Brief Report: Time-varying Social Support and Time to Death in The Cardiovascular Health Study

Abstract

Objectives.

There is a consensus that social connectedness is integral for a long, healthy life. However, studies of social support and survival have primarily relied on baseline social support measures, potentially missing the effects of fluctuations of perceived support over time. This is especially important for older adults who experience increased changes in disability. This study examined whether among older adults time-varying perceived social support was associated with time to death (main effect model of support) and whether time-varying disability was a modifier (stress-buffering model of support). Gender and marital status were also examined as modifiers.

Methods.

Older adults in the Cardiovascular Health Study (N=5,201) completed self- report measures of demographics and psychological health and clinical risk factors for mortality at baseline (1989–1990). Perceived social support and disability were measured from baseline through Wave 11 (1998–1999). Cox regression of time to death with time-varying covariates was performed.

Results.

Time-varying as well as baseline-only perceived social support were associated with greater survival in the unadjusted models but not after adjustment. Gender, marital status, and time-varying disability were not significant modifiers.

Conclusions.

In contrast with the previously reported association between baseline individual differences in perceived social support and time to death, older adults’ baseline-only and fluctuating perceptions of perceived support over time were not associated with time to death after adjustment for other clinical physical and psychological risk factors. Research is needed to identify other relationship factors that may be more informative as time-varying predictors of health and longevity in large longitudinal datasets.

There is now a consensus that social connectedness is beneficial for health across the lifespan, with studies showing greater effect sizes for several aspects of social relationships regarding survival than other established risk factors, such as physical activity, smoking, and alcohol consumption (Holt-Lunstad, Smith, & Layton, 2010). A great deal of research on social relationships and health has focused on older adults, partly because as people become older they are more likely to become dependent on others (Janssen, Heymsfield, & Ross, 2002). Most studies of older adults’ social connectedness and mortality focus on feelings of loneliness, whether or not people live alone, marital status, and social network size (Teguo et al., 2016). Less research has examined perceived social support, or the perceived availability of emotional (e.g. belonging), informational (e.g. advice), and instrumental (e.g. tangible assistance) support from social network members (Cohen, Mermelstein, Kamarck, & Hoberman, 1985). Although there is some evidence that perceived social support measured at one time-point is beneficial for older adults’ longevity (Lyyra & Heikkinen, 2006), past findings have been mixed regarding whether perceived social support has a direct or a stress-buffering effect, whether there are gender differences, and whether marital status is a moderator. Furthermore, no study to our knowledge has examined the association between perceived social support and survival in a large sample of older adults in which support and disability are measured at multiple time points and are accounted for as time-varying covariates. This is important because examining time-varying support allows researchers to better understand whether temporal changes in perceived support provide more information than a static measure of baseline support. The Cardiovascular Health Study (CHS) provides a unique opportunity to fill these research gaps. In a 1999 study using CHS data, Martire and colleagues showed that while social network size did not significantly change over time, multiple aspects of perceived support did (Martire, Schulz, Mittelmark, & Newsom, 1999). Another advantage of analyzing this cohort is that the CHS includes updated values of an extensive battery of clinical health indicators that can be used as time-varying covariates.

In this report we tested several models of mortality that regressed on perceived support and health. First, we aimed to replicate the “main effect model of social support” finding that baseline perceived social support is positively associated with survival (Cohen & Wills, 1985; Lyyra & Heikkinen, 2006). Second, we examined a main effect model of mortality by regressing on time-varying perceived social support. The main effect model posits that social support protects health through direct means, such as cognitive, emotional, behavioral and biological influences not explicitly intended as help or support. For example, being around others can lead to better access to health information (Berkman & Syme, 1979) and constructive health behaviors while increasing a person’s sense of belonging (Thoits, 1983). Third, we examined the “stress-buffering model of social support” of mortality by evaluating the association of the interaction between time-varying perceived support and disability. The stress-buffering model states that social support is protective for health by its mitigation of the deleterious effects of stressors in a person’s life (Uchino, 2004). Because disability is a common stressor for older adults, we were interested in whether perceived support modifies the effects of disability on mortality. Previous research has conceptualized disability as a stressor negatively associated with survival (Newman et al., 2006; St. John, Tyas, Menec, & Tate, 2014). After replicating the finding that baseline-only perceived social support is positively associated with survival, we examined whether there was an association between time-varying social support and time to death (Hypothesis 1) and whether there was an association between the interaction between time-varying support and disability and time to death (Hypothesis 2).

In addition, we examined gender and marital status as moderators of time-varying perceived support for the following reasons. First, it remains unclear how the association between perceived support and mortality may differ for men and women. Whereas Jylha & Aro (1989) and Seeman and colleagues (1993) found no significant gender differences in associations between support (e.g., marital status, number of social ties, social participation) and mortality, Lyyra and Heikkinen (2006) found significant survival effects of perceived support only among women. Theorists suggest that women are more sensitive than men to perceived support because they are socialized to be interdependent with others (Cross & Madson, 1997). Here, we hypothesized that women would benefit more from time-varying perceived support than men in terms of time to death (Hypothesis 3). In terms of marital status, researchers have suggested that social support may protect individuals who are unmarried from negative health outcomes (Holt-Lunstad, Birmingham, & Jones, 2008). For this reason we hypothesized that time-varying perceived social support would be more predictive of time to death for unmarried individuals than for married individuals (Hypothesis 4).

Methods

Participants

Data for this study were drawn from the Cardiovascular Health Study (CHS), a large, multi-site population-based sample of older adults. Participants in the CHS were identified from Medicare eligibility lists of the Health Care Financing Administration and were recruited from four U.S. communities: Forsyth County, North Carolina; Washington County, Maryland; Sacramento County, California; and Pittsburgh (Allegheny County), Pennsylvania. Participants were selected and stratified by age and sex at each site in order to produce a cohort with a 60: 40 female to male ratio in each of the following age groups: 65–69 years of age, 70–74 years, 75–79 years, and aged 80 and older. Those eligible to participate included all persons who (a) were 65 years of age or older at the time of the baseline interview; (b) were non-institutionalized; (c) were expected to remain in the area for the next 3 years; and (d) were able to give informed consent and did not require a proxy respondent at baseline. Persons who were wheelchair bound in the home at baseline or who were receiving hospice treatment, radiation therapy, or chemotherapy for cancer were excluded. A total of 5,201 individuals were recruited and interviewed at baseline, with an additional 687 African Americans enrolled in 1992/1993. Information regarding sampling and recruitment are in the original CHS design articles (Fried et al., 1991; Tell et al., 1993). The present study used the original cohort’s baseline data (1989/1990) and Waves 3–11 (1998/99) when perceived social support was measured. We did not include the African American cohort data because its social support data was only measured at Waves 5, 6, and 11 and because their distinct baseline year limited our ability to test the main hypothesis. The CHS study was approved by the institutional ethical review boards at each study site, and the present analysis was exempted by Yale’s institutional review board.

Measures

Mortality.

Time to death was the number of days between the baseline interview and death or ten years, with persons surviving through the end of study being censored. Deaths were confirmed by review of medical records and death certificates, as well as the Health Care Financing Administration Medicare health care utilization database for hospitalizations (Ives et al., 1995).

Sociodemographic characteristics at baseline.

Participants’ gender was coded as 0/1 (female/male). Race (White, Black and other) and marital status (married=1, not married=0) were categorical. Age was treated as a continuous variable while education was coded as the number of years of education. Income was a categorical ordinal variable. All socio-demographics were self-reported.

Perceived support.

Five items from the 6-item version of the Interpersonal Support Evaluation List (ISEL)(Cohen et al., 1985) were used to assess the perceived emotional, informational, and instrumental support. This scale consists of two items assessing each of the three different types of support. Examples of items assessing emotional support and informational support, respectively, are: “When I feel lonely, there are several people I can talk to,” and “When I need suggestions on how to deal with a personal problem, I know someone I can turn to.” The instrumental items were “If I had to go out of town for a few weeks, it would be difficult to find someone who would look after my house or apartment” and “If I were sick, I could easily find someone to help me with my daily chores.” We used only the latter instrumental item as the two items did not form a reliable scale (Martire et al., 1999; Newsom & Schulz, 1996). All items were rated on a 4-point scale from 1 “definitely false” to 4 “definitely true,” and scores for each of the types of support were averaged for each participant. Cronbach’s alpha for the emotional support scale ranged from 0.60 to 0.63, and the alpha for the informational support scale ranged from 0.66 to 0.71. Perceived support was measured at all waves.

Disability.

Disability was assessed with self-reported Instrumental Activity of Daily Living (IADL) and Activities of Daily Living (ADL) needs (Fried, Ettinger, Lind, Newman, & Gardin, 1994). IADLs were defined as difficulty or inability to perform any of the following: heavy housework, light housework, shopping, preparing meals, paying bills or using the phone. ADLs were: difficulty or an inability with walking around the home, getting out of bed, eating, dressing, bathing or using the toilet. IADL items were coded as 0 (no difficulty) and 1 (any difficulty) and summed to create an index. ADLs and IADLs were measured at all waves. Because there were low levels, ADLs and IADL scores were summed and then dichotomized. A score of zero meant no IADL or ADL disability and a score greater than zero indicated any IADL or ADL disability.

Covariates.

In addition to the demographic variables, baseline covariates included stressful life events (Wells, 1985), depressive symptoms (Orme, Reis, & Herz, 1986), the presence of major chronic conditions (e.g., lung disease, nervous disorders, arthritis, kidney disease, diabetes, stroke), subclinical cardiovascular disease, congestive heart failure status, coronary heart disease status, cognitive impairment (normal, moderate, severe).

Analysis

We first calculated the means and frequencies of all variables and evaluated their respective percentages of missing data. To replicate previous findings, we first performed proportional hazards models (unadjusted and adjusted) to examine the associations between baseline-only perceived social support and survival. To evaluate perceived social support and disability as time-varying predictors, we constructed a counting process dataset (Allison, 2010) using last value carried forward for the missing data. For hypothesis 1, we used both unadjusted and adjusted Cox models to test the associations between time-varying social support and mortality. In hypothesis 2, to evaluate whether time-varying perceived social support modified the effect of time-varying disability on survival, the multivariable Cox model included their interaction. For hypothesis 3, we examined the two-way interaction between gender and time-varying social support and the three-way interaction among gender, disability, and social support for their respective associations with time to death. For hypothesis 4 the same procedure was repeated to test whether marital status was a modifier. Functional form and compliance with the assumption of proportional hazards were verified with cumulative sums of martingale residuals (Therneau, Grambsch, & Fleming, 1990). We checked for linearity by plotting the log negative log of survival versus quintiles of social support and checking the associated (adjusted) r-square value. All multivariable models used the same covariates and statistical significance was defined as a two-sided p-value less than 0.05. All analyses were performed in SAS 9.4.

Results

The average age of participants was 72.8 (95% CI 72.6–72.9). Ninety-five percent of the participants were White, 4.7% were Black, and 0.6% were of other races. The full sample consisted of 57% women and 43% men, of whom 69% were married. On average, participants had 13.9 years of education. Table 1 shows descriptive statistics for the explanatory variables at each wave. Perceived social support (p=0005) and disability (p<0.001) each fluctuated over time. A total of 1,806 events occurred. As shown in Table 2, there were significant associations of baseline-only and time-varying perceived social support in the unadjusted models, but neither retained significance after adjustment. The covariates male gender, higher age, lung disease/ bronchitis/ emphysema, CHF status, diabetes and any subclinical CVD were each highly significant in all models and are responsible for eliminating the significance of the association between exposure and the outcome. Time-varying disability (hazard ratio: 1.03, 95% CI 0.97, 1.09), gender (hazard ratio: 1.02, 95% CI 0.96, 1.08), and marital status (hazard ratio: 0.96, 95% CI 0.91, 1.02) did not significantly modify the association between time-varying social support and time to death.

Table 1.

Means and Standard Deviations of Explanatory Variables at Each Wave

	n	Mean	SD	% missing
Perceived Social Support (0-24)1
Baseline	5183	8.27	2.60	0.35
Wave 3	5101	8.44	2.72	4.19
Wave 4	4988	8.27	2.72	7.04
Wave 5	4848	8.38	2.62	10.41
Wave 6	4695	8.45	2.57	17.82
Wave 11	2987	8.6	2.6	42.7
	N	Disability (n)	Disability (%)	% missing
Disability in either ADLs or IADLs (0 or 1)2
Baseline	5192	1403	27.02	0.17
Wave 3	5101	1378	27.01	4.8
Wave 4	4988	1368	27.43	7.6
Wave 5	4848	1257	25.93	9.32
Wave 6	4695	1232	26.24	15.06
Wave 11	3124	1405	27.0	39.9

Note: Raw numbers of participants are reported. The percent missing excludes deceased individuals.

Abbreviations: ADLs= activities of daily living; IADLS= instrumental activities of daily living.

¹Range for social support with higher scores indicating higher values of social support.

²Binary indicator of any disability in either ADLs or IADLs

Table 2.

Unadjusted and Adjusted Cox Regression Models of the Association between Perceived Support and Mortality over Seven Years: Hazard Ratios and 95% Confidence Intervals (N=5,201)

	Fixed		Hypothesis 1		Hypothesis 2
	Unadjusted Baseline only model	Baseline only covariate model	Unadjusted Time-varying covariate model	Main effects time-varying covariates model	Time-varying covariates Interaction model
	HR (95% CI)	HR (95% CI)	HR (95% CI)	HR (95% CI)	HR (95% CI)
Primary Exposure Variables
Perceived social support	1.04 (1.03,1.06)****	1.01 (0.99,1.03)	1.04 (1.01,1.05)****	1.01 (0.99, 1.03)	0.98 (0.93, 1.03)
Disability (ref: no disability)		1.39 (1.24,1.56)****		2.24 (2.01, 2.51)****	1.62 (0.95, 2.79)
Perceived social support* Disability (Interaction)					1.03 (0.97, 1.09)
Covariates
Male gender (ref: Female)		1.81 (1.62, 2.03) ****		1.86 (1.66, 2.08)****	2.28 (1.88, 2.75)****
Not married (ref: Married)		1.22 (1.08, 1.37)***		1.22 (1.09, 1.38)****	1.17 (0.96, 1.42)
Age		1.09 (1.08, 1.1)****		1.08 (1.07, 1.09)****	1.07 (1.06,1.09)****
Education (years)		0.99 (0.98, 1.01)		0.96 (0.92, 1.00)	0.98 (0.96,1.00)*
Life Events Score		0.95 (0.91, 1.00)		0.96 (0.92, 1.00)	0.95 (0.88, 1.02)
Depression Scale		1.01 (1.00, 1.02)		1.01 (0.99,1.02)	1.00 (0.98, 1.02)
Race: Black (ref: White)		1.01 (0.8, 1.27)		0.98 (0.78,1.23)	0.93 (0.63, 1.37)
Other Race		0.84 (0.42, 1.68)		0.89 (0.44,1.79)	1.78 (0.7,4.32)
Income: ≥25,000 (ref: median < 25,000)		0.83 (0.73, 0.94)		0.86 (0.76,0.97)*	0.94 (0.77, 1.15)
Lung disease Emphysema, or Bronchitis		1.45 (1.24, 1.69)****		1.41 (1.21,1.64)****	1.42 1.13, 1.79)**
Nervous or emotional disorder		1.08 (0.90, 1.29)		1.07 (0.89,1.27)	1.34 (1.02, 1.75)*
Kidney disease		1.08 (0.9, 1.43)		1.06 (0.81, 1.39)	0.92 (0.59, 1.46)
Heart disease		1.02 (0.88, 1.17)		1.03 (0.89, 1.19)	1.04 (0.83, 1.31)***
Diabetes		1.66 (1.44,1.92)****		1.53 (1.33, 1.77)****	1.52 (1.21, 1.91)***
Arthritis		0.86 (0.77, 0.95)***		0.81 (0.73, 0.90)*	0.96 (0.81, 1.13)
Stroke		1.56 (1.18, 2.06) **		1.44 (1.09, 1.91)*	1.24 (0.81, 1.90)
Any subclinical CVD		1.61 (1.42 1.82)****		1.56 (1.38, 1.76)****	1.76 (1.41, 2.19)****
CHF status		1.77 (1.47, 2.15)****		1.74 (1.44, 2.10)****	2.22 (1.72, 2.88)****
CHD status		1.13 (0.98, 1.31)		1.08 (0.93, 1.25)	1.14 (0.91, 1.42)
Mild cognitive impairment (ref: normal)		1.16 (1.01, 1.33)*		1.16 (1.01, 1.33)*	1.00 (0.79, 1.25)
Moderate impairment (ref: normal)		1.32 (1.09,1.62)**		1.27 (1.04, 1.55)*	1.23 (0.91, 1.66)
Severe impairment		1.39 (0.89, 2.17)		1.40 (0.89, 2.19)	1.26 (0.66,2.39)

Notes:

^*p-value < 0.05,

^**p-value < 0.01,

^***p-value <0.001,

^****p-value<.0001

Abbreviations: CESD Center for Epidemiologic Studies Depression Scale; CVD= Cardiovascular disease; CHF= congestive heart failure; CHD= Confirmed myocardial infarction/angina,

Discussion

The results of this study did not support the hypothesis that time-varying perceived social support increases older adults’ longevity after adjustment for other known predictors of mortality. There was also no evidence that time-varying disability modified the effects of time-varying perceived support on survival or that either gender or marital status were significant modifiers. To our knowledge, this is the first study of mortality to examine time-varying perceived social support and its interaction with time-varying disability in a large longitudinal dataset of older adults with adjustment for other known risk factors such as chronic conditions (e.g. cancer, diabetes), cognitive impairment, cardiovascular disease, depressive symptoms, and stressful life events.

These results do not support the main effects model or the stress-buffering model of time-varying perceived support. Although perceived support changes over time for older adults (Martire et al., 1999), there is no evidence to suggest that taking these changes into account or examining how they interact with gender or marital status provides added information about survival. We also did not replicate prior findings showing that baseline-only perceived support was a significant predictor of mortality after adjustment for covariates, suggesting that neither baseline nor time-varying perceived support are as influential on mortality as other clinical markers of health. Past studies that showed positive effects of social support have not included objective clinical markers of health or specific health conditions (Lyyra & Heikkinen, 2006).

There were several notable strengths of this study. Firstly, this was a large sample of older adults, a larger sample than most studies examining perceived support and health, who were in very good health and community-dwelling at baseline. Secondly, perceived social support, as well as other psychosocial factors, and extensive clinical and self-report assessments of physical health were measured at multiple, consecutive, time points for a large portion of ten years.

Limitations of this study include that the CHS was conducted about 20 years ago with a primarily white sample. Although the CHS oversampled and added an African American cohort, perceived social support was not assessed at time points that matched the original baseline sample. Also, there was a gap in reports of perceived social support between Wave 6 and 11 where our models have to assume that social support remains constant over that period. Our approach should be replicated in other more recent datasets of heterogenous populations with more frequent longitudinal measures of perceived social support over time.

Taken together, the CHS does not provide evidence that perceived support, as it changes over time and interacts with changes in disability, is associated with increased survival among older adults after adjustment for known clinical risk factors. It is important to report null findings when they arise, as meta-analyses describing the influences of relationships on health make important contributions to public health research. Research is needed to identify other factors that capture how older adults’ relationships impact their longevity in large samples that include social contextual factors and multiple indicators of physical and psychological health.