderate levels of all five indicators of social relationships. Wealth ranged widely from negative 2.2 million to nearly 37 million dollars in net worth, indicating both assets and debts. Examining health behaviors, 13 percent of the sample were current smokers, 60 percent were physically active, and 10 percent engaged in binge drinking. Compared to those who remained in the sample (N = 6,816), those lost to follow-up due to death or non-response (N = 5,127) had higher CMR, less positive social relationships, lower wealth, were older and more likely to smoke at the baseline (see Appendix A for more detail).

Table 3.

Baseline Sample Characteristics, Health and Retirement Study 2006 – 2016, (N = 11,943)

	Mean/% (SD)	Min	Max
Biomarkers
Cardiometabolic Risk	2.53 (1.43)	0	5
Blood Pressure	0.67	0	1
HbA1c	0.13	0	1
HDL	0.31	0	1
Total Cholesterol	0.18	0	1
Waist Circumference	0.65	0	1
Cystatin C	0.27	0	1
CRP	0.31	0	1
Social Relationships
Overall Social Relationships PCA Score	0.02 (0.99)	−5.0	2.1
Positive Social Relationships PCA Score	0.03 (0.99)	−4.4	2.4
Negative Social Relationships PCA Score	0.01 (0.99)	−4.7	1.4
Social Support	3.14 (0.52)	1	4
Perceived Neighborhood Social Cohesion	5.51 (1.37)	1	7
Social Integration	2.58 (1.09)	0	5
Social Strain	3.35 (0.47)	1	4
Loneliness	2.52 (0.54)	1	3
SES
Wealth	526,775 (1,139,559)	−2,199,392	37,050,000
IHS Transformed Wealth	11.70 (5.00)	−15.3	18.1
Sociodemographic status
Age	68.35 (9.64)	50	95
Female (%)	0.59	0	1
Race (%)
Non-Hispanic White	0.78	0	1
Non-Hispanic Black	0.12	0	1
Hispanic	0.08	0	1
Other	0.02	0	1
Health Behavior
Smoking	0.13	0	1
Physical Activity	0.60	0	1
Binge Drinking	0.10	0	1

Modeling of Growth Trajectories of CMR

Table 4 presents growth curve model estimates of the CMR intercept and slope of increase with age in association with overall social relationships and wealth. Model 1 shows that CMR increased significantly as individuals aged. Model 2 shows that consistent with our hypothesis, more positive social relationships and higher wealth were associated with lower mean CMR (Model 2). In particular, a 1-SD increase in the baseline overall PCA score was associated with a 0.13-unit (p<.001) decrease in mean CMR; and a 1-SD increase in IHS transformed wealth score was associated with a 0.12-unit (p<.001) decrease in mean CMR. Model 3 further shows a significant interaction between overall social relationships and wealth, indicating that the social relationship gradient in mean CMR (greater CMR for poor social relationships) was larger in the wealthier group (p<.001). The inclusion of this interaction improved the model fit (smaller BICs for Model 3 compared to Model 2).

Table 4.

Linear Mixed Models of Cardiometabolic Risk, Overall PCA Social Relationships, and Wealth

	Model 1	Model 2	Model 3
Fixed Effects
Constant	2.22*** (2.15, 2.30)	2.22*** (2.15, 2.29)	2.23*** (2.16, 2.30)
Age/10	0.22*** (0.20, 0.24)	0.23*** (0.21, 0.25)	0.23*** (0.21, 0.25)
Social Relationship PCA		−0.12*** (−0.17, −0.07)	−0.13*** (−0.18, −0.08)
IHS Wealth		−0.09*** (−0.15, −0.04)	−0.12*** (−0.18, −0.07)
Social Relationships PCA × Age		0.02 (−0.01, 0.04)	0.02 (−0.001, 0.04)
IHS Wealth × Age		−0.02 (−0.04, 0.01)	−0.01 (−0.04, 0.01)
Social Relationships PCA × IHS Wealth			−0.04*** (−0.07, −0.02)
Random Effects
Level 1 residual	0.66 (0.01)	0.66 (0.01)	0.66 (0.01)
Level 2 intercept	1.67 (0.08)	1.64 (0.08)	1.63 (0.08)
Level 2 age	0.14 (0.02)	0.13 (0.02)	0.13 (0.02)
Goodness of fit
BIC	91029	90880	90874

Note:

^***p<0.001

^**p<0.01

^*p<0.05, two tailed test

Robust confidence interval in parentheses for fixed effects, robust standard errors in parentheses for random effects

Models control for gender, race, smoking, physical activity, and binge drinking.

We illustrated the results from Model 3 in Figure 1 which shows the model predicted age trajectories of CMR by social relationship and wealth levels. The high (top 10 percentiles) social relationship PCA group displayed lower CMR trajectories for most ages compared to the low (bottom 10 percentiles) group for both the low (bottom 10 percentiles) and high (top 10 percentiles) wealth groups. For example, those who are in the bottom 10^th percentile of the social relationship PCA have a predicted CMR at age 65 of 2.66 for the low wealth group and 2.55 for the high wealth groups. Meanwhile, those in the top 10 percentile of the social relationship PCA have a predicted CMR of 2.48 and 2.22 at age 65 for the low and high wealth groups, respectively. Overall, this demonstrates that social relationships have a protective effect on CMR across low and high wealth groups. However, the gap in CMR between low and high social relationships was significantly greater for the high wealth group (difference of 0.33) compared to the low wealth group (difference of 0.18), indicating that positive social relationships may be less health-enhancing for low wealth individuals. Furthermore, the gap between the high and low social relationship groups in CMR decreased with age, suggesting potentially less protective effect of overall social relationships on CMR at older ages. The interaction effect of social relationship PCA and slope of CMR change is not statistically significant, however.

Figure 1. Association of age with cardiometabolic risk (CMR) for participants with low (left) and high (right) levels of wealth.

Data are plotted separately for participants with overall social relationships (SR) PCA scores at the 10^th percentile (blue dots) and 90^th percentile (red dots) of the sample distribution. Low wealth represents the 10^th percentile of the sample distribution. High wealth represents the 90^th percentile of the sample distribution. The y-axis of the figure shows the predicted count of cardiometabolic risk markers. The x-axis of the figure shows chronological age. Predicted values are derived from Model 3 reported in Table 4. The figure shows that social relationships have a protective effect on CMR for both low and high wealth groups, but the protective effect is lessened for low wealth groups.

The results for parallel analyses of the positive relationship and negative relationship PCA scores as the outcomes are shown in Tables 5 and 6, respectively. Negative social relationships are reverse coded that a lower score indicates poorer social relationships. Overall, the results regarding both the main associations of social relationships and wealth and their interactions with CMR are similar to those shown in Table 4 for overall social relationships. The negative relationship PCA score, however, shows a smaller association (Table 6 Model 3: coef. = −.08, p<.01) with the intercept or mean CMR than overall and positive relationship PCA scores. This suggests that positive social relationships may provide a stronger stress buffer than the absence of negative social relationships. In addition, while the slope of increase in CMR with age was not significantly associated with overall or negative social relationship PCA scores, it was with the positive social relationship PCA score (Table 5 Model 3 coef. = .04, p<.001). The results using the five standardized individual social relationship measures are consistent with those reported above (see Appendix Table B). In comparison, the social relationships PCA scores had stronger associations with CMR than the individual measures alone. These findings suggest that composite social relationship measures may be better at summarizing the multiple pathways by which social relationships impact health than individual measures.

Table 5.

Linear Mixed Models of Cardiometabolic Risk, Positive PCA Social Relationships, and Wealth

	Model 1	Model 2	Model 3
Fixed Effects
Constant	2.22*** (2.15, 2.30)	2.23*** (2.16, 2.31)	2.24*** (2.17, 2.32)
Age/10	0.22*** (0.20, 0.24)	0.22*** (0.20, 0.25)	0.22*** (0.20, 0.25)
Social Relationship PCA		−0.14*** (−0.20, −0.09)	−0.16*** (−0.21, −0.10)
IHS Wealth		−0.09*** (−0.14, −0.04)	−0.11*** (−0.16, −0.05)
Social Relationships PCA × Age		0.03** (0.01, 0.06)	0.04*** (0.02, 0.06)
IHS Wealth × Age		−0.02 (−0.05, 0.002)	−0.02 (−0.05, 0.001)
Social Relationships PCA × IHS Wealth			−0.04*** (−0.07, −0.02)
Random Effects
Level 1 residual	0.66 (0.01)	0.66 (0.01)	0.66 (0.01)
Level 2 intercept	1.67 (0.08)	1.64 (0.08)	1.64 (0.08)
Level 2 age	0.14 (0.02)	0.13 (0.02)	0.14 (0.02)
Goodness of fit
BIC	91029	90888	90883

Note:

^***p<0.001

^**p<0.01

^*p<0.05, two tailed test

Robust confidence interval in parentheses for fixed effects, robust standard errors in parentheses for random effects

Models control for gender, race, smoking, physical activity, and binge drinking.

Table 6.

Linear Mixed Models of Cardiometabolic Risk, Negative PCA Social Relationships, and Wealth

	Model 1	Model 2	Model 3
Fixed Effects
Constant	2.22*** (2.15, 2.30)	2.22*** (2.15, 2.30)	2.23*** (2.16, 2.30)
Age/10	0.22*** (0.20, 0.24)	0.24*** (0.21, 0.26)	0.23*** (0.21, 0.26)
Social Relationship PCA		−0.07** (−0.12, −0.02)	−0.08** (−0.13, −0.03)
IHS Wealth		−0.11*** (−0.16, −0.06)	−0.13*** (−0.19, −0.08)
Social Relationships PCA × Age		−0.003 (−0.03, 0.02)	−0.001 (−0.02, 0.02)
IHS Wealth × Age		−0.02 (−0.04, 0.01)	−0.01 (−0.03, 0.02)
Social Relationships PCA × IHS Wealth			−0.04** (−0.06, −0.01)
Random Effects
Level 1 residual	0.66 (0.01)	0.66 (0.01)	0.66 (0.01)
Level 2 intercept	1.67 (0.08)	1.64 (0.08)	1.64 (0.08)
Level 2 age	0.14 (0.02)	0.13 (0.02)	0.13 (0.02)
Goodness of fit
BIC	91029	90895	90895

Note:

^***p<0.001

^**p<0.01

^*p<0.05, two tailed test

Robust confidence interval in parentheses for fixed effects, robust standard errors in parentheses for random effects

Models control for gender, race, smoking, physical activity, and binge drinking.

We conducted sensitivity analyses including education as a covariate and estimating models without IPAW. Education was not included in our final models because its coefficient was not significant in the presence of wealth and the addition of education as a covariate did not significantly change the results reported above. Additionally, the results do not differ in any meaningful way without and with the use of IPAW (see Appendix C for results of overall social relationships as compared to Table 4 Model 3). It is therefore unlikely that attrition due to lost to follow-up lead to serious bias in the estimates.

DISCUSSION

Research has long documented the associations between social relationships and various health outcomes at different points in life. However, most of these studies examine a singular dimension of social relationships, use cross-sectional data, or do not address differential cardiometabolic health benefits of social relationships across SES levels (Ford et al., 2000; Ford et al., 2006; Goldman, 2016; Holt-Lunstad et al., 2010; Seeman et al., 2014). This study used longitudinal data from a nationally representative sample of older adults to shed new light on how multidimensional measures of social relationships are associated with age trajectories of cardiometabolic risk (CMR) in late life. We highlight three key findings.

First, we found strong connections of CMR with social relationships comprehensively assessed by composite measures that integrated both structural and functional aspects of social relationships. In fact, such associations are stronger than those assessed by a subset of relationship factors or single indicators of social relationships which are more commonly used in previous research. Results from both principal component analysis and growth curve models show support for the use of the overall composite score as it provided a superior model fit than either a single indicator or positive/negative dimensions of social relationships. Results using the positive and negative social relationship PCA scores indicate differences in their relationship to CMR which may suggest different pathways linking specific relationship characteristics to CMR. For example, while negative social relationships serve as a stressor, positive social relationships serve as a stress-buffer (Umberson & Montez, 2010). While it is possible that positive, supportive, and stress-buffering relationships may have differential health impacts from negative, straining, and stress-inducing relationships, the precise reason for their stronger association with CMR than the latter is unclear and needs to be further investigated. These findings clearly suggest that future research would benefit from including multiple measures simultaneously to fully capture the health benefits of social relationships and aid in the identification of multiple mechanisms by which social relationships lead to health change (Holt-Lunstad, 2018).

Second, in addition to support for the hypothesized independent protective effects of social relationships and wealth on CMR, we found evidence for a moderating effect of wealth on social relationship differences in CMR. While having better social relationships is associated with a lower CMR on average, this protection is weaker for less wealthy older adults. This suggests that there is a double disadvantage to having poor social relationships and low wealth that can compound to result in a higher CMR in late life. Having both poor social relationships and low wealth can intensify stress, leading to greater CMR when experiencing both stressors compared to experiencing them individually (Thoits, 2010). Additionally, the lack of stress mediators, such as social support, may exacerbate the cardiometabolic risks of stress that are experienced among older adults with low levels of wealth (Pearlin, 1989, 1999). The higher burden of stress in less wealthy older adults may outweigh any stress-buffering effects of beneficial relationships. Lower social support for less wealthy older adults may also provide lower quality of stress coping resources than those of wealthy older adults. Our study thus suggests that social relationship related CMR differences are not universally observed across wealth levels. Future research should continue to investigate the potential role of other SES related contexts in moderating the association between social relationships and other health indicators in old age.

Third, we found longitudinal associations between social relationships, wealth, and age trajectories of change in CMR that could not be discerned in cross-sectional studies. Similar to what has been found in previous literature (Ford et al., 2000; Goldman, 2016; Seeman et al., 2014; Shiovitz-Ezra & Parag, 2019; Yang et al., 2013), our analyses found that having greater overall and positive social relationships lowered mean CMR and was protective of CMR on average. However, we further found that this protective effect waned with advancing age as shown by positive associations of social relationship measures with the slope of increase in CMR. The declining cardiometabolic benefits of more positive social relationships over time may suggest the greater force of frailty or biological aging over the influences of psychosocial factors as aging progresses (Hoffmann, 2011; Taylor, 2010). Alternatively, those with better social relationships have more friends and family in their network, who are also growing older. The increasing age of friends and family members could lead to increasing comorbidities among network members, resulting in less supportive and even more stressful relationships over time. We further note that this result was not statistically significant across all relationship measures, such as overall and negative social relationships. We therefore caution against overinterpretation or generalization of this finding and call for its corroboration from additional studies. In all, it seems imperative for future research to employ longitudinal data and analysis of similar and other health markers to investigate the temporal specificity of health impacts of social relationships as well as other social determinants.

Our study has several limitations. First, although we used enriched measures of social relationships, these measures are not exhaustive and may leave out other potentially relevant factors, such as relationship quality perceived by individuals. For CMR, we focused on a summary count indicating the overall risk burden that is based on cut-points capturing risk thresholds. Therefore, it does not provide the finer gradations in each continuous measure of biomarker for testing linear effects. The measures we did include, however, represent a wide range of major indicators of the overall theoretical constructs of social relationships and CMR which can serve as a foundation for future research using additional measures. Second, we only examined baseline measures of social relationships and wealth, future investigations should include additional follow-up data on either or both exposure variables to further investigate their longitudinal associations with CMR change. Additionally, while this research focused on late life, gaps remain in investigating longitudinal association of social relationships with CMR change in younger adults (Seeman et al., 2014). Third, we did not examine potential explanatory mechanisms that account for the independent and interactive associations between social relationships and wealth with CMR. Future studies are needed to further examine pathways linking social relationships to CMR and variations in these links by wealth status.

In conclusion, our study suggests that policies and programs aimed at reducing cardiometabolic risks and related morbidities and mortality should be directed toward improving both social relationships and financial conditions in older adults. Because we found that more positive social relationships are less health-enhancing for those with low wealth, future research should investigate in greater detail social network characteristics unique to the low-wealth group that hinder the health benefits of social connection. The blunted protective effects of social relationships may be one pathway through which SES inequalities in health persist. Low wealth may be particularly detrimental to older adults who lack resources to create and maintain social networks and support after retirement and the loss of family and friends. A better understanding of what social contexts may condition the health impacts of social relationships is needed to better target interventions and promote healthy aging of older adults with poor social connections and SES related disadvantages.

Appendix Table A. Baseline Sample Characteristics by Attrition Status, Health and Retirement Study 2006 – 2016, (N = 11,943)

	Mean/% (SD)
	Remained	Lost to Follow-Upb
Biomarkers
Cardiometabolic Risk	2.39 (1.42)	2.72 (1.42)a
Blood Pressure	0.63	0.71a
HbA1c	0.12	0.15a
HDL	0.30	0.33a
Total Cholesterol	0.19	0.17
Waist Circumference	0.66	0.64
Cystatin C	0.19	0.38a
CRP	0.30	0.33a
Social Relationships
Overall Social Relationships PCA Score	0.05 (0.98)	−0.01 (1.01)a
Positive Social Relationships PCA Score	0.09 (0.97)	−0.05 (1.01)a
Negative Social Relationships PCA Score	0.004 (0.98)	0.03 (1.01)
Social Support	3.14 (0.51)	3.15 (0.54)
Perceived Neighborhood Social Cohesion	5.53 (1.32)	5.49 (1.43)a
Social Integration	2.72 (1.09)	2.39 (1.06)a
Social Strain	3.33 (0.47)	3.38 (0.47)a
Loneliness	2.54 (0.53)	2.50 (0.55)a
SES Variables
Wealth	545,895 (1,120,583)	510,361 (1,163,938)a
IHS Transformed Wealth	11.83 (5.04)	11.54 (4.94)a
Sociodemographic Variables
Age	65.82 (8.41)	71.66 (10.13)a
Female (%)	0.61	0.57
Race (%)
Non-Hispanic White	0.78	0.78
Non-Hispanic Black	0.12	0.12
Hispanic	0.08	0.08
Other	0.02	0.02
Health Behavior Variables
Smoking	0.12	0.14a
Physical Activity	0.65	0.54a
Binge Drinking	0.12	0.09a
N	6,816	5,127

^ap<.05, two-tailed test

^bincludes the deceased (N = 2,628) and non-responses (N = 2,499) who did not differ significantly in baseline characteristics.

Appendix Table B. Linear Mixed Models of Cardiometabolic Risk, Individual Social Relationship Measures, and Wealth

	Social Support	Perceived Neighborhood Social Cohesion	Social Integration	Social Strain	Loneliness
Fixed Effects
Constant	2.22*** (2.15, 2.30)	2.24*** (2.16, 2.31)	2.28*** (2.21, 2.36)	2.23*** (2.16, 2.30)	2.24*** (2.16, 2.31)
Age/10	0.23*** (0.21, 0.25)	0.23*** (0.21, 0.25)	0.21*** (0.19, 0.23)	0.23*** (0.21, 0.26)	0.23*** (0.21, 0.25)
Social Relationship PCA	−0.12*** (−0.17, −0.06)	−0.06* (−0.12, −0.01)	−0.13*** (−0.18, −0.08)	−0.06* (−0.11, −0.01)	−0.07** (−0.12, −0.02)
IHS Wealth	−0.12*** (−0.17, −0.07)	−0.12*** (−0.18, −0.07)	−0.10*** (−0.15, −0.05)	−0.14*** (−0.19, −0.09)	−0.11*** (−0.17, −0.06)
Social Relationships PCA × Age	0.04** (0.01, 0.06)	0.01 (−0.02, 0.03)	0.03** (0.01, 0.06)	−0.002 (−0.02, 0.02)	0.004 (−0.02, 0.03)
IHS Wealth × Age	−0.02 (−0.04, 0.01)	−0.01 (−0.04, 0.01)	−0.02 (−0.05, 0.0002)	−0.01 (−0.03, 0.02)	−0.01 (−0.04, 0.01)
Social Relationships PCA × IHS Wealth	−0.03* (−0.05, −0.01)	−0.03** (−0.05, −0.01)	−0.03* (−0.06, −0.001)	−0.04*** (−0.06, −0.02)	−0.02 (−0.04, 0.01)
Random Effects
Level 1 residual	0.66 (0.01)	0.66 (0.01)	0.66 (0.01)	0.66 (0.01)	0.66 (0.01)
Level 2 age	0.13 (0.02)	0.14 (0.02)	0.13 (0.02)	0.13 (0.02)	0.13 (0.02)
Level 2 intercept	1.64 (0.08)	1.66 (0.08)	1.65 (0.08)	1.64 (0.08)	1.65 (0.08)
Goodness of fit
BIC	90921	90923	90906	90909	90913

^***p<0.001

^**p<0.01

^*p<0.05, two tailed test

Robust confidence interval in parentheses for fixed effects, robust standard errors in parentheses for random effects

Models control for gender, race, smoking, physical activity, and binge drinking

Social strain and loneliness are reverse coded to make all social relationship values have a positive direction.

Appendix Table C. Linear Mixed Models of Cardiometabolic Risk, Overall PCA Social Relationships, and Wealth: No IPAW

	Model 1	Model 2	Model 3
Fixed Effects
Constant	2.25*** (2.18, 2.32)	2.24*** (2.17, 2.31)	2.25*** (2.18, 2.32)
Age/10	0.21*** (0.19, 0.23)	0.22*** (0.20, 0.24)	0.22*** (0.20, 0.24)
Social Relationship PCA		−0.13*** (−0.18, −0.08)	−0.14*** (−0.19, −0.10)
IHS Wealth		−0.09*** (−0.14, −0.05)	−0.12*** (−0.17, −0.08)
Social Relationships PCA × Age		0.02* (0.00, 0.04)	0.03* (0.01, 0.05)
IHS Wealth × Age		−0.02 (−0.04, 0.001)	−0.01 (−0.04, 0.01)
Social Relationships PCA × IHS Wealth			−0.04*** (−0.07, −0.02)
Random Effects
Level 1 residual	0.81 (0.01)	0.81 (0.01)	0.81 (0.01)
Level 2 intercept	1.20 (0.08)	1.17 (0.08)	1.17 (0.07)
Level 2 age	0.02 (0.02)	0.02 (0.02)	0.02 (0.02)
Goodness of fit
BIC	82106	81942	81934

Note:

^***p<0.001

^**p<0.01

^*p<0.05, two tailed test

Robust confidence interval in parentheses for fixed effects, robust standard errors in parentheses for random effects

Models control for gender, race, smoking, physical activity, and binge drinking.