Assessment for Social Support

Reading2week3.pdf

Home >Social Science homework help >Assessment for Social Support

Social Integration, Social Support and Mortality in the US National Health Interview Survey STEVEN D. BARGER, PHD

Background: Social relationship quantity and quality are associated with mortality, but it is unclear whether each relationship dimension is equally important for longevity and whether these associations are sensitive to baseline health status. Methods: This study examined the individual and joint associations of relationship quantity (measured using a social integration score) and quality (measured by perceived social support) with mortality in a representative US sample (n = 30,574). The study also evaluated whether these associations were consistent across individuals with and without diagnosed chronic illness and whether they were independent of socioeconomic status (SES; education, income, employment, and wealth). Baseline data were collected in 2001 and were linked to vital status records 5 years later (1836 deaths). Results: Both social integration and social support were individually related to mortality (hazard ratios [HRs] = 0.83 [95% confidence interval {CI} = 0.80Y0.85] and HR = 0.94 [95% CI = 0.89Y0.98], respectively). However, in multivariate models including demographic and SES variables, social integration (HR = 0.86, 95% CI = 0.83Y0.89) but not social support (HR = 1.03, 95% CI = 0.98Y1.08) was associated with mortality. The social integration association was linear and consistent across baseline health status and men and women. Conclusions: Social integration but not social support was independently asso ciated with mortality in the US sample. This association was consistent across baseline health status and not accounted for by SES. Key words: mortality determinants, population, social networks, social support, socioeconomic factors, NHIS.

SES = socioeconomic status; NHIS = National Health Interview Survey; HR = hazard ratio.

INTRODUCTION

Having and maintaining social relationships are fundamental human motives (1). Higher-quality relationships and more frequent social contacts are associated with better health. Re lationship quality, broadly labeled functional social relation ships, reflects the social and emotional resources that people have or perceive to have available to them (2). Relationship quantity, or structural social relationships, reflects participation in a broad range of social relationships (3).

A meta-analytic review of 148 studies reported that both functional and structural relationships were inversely associ ated with mortality, with effect sizes comparable with health risks such as smoking (4). Meta-analysis is considered a high- quality research design (5), and this evidence has been cited in support of the claim that social relationships, particularly functional relationships, are important for health (6). However, there are theoretical and empirical reasons to reexamine whether functional and structural dimensions are equally im portant for mortality. For example, some theoretical models assert that the physical health benefits of structural social re lationships are a consequence of social participation itself, not the supportive functions that social relationships may provide (7). Other theories exclude supportive functions altogether (8), instead emphasizing the importance of structural social re lationships (e.g., social contact frequency) for health. Thus, several perspectives suggest that functional relationships may

From the Department of Psychology, Northern Arizona University, Flagstaff, Arizona.

Address correspondence and reprint requests to Steven D. Barger, PhD, Department of Psychology, Northern Arizona University, PO Box 15106, Flagstaff, AZ 86011. E-mail: Steven.Barger@nau.edu.

Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s Web site (www.psychosomaticmedicine.org).

Received for publication July 26, 2012; revision received January 25, 2013. DOI: 10.1097/PSY.0b013e318292ad99

not represent the social relationship dimension most relevant to mortality.

These theoretical assertions can be evaluated by concurrently comparing these social relationship dimensions in studies that included both functional and structural relationships. Such studies show a consistent association for structural relation ships, whereas the association seems to be sample size depen dent for functional relationships. For example, smaller studies (averaging G60 mortality events) find that both structural and functional relationships are inversely associated with mortality (9,10), whereas larger studies (averaging 9700 events) show no association with functional relationships when structural social relationship measures are included (11Y14). This inverse asso ciation between effect size and study size for functional re lationships signals a statistical artifact, that is, inflated effect estimates caused by small sample sizes (15Y18).

Alternatively, the association of functional relationships with mortality could be dependent on initial health status, in that the association occurs only among patient groups or those who have experienced a serious medical event.1 Patient sam ples comprise most studies (18/24) that include only functional relationship measures (4) and thus can more directly address whether the apparent survival benefit is restricted to initially unhealthy samples.

These studies also are consistent with the statistical artifact hypothesis (e.g., an inverse association between effect size and sample size) rather than the hypothesis that these associ ations are limited to unhealthy samples. Among the 24 studies in the meta-analysis, 14 found no association of functional relationships with mortality. For the remaining 10 studies (9 with patient samples, the 10th was an elderly sample averaging 85 years old), the largest 2 (with 9250 events) (19,20) reported the smallest effects, consistent with the statistical artifact ex planation (15,17,21). Moreover, age adjustment eliminated the association in one study (19), and the other study (20) was ambiguous because structural social relationship content was

1 The author is grateful to an anonymous reviewer for making this suggestion.

Psychosomatic Medicine 75:510Y517 (2013) 0033-3174/13/7505Y0510 Copyright * 2013 by the American Psychosomatic Society

510

http://www.psychosomaticmedicine.org

mailto:Steven.Barger@nau.edu

SOCIAL RELATIONSHIPS AND MORTALITY

included in the functional support measure (e.g., ‘‘I regularly meet or talk with members of my family or friends’’) (22). The strong inverse association between effect size and sample size in these 10 studies (Spearman r = j0.77, p = .009) is consistent with the statistical artifact interpretation (16,23).

The remaining seven studies also examined initially un healthy samples, but effect estimates in those studies are likely to be biased because of model overfitting (the eighth (24) in cluded SES as part of the social relationship assessment and is not considered further). Overfit regression models have an in sufficient number of events relative to the number of covariates (25). In mortality studies, the limiting sample size is determined by the number of events rather than by the total number of participants (25,26), and a ratio of 10 to 15 events per predictor is the minimum necessary to produce unbiased estimates (27). For five of these seven studies (averaging G60 events), the event per predictor ratio was 4 or less (28Y32), indicating substantial unreliability in the estimates (27). In the remaining two studies (averaging 174 events), the ratio was less than 15 (33,34). Es timates derived from a small event to predictor ratio are unlikely to replicate (18,25), an expectation confirmed by the 60% of studies in the meta-analysis detecting no association between functional social relationships and mortality. In sum, these sta tistical artifacts undermine confidence in the putative associa tion of functional relationships with mortality, and these artifacts persist when considering baseline health status. Al though meta-analytic summaries cannot overcome these limi tations (15,16,23), large, preferably representative samples should provide stable and less biased effect estimates (15Y18).

The present study evaluated the association of functional and structural social relationships with 5-year mortality in a nationally representative US sample. This sample has a large number of participants with (95000) and without (925,000) diagnosed illness, permitting comparison of these social rela tionship dimensions across baseline health status. Multivariate evaluation of other important mortality determinants, such as socioeconomic status (SES), is facilitated by the large number of mortality events (91800). SES is particularly important because it is inversely associated with mortality (35) and positively associated with social relationships (36,37). SES was assessed using education and a number of indicators of material resources (income, wealth [home ownership], and employment status) (38). Wealth and employment status measures are rarely included in this literature, but both are associated with mortality (39,40) and employment status is particularly relevant because employment provides both eco nomic and social interaction opportunities. The primary re search questions were as follows: (1) do functional and structural social relationships predict mortality individually and/ or independently? and (2) are these associations modified by initial health status or SES? Functional and structural relation ships were measured by perceived social support and social in tegration, respectively. This study also evaluated whether the form of the social relationshipYmortality association is linear or threshold (41) in addition to whether the association is consistent for men and women (11,12).

METHODS Data Source The National Health Interview Survey (NHIS) is an annual, in-person cross-

sectional interview of US households. It is the primary source of health infor mation for the noninstitutionalized US population (42). Analyses are based on NHIS sample adult participants (n = 33,326; response rate, 73.8%; aged 18Y85+ years) interviewed in 2001 who were eligible for mortality follow-up in 2006 (n = 31,358; see below). All participants provided informed consent and completed the interview in their residence. This study was exempt from human subjects review because it involved secondary analysis of publicly available data lacking identifying information.

Mortality The NHIS submitted survey records to the National Death Index for

matching and subsequent vital status ascertainment (43). This procedure cor rectly matches 98.5% of those eligible for mortality follow-up (44). In 2001, 94% (n = 31,358) of sample adult participants were eligible for mortality follow-up. The remaining 6% did not have the minimal identification data requirements for reliable matching and thus were ineligible for vital status ascertainment (43). New sample weights were created for the eligible subsample to represent the noninstitutionalized US population. Death was coded by year and quarter and included vital status follow-up through December 31, 2006. During the follow-up, 1937 people died.

Social Relationship Assessments Social support, reflecting the social resources that people perceive to be

available or are actually provided to them (2), represented the functional social relationship dimension. Social support was assessed with the question ‘‘How often do you get the social and emotional support you needValways, usually, sometimes, rarely, or never?’’ Participants with missing social support re sponses (G2%; n = 534) were excluded.

Social integration, which reflects participation in a broad range of social relationships (3), represented structural social relationships. Eight binary questions, scored 0 being no and 1 being yes, were summed to create an overall social integration score. Four questions assessed recent contacts with friends or relatives, either over the telephone or in person, excluding persons living with the respondent. Three other questions assessed attending a group social activity, a religious service, or going out to eat. All seven questions referred to activity in the past 2 weeks. The final social integration item was marital status, defined as whether respondents were married/cohabiting or not. Although marital status by itself is associated with mortality (12,45), it was included in the social inte gration score to parallel prior work showing an inverse association between social integration and mortality (40,46,47).

Owing to low frequencies in the zero and one social integration categories, these two categories were combined. Thus, social integration scores could range from 0/1 to 8. Participants received a social integration score if they had six or more valid values on the eight itemsVotherwise, they were excluded (n = 263). Missing social support and social integration values reduced the number of deaths to 1849.

SES and Demographic Variables Indicator variables were used to code years of education (less than high school,

high school diploma or equivalent, some college, college graduate or higher), household income in 2001 (US$ 0Y$24,999, $20,000Y$34,999, $35,000Y $64,999, 9$65,000), and employment (working, retired, out of work, or never worked). Wealth was indicated by home ownership (own versus renting or some other arrangement). All SES variables were retained in models regardless of sta tistical significance. Demographic variables included age, sex, and race/ethnicity (non-Hispanic white, non-Hispanic black, Hispanic, other non-Hispanic).

Household income had a large number of missing values (21%). A large proportion of missing predictors reduce the effective sample size and may result in biased and/or inefficient estimates (48,49). To overcome these potential limitations, the author used five multiply imputed family income values pro vided by the data producer (50) for SES analyses. These imputations accom modate the complex survey design, add stochastic error variability to estimates, and incorporate specialized, nonpublic survey information in the imputation

Psychosomatic Medicine 75:510Y517 (2013)

511

S. D. BARGER

procedure (e.g., income mean and standard deviation within small household TABLE 1. Baseline Demographic, Economic, and Social Characteristics area sampling units) (50). Imputed income restored the effective sample size for of 2001 US National Health Interview Survey Participants With 5-year fully adjusted regression models to 30,574 (97.5% of those eligible for mor tality follow-up, 94.8% [n = 1836] of those with ascertained vital status).

Statistical Analysis Survival time was defined as time since birth. This time scale is preferable to

one based on follow-up time (i.e., time from the baseline survey to mortality or censoring) because it provides less biased regression coefficients (51) and is preferred when age confounding is a concern (52). Analyses were stratified by 5-year birth intervals to control for cohort effects (53), and baseline age was included as a covariate. In Step 1, social support and social integration were entered individually into Cox regression models predicting survival. In Step 2, both social relationship variables were entered together. Models were adjusted for demographics in Step 3 and then SES in Step 4. To address whether the social relationshipYmortality association is dependent on initial health status, analyses were repeated for healthy and unhealthy subgroups (participants who reported at least one chronic disease at baseline). Ancillary analyses of social support only were also conducted across the healthy and unhealthy groups. All analyses in corporated the complex survey design (strata, clusters, and weights). Statistical tests were two tailed, were considered statistically significant if p e .05, and were conducted with Stata 11.2 (Stata Corp., College Station, TX).

Model adequacy was evaluated statistically and graphically. Nonlinear (squared) predictors were evaluated and discarded because they did not sig nificantly improve prediction. The proportional hazards assumption (incorpo rating clustering and weighting but not strata) for the full model was satisfied ( p = .50), and graphical inspection of social relationship residuals confirmed slopes at or very near zero. Social support and social integration were modestly correlated (r = 0.25, p G .001) and were of similar magnitude to values reported previously (4). High tolerance values (the reciprocal of the variance inflation factor) for social integration (0.85) and social support (0.92) denote the large amount of unique variance in mortality explained by these measures relative to all other predictors in multivariate models. Regression coefficients and statistical conclusions were similar to Cox models when analyzing mortality using a person/ time metric with complementary log-log regression (data not shown).

The primary outcome was all-cause mortality. To address the possibility that poor health status increases both social isolation and early mortality, sensitivity analyses were conducted, (1) excluding participants who died within 1 year after the interview and (2) including only participants free of reported disease at baseline (i.e., stroke, myocardial infarction, other coronary heart disease, or cancer, excluding nonmelanoma skin cancer). Additional analyses were re stricted to participants of working age (G65 years).

Both social relationship variables met an interval assumption, and thus, each was used as single variables in the regressions (54). However, to illustrate the form of the association, hazard ratios (HRs) are presented using indicator variables for both social support and social integration.

RESULTS Participant characteristics are presented in Table 1. Unad

justed death rates per 10,000 person-years by social support and social integration are presented in Table 2, with rates for educa tion and income provided for comparison. Social support, social integration, and SES were each inversely associated with mor tality. As expected (36,37), social relationship resources were greater at higher levels of each SES marker (see Table, Supple mental Digital Content 1, http://links.lww.com/PSYMED/A70).

When analyzed individually, social support and social in tegration were inversely associated with mortality. When both social relationship variables were entered together, social inte gration but not social support was inversely associated with mortality risk. These findings were unaffected by adjustment for age at study entry (dummy categories in addition to stratification by birth cohort), sex, and race/ethnicity and by additional ad-

Vital Status Ascertainment (n = 31,358)

Participant Characteristic M (SD) No. Weighteda %

Age, y 46.3 (17.8) 18Y24 3311 13.2 25Y34 6131 18.2 35Y44 6641 21.8 45Y54 5622 18.8 55Y64 3849 11.9 Q65 5804 16.1

Sex Women 17,694 52.0 Men 13,664 48.0

Race/Ethnicity Hispanic 5266 10.8 Non-Hispanic white 20,662 73.6 Non-Hispanic black 4324 11.3 Other non-Hispanic 1106 4.3

Educational level Less than high school 6411 17.6 High school 8905 29.3 Some college 8903 29.1 College graduate or higher 6942 23.4 Missing 197 0.6

Annual household income $0Y$19,999 6712 14.9 $20,000Y$34,999 5259 15.0 $35,000Y$64,999 6795 23.3 Q$65,000 6019 25.2 Missing 6573 21.6

Employment status Employed 20,094 66.3 Retired 5054 14.4 Not currently working 4717 14.9 Has never worked 1453 4.1 Unknown 40 0.1

Home tenure Own home 19,502 70.2 Rent/Other arrangement 11,784 29.6 Missing 72 0.2

Social support Never 861 2.4 Rarely 1114 3.0 Sometimes 4214 12.1 Usually 10,544 33.9 Always 14,091 46.9 Missing 534 1.6

Social integration score 0/1 589 1.6 2 841 2.3 3 1581 4.5 4 2912 8.2 5 5066 15.0 6 7571 23.9 7 7993 26.2 8 4542 17.6 Missing 9 2 items 263 0.8

SD = standard deviation. a Percentages are weighted to represent the noninstitutionalized US population.

Psychosomatic Medicine 75:510Y517 (2013)

512

http://links.lww.com/PSYMED/A70

SOCIAL RELATIONSHIPS AND MORTALITY

TABLE 2. Crude Mortality Rates (per 10,000) by Social Support, Social Integration, Education, and Income in the 2001 US National Health

Interview Survey

Rate 95% CI

Social support

Never 3.57 2.80Y4.63

Rarely 3.04 2.41Y3.90

Sometimes 2.83 2.50Y3.23

Usually 2.29 2.09Y2.51

Always 2.67 2.49Y2.88

Social integration score

0/1 7.39 5.99Y9.23

2 6.35 5.30Y7.68

3 5.12 4.44Y5.95

4 4.79 4.28Y5.39

5 3.28 2.95Y3.66

6 2.18 1.95Y2.45

7 1.67 1.48Y1.90

8 1.32 1.10Y1.59

Years of education

Less than high school 5.17 4.80Y5.58

High school 2.59 2.37Y2.84

Some college 1.88 1.68Y2.11

College or higher 1.53 1.34Y1.76

Family income

$0Y$19,999 4.96 4.61Y5.34

$20,000Y$34,999 3.17 2.83Y3.57

$35,000Y$64,999 1.44 1.24Y1.69

Q$65,000 1.17 0.97Y1.41

CI = confidence interval.

justment for indicators of SES (see Table 3). A similar pattern was observed when analyses were repeated for working-aged adults (G65 years; n = 543 deaths) and for participants free of baseline chronic disease (n = 875 deaths). The linear pattern of reduced risk among more socially integrated participants persisted after excluding those who died within a year of the interview (Table 3) and when limiting the sample to participants free of any func tional limitation (data not shown). Thus, the association of social integration and mortality was robust to SES adjustment and did not seem to be driven by poor initial health status causing both reduced social integration and increased mortality risk.

To illustrate the form of the association, social relationship indicator variables were entered into fully adjusted models. Those with the highest social integration had 64% lower mortality risk compared with the lowest integration category (HR = 0.36, 95% confidence interval [CI] = 0.26Y0.49). This pattern was consistent for men and women, although in these analyses, CIs excluded 1.0 only for social integration scores of 5 or more (Fig. 1). Similar results were observed when marital status was removed from the social integration measure and entered separately (see Table, Supplemental Digital Content 2, http://links.lww.com/PSYMED/A71).

When excluding social integration and restricting analyses to initially unhealthy participants, social support was inversely associated with mortality (Table 4).

This association persisted after adjustment for demographics and marital status but was much weaker when SES was included. A positive association between social support and mortality was observed when limiting analyses to healthy participants, of sim ilar direction and magnitude found in another nationally repre sentative US sample (55).

DISCUSSION This study found that social integration but not social sup

port was robustly associated with mortality. Although higher social support was associated with reduced mortality, this as sociation was fully accounted for by social integration. The greater predictive strength of social integration relative to so cial support is consistent with other large studies that jointly (11,13,14,56) or separately (12) evaluated functional and structural social relationships. Similarly, the bulk of studies examining solely functional support were conducted with un healthy participants, and in the present data, functional support was inversely associated with mortality among unhealthy but not healthy participants when social integration was excluded. However, this association was weaker and was no longer signif icant after SES adjustment. Studies that include both dimensions provide the best evidence regarding the relative importance of social relationship dimensions, and to date, such studies indicate that structural social relationships are most consistently associ ated with longevity. This could reflect the fact that maintaining some level of relationship activity is more important for mortality than the quality of those relationships (11). It is also possible that functional relationships are associated with decreased mortality risk in acute illness contexts, that is, in the months after an acute serious health event (see Footnote 1).

This study contributes to a substantial empirical literature showing decreased mortality risk among those who are more socially integrated (39,40,47,56,57) and extends that literature to show that the association exists in the general US population and is consistent across baseline health status. Inclusion of a broad range of mortality-relevant SES markers helps establish the independence of these associations from SES. The present evaluation of SES confounding is particularly diagnostic be cause, as expected (7), SES was inversely associated with both mortality and social relationships. This pattern was not observed in a similar study addressing whether the social environment- mortality association is independent of SES2 (12). Multiply im puted household income was used to manage the substantial nonresponse observed for this important SES marker, and therefore, these estimates should be more efficient and less biased compared with complete-case analyses (48,49). Although no set

2 An anonymous reviewer noted that those analyses were sex specific, revealing an inverse association with mortality for men but a direct association for women. In the present data, a consistent inverse association of education, income, and home ownership with mortality was observed for both men and women (data not shown).

Psychosomatic Medicine 75:510Y517 (2013)

513

http://links.lww.com/PSYMED/A71

http:0.26Y0.49

S. D. BARGER

TABLE 3. Cox Proportional Hazard Ratios (HRs) for All-Cause Mortality by Social Support and Social Integration, 2001 US National Health Interview Survey

Social Support Social Integration Score

HR 95% CI HR 95% CI

Model 1: social relationship variable by itself 0.94 0.89Y0.98 0.83 0.80Y0.85

Model 2: both social relationship variables 1.01 0.96Y1.06 0.83 0.80Y0.85

Model 3: additionally adjusted for demographicsa 1.00 0.95Y1.05 0.83 0.80Y0.85

Model 4: additionally adjusted for SESb 1.03 0.98Y1.08 0.86 0.83Y0.89

Model 4: sensitivity analyses: limit sample to:

Participants G65 years 1.11 1.01Y1.22 0.86 0.81Y0.91

Participants free of baseline chronic diseasec 1.03 0.96Y1.12 0.88 0.84Y0.92

Participants with baseline chronic disease 1.03 0.96Y1.11 0.85 0.81Y0.88

Participants who died 91 y after the interview 1.00 0.94Y1.05 0.87 0.84Y0.90

CI = confidence interval; SES = socioeconomic status. a Demographic variables included age, sex, and race/ethnicity. b SES variables included years of education, employment status (unemployed, employed, never worked, retired), home ownership, and multiply imputed household income. c Chronic diseases included prior myocardial infarction, stroke, cancer (excluding nonmelanoma skin cancer), or other coronary heart disease.

of indicators can decisively rule out SES confounding, the HRs for social integration were insensitive to SES adjustment.

The form of the association between social integration and mortality was linear when using a summary social integration variable and was statistically significant above a social inte gration score of 4 when using categorical indicators. When the number of events was reduced in subgroup analyses, the linear form persisted, but CIs were wider and only the highest social contact categories had CIs that did not include 1.0 (data not shown). This apparent difference may be caused by the large number of events necessary to distinguish adjacent social in tegration intervals. From a practical standpoint, it is reason

able to conclude that those with moderate to high levels of social integration are at lower risk for mortality.

Strengths and Limitations This study used a nationally representative adult sample of

more than 30,000 participants followed for 5 years. To date, this is the largest population-based study of functional and structural social relationships and mortality (see Hummer et al. (56) for a closely related US study). Both social relationship dimensions were tested simultaneously, and the large number of mortality events permitted a statistically appropriate evaluation of a di verse set of mortality-relevant SES covariates, many of which

Figure 1. Mortality hazard ratios for social support (circles) and social integration (triangles) in the 2001 National Health Interview Survey. Vertical bars represent 95% confidence intervals. A, For the full sample (n = 30,574; 1836 events). B and C, For women (929 events) and men (907 events), respectively. All analyses adjusted for demographics and socioeconomic status including multiply imputed income values.

Psychosomatic Medicine 75:510Y517 (2013)

514

SOCIAL RELATIONSHIPS AND MORTALITY

TABLE 4. Cox Proportional Hazard Ratios (HR) for 5-Year All-Cause Mortality by Social Support and Baseline Health Status: 2001 US National Health Interview Survey

Model 1: Social Support Alone

Model 2: + Demographics Model 3: + Marital Status Model 4: + SES

HR 95% CI HR 95% CI HR 95% CI HR 95% CI

Baseline unhealthya

Social support

Never 1.0 Reference 1.0 Reference 1.0 Reference 1.0 Reference

Rarely 0.66 0.41Y1.06 0.72 0.45Y1.14 0.69 0.44Y1.10 0.81 0.51Y1.29

Sometimes 0.70* 0.50Y0.98 0.73 0.52Y1.03 0.72 0.51Y1.02 0.79 0.55Y1.12

Usually 0.61** 0.45Y0.83 0.62** 0.46Y0.85 0.64** 0.46Y0.88 0.74 0.53Y1.04

Always 0.65** 0.48Y0.89 0.65** 0.48Y0.89 0.69* 0.50Y0.95 0.81 0.58Y1.13

Observations 5257 5257 5257 5232

Baseline healthy

Social support

Never 1.0 Reference 1.0 Reference 1.0 Reference 1.0 Reference

Rarely 1.45 0.82Y2.57 1.45 0.82Y2.58 1.45 0.83Y2.56 1.44 0.83-2.53

Sometimes 1.24 0.80Y1.92 1.26 0.82Y1.94 1.28 0.83Y1.96 1.25 0.82Y1.93

Usually 1.09 0.71Y1.68 1.12 0.73Y1.72 1.18 0.77Y1.81 1.27 0.83Y1.94

Always 1.01 0.65Y1.56 1.03 0.66Y1.59 1.11 0.71Y1.72 1.21 0.78Y1.87

Observations 25,567 25,567 25,567 25,429

SES = socioeconomic status; CI = confidence interval. * p G .05, ** p G .01, *** p G .001. a Unhealthy participants reported at least one of the following conditions: prior myocardial infarction, stroke, cancer (excluding nonmelanoma skin cancer), or other coronary heart disease. Healthy participants reported none of these conditions. Demographic variables included continuous age, sex, and race/ethnicity. SES variables included multiply imputed household income (see ‘‘Methods’’), workforce status (employed, unemployed, retired, never worked), and home ownership (yes or no). Entering the social integration variable in every model eliminated the social support/mortality association (data not shown).

have not been previously considered together. This study also evaluated the consistency of the social relationshipYmortality association across large subsamples of both initially healthy and unhealthy participants, a unique contribution to the literature.

The one-time assessment of functional and structural social relationships is a potential limitation of the study. Although single baseline assessments are common in the literature (4) and social relationship resources are reasonably stable (58,59), repeated social relationship measurements may improve expo sure classification and provide stronger effect estimates. Con versely, the 5-year follow-up interval in this study should increase the accuracy of participants’ social relationship classi fication, and the effect estimates were similar to those observed in studies with longer follow-up intervals (11,14). It is also pos sible that health selection effects persist within this follow-up time frame (60) and that future studies with more detailed or alternative social relationship assessments could detect a mor tality association.

Emotional support was measured with a single item, which is potentially less reliable and therefore less predictive when com pared with a social integration measure based on multiple items. However, other evidence shows that the emotional support item used in this study has excellent predictive validity. For example, in a multivariate model predicting psychological well-being, re moving the one-item emotional support variable decreased the explained variance by 8% (0.25Y0.17), whereas removing the

eight-item social integration variable from the same model re duced the explained variance by only 2% (0.25Y0.23) (cf. Barger et al. (61)). This emotional support item was also independently associated with objectively assessed health knowledge in the 2001 NHIS and that association was replicated in a separate study of more than 87,000 US adults (62). A third study showed that this item was associated with mental health, activity limitations, and health risk behavior in a nationally representative sample of more than 330,000 adults (63). Thus, the imbalance in the number of items representing structural and functional di mensions does not result in a global insensitivity for predicting health end points. Perhaps a more parsimonious explanation is that social support is independently associated with health outcomes other than mortality (61,62,64Y66), whereas social integration is more consistently associated with mortality (11,12,40,47,56).

Finally, these data are observational, and therefore, causality cannot be confidently established. In contrast, randomized controlled trials (RCTs) provide a stronger foundation for causal inference, but the availability of such trials is limited. On the other hand, effect estimates for well-designed observational studies closely match those of RCTs (67,68), and observational studies of representative samples may provide more ecologi cally valid estimates. Given the barriers to conducting an RCT in a comparable population-based setting, readers should weigh the observational evidence in light of sample sizes, adequacy of

Psychosomatic Medicine 75:510Y517 (2013)

515

http:0.25Y0.23

http:0.25Y0.17

S. D. BARGER

control for plausible confounding variables, and sample char acteristics to determine which aspects of the social environment are most important for mortality.

CONCLUSIONS Social integration was strongly and consistently associated

with 5-year mortality risk in the US population, whereas social support had a weaker association that was fully explained by social integration. Although SES was associated with both social relationships and mortality, SES did not alter the robust linear association of social integration with mortality. Consis tent with theoretical predictions (8,69), social resources predict mortality regardless of sex, initial health status, and other established mortality determinants.

Source of Funding and Conflicts of Interest: The author reports no source of funding and no conflicts of interest.

REFERENCES 1. Baumeister RF, Leary MR. The need to belong: desire for interpersonal

attachments as a fundamental human motivation. Psychol Bull 1995;117: 497Y529.

2. Cohen S, Gottlieb BH, Underwood LG. Social relationships and health. In: Cohen S, Gottlieb BH, Underwood LG, editors. Measuring and In tervening in Social Support. New York: Oxford University Press; 2000: 3Y25.

3. Cohen S. Social relationships and health. Am Psychol 2004;59:676Y84. 4. Holt-Lunstad J, Smith TB, Layton JB. Social relationships and mortality

risk: a meta-analytic review. PLoS Med 2010;7:e1000316. 5. Patsopoulos NA, Analatos AA, Ioannidis JP. Relative citation impact of

various study designs in the health sciences. JAMA 2005;293:2362Y6. 6. Uchino BN, Bowen K, Carlisle M, Birmingham W. Psychological path

ways linking social support to health outcomes: a visit with the ‘‘ghosts’’ of research past, present, and future. Soc Sci Med 2012;74:949Y57.

7. Berkman LF, Glass T, Brissette I, Seeman TE. From social integration to health: Durkheim in the new millennium. Soc Sci Med 2000;51:843Y57.

8. Oakes JM, Rossi PH. The measurement of SES in health research: current practice and steps toward a new approach. Soc Sci Med 2003;56:769Y84.

9. Blazer DG. Social support and mortality in an elderly community popu lation. Am J Epidemiol 1982;115:684Y94.

10. Hanson BS, Isacsson S-O, Janzon L, Lindell S-E. Social network and social support influence mortality in elderly men. Am J Epidemiol 1989; 130:100Y11.

11. House JS, Robbins C, Metzner HL. The association of social relationships and activities with mortality: prospective evidence from the Tecumseh Community Health Study. Am J Epidemiol 1982;116:123Y40.

12. Stringhini S, Berkman L, Dugravot A, Ferrie JE, Marmot M, Kivimaki M, Singh-Manoux A. Socioeconomic status, structural and functional mea sures of social support, and mortality. Am J Epidemiol 2012;175:1275Y83.

13. Fuhrer R, Dufouil C, Antonucci TC, Shipley MJ, Helmer C, Dartigues JF. Psychological disorder and mortality in French older adults: do social relations modify the association? Am J Epidemiol 1999;149:116Y26.

14. Helweg-Larsen M, KjLller M, Thoning H. Do age and social relations moderate the relationship between self-rated health and mortality among adult Danes? Soc Sci Med 2003;57:1237Y47.

15. Bakker M, van Dijk A, Wicherts JM. The rules of the game called psy chological science. Perspect Psychol Sci 2012;7:543Y54.

16. Ferguson CJ, Heene M. A vast graveyard of undead theories: publication bias and psychological science’s aversion to the null. Perspect Psychol Sci 2012;7:555Y61.

17. Ioannidis JP, Panagiotou OA. Comparison of effect sizes associated with biomarkers reported in highly cited individual articles and in subsequent meta-analyses. JAMA 2011;305:2200Y10.

18. Ioannidis JP. Why most published research findings are false. PLoS Med 2005;2:e124.

19. Maier H, Smith J. Psychological predictors of mortality in old age. J Gerontol B Psychol Sci Soc Sci 1999;54:P44Y54.

20. Brummett BH, Mark DB, Siegler IC, Williams RB, Babyak MA, Clapp- Channing NE, Barefoot JC. Perceived social support as a predictor of mortality in coronary patients: effects of smoking, sedentary behavior, and depressive symptoms. Psychosom Med 2005;67:40Y5.

21. Ferguson CJ, Brannick MT. Publication bias in psychological science: prevalence, methods for identifying and controlling, and implications for the use of meta-analyses. Psychol Methods 2012;17:120Y8.

22. Cohen S, Mermelstein R, Kamarck TW, Hoberman HM. Measuring the functional components of social support. In: Sarason IG, Sarason BR, editors. Social Support: Theory, Research and Applications. New York: Springer-Verlag; 1985:73Y94.

23. Levine TR, Asada KJ, Carpenter C. Sample sizes and effect sizes are negatively correlated in meta-analyses: evidence and implications of a publication bias against nonsignificant findings. Commun Monogr 2009;76:286Y302.

24. Chacko RC, Harper RG, Gotto J, Young J. Psychiatric interview and psychometric predictors of cardiac transplant survival. Am J Psychiatry 1996;153:1607Y12.

25. Babyak MA. What you see may not be what you get: a brief, nontechnical introduction to overfitting in regression-type models. Psychosom Med 2004;66:411Y21.

26. Hsieh FY, Lavori PW. Sample-size calculations for the Cox proportional hazards regression model with nonbinary covariates. Control Clin Trials 2000;21:552Y60.

27. Peduzzi P, Concato J, Feinstein AR, Holford TR. Importance of events per independent variable in proportional hazards regression analysis. II. Ac curacy and precision of regression estimates. J Clin Epidemiol 1995;48: 1503Y10.

28. Christensen AJ, Wiebe JS, Smith TW, Turner CW. Predictors of survival among hemodialysis patients: effect of perceived family support. Health Psychol 1994;13:521Y5.

29. Colon EA, Callies AL, Popkin MK, McGlave PB. Depressed mood and other variables related to bone marrow transplantation survival in acute leukemia. Psychosomatics 1991;32:420Y5.

30. Kimmel PL, Peterson RA, Weihs KL, Shidler N, Simmens SJ, Alleyne S, Cruz I, Yanovski JA, Veis JH, Phillips TM. Dyadic relationship conflict, gender, and mortality in urban hemodialysis patients. J Am Soc Nephrol 2000;11:1518Y25.

31. Lee M, Rotheram-Borus MJ. Challenges associated with increased survival among parents living with HIV. Am J Public Health 2001;91:1303Y9.

32. Lehto US, Ojanen M, Dyba T, Aromaa A, Kellokumpu-Lehtinen P. Baseline psychosocial predictors of survival in localized melanoma. J Psychosom Res 2007;63:9Y15.

33. Coyne JC, Rohrbaugh MJ, Shoham V, Sonnega JS, Nicklas JM, Cranford JA. Prognostic importance of marital quality for survival of congestive heart failure. Am J Cardiol 2001;88:526Y9.

34. Woloshin S, Schwartz LM, Tosteson AN, Chang CH, Wright B, Plohman J, Fisher ES. Perceived adequacy of tangible social support and health out comes in patients with coronary artery disease. J Gen Intern Med 1997; 12:613Y8.

35. Mackenbach JP, Stirbu I, Roskam AJ, Schaap MM, Menvielle G, Leinsalu M, Kunst AE. Socioeconomic inequalities in health in 22 European countries. N Engl J Med 2008;358:2468Y81.

36. O’Brien KM. Healthy, wealthy, wise? Psychosocial factors influencing the socioeconomic statusYhealth gradient. J Health Psychol 2012;17:1142Y51.

37. Klein J, Vonneilich N, Baumeister S, Kohlmann T, von dem Knesebeck O. Do social relations explain health inequalities? Evidence from a longitu dinal survey in a changing eastern German region. Int J Public Health 2012;57:619Y27.

38. Galobardes B, Shaw M, Lawlor DA, Lynch JW, Davey Smith G. Indicators of socioeconomic position (part 1). J Epidemiol Community Health 2006; 60:7Y12.

39. Malmstrom M, Johansson SE, Sundquist J. A hierarchical analysis of long-term illness and mortality in socially deprived areas. Soc Sci Med 2001;53:265Y75.

40. Orth-Gomer K, Johnson JV. Social network interaction and mortality. A six year follow-up study of a random sample of the Swedish population. J Chronic Dis 1987;40:949Y57.

41. House JS. Social isolation kills, but how and why? Psychosom Med 2001;63:273Y4.

42. Centers for Disease Control and Prevention. About the National Health In terview Survey. Available at: http://www.cdc.gov/nchs/nhis/about_nhis.htm: Accessed. May 9, 2012.

43. National Center for Health Statistics. Office of Analysis and Epidemiology. The National Health Interview Survey (1986Y2004) Linked Mortality

Psychosomatic Medicine 75:510Y517 (2013)

516

http://www.cdc.gov/nchs/nhis/about_nhis.htm

SOCIAL RELATIONSHIPS AND MORTALITY

Files, mortality follow-up through 2006: matching methodology, May 2009, Hyattsville, MD; 2009.

44. Ingram DD, Lochner KA, Cox CS. Mortality experience of the 1986Y2000 National Health Interview Survey Linked Mortality Files participants. Vital Health Stat 2 2008:1Y37.

45. Rutledge T, Matthews K, Lui LY, Stone KL, Cauley JA. Social networks and marital status predict mortality in older women: prospective evidence from the Study of Osteoporotic Fractures (SOF). Psychosom Med 2003;65:688Y94.

46. Berkman LF, Melchior M, Chastang JF, Niedhammer I, Leclerc A, Goldberg M. Social integration and mortality: a prospective study of French employees of Electricity of FranceYGas of France: the GAZEL Cohort. Am J Epidemiol 2004;159:167Y74.

47. Kawachi I, Colditz GA, Ascherio A, Rimm EB, Giovannucci E, Stampfer MJ, Willett WC. A prospective study of social networks in relation to total mortality and cardiovascular disease in men in the USA. J Epidemiol Com munity Health 1996;50:245Y51.

48. Donders AR, van der Heijden GJ, Stijnen T, Moons KG. Review: a gentle in troduction to imputation of missing values. J Clin Epidemiol 2006;59:1087Y91.

49. Schafer JL. Multiple imputation: a primer. Stat Methods Med Res 1999; 8:3Y15.

50. Schenker N, Raghunathan TE, Chu P-L, Makuc DM, Zhang G, Cohen AJ. Multiple imputation of family income and personal earnings in the Na tional Health Interview Survey: methods and examples. Available at: http:// www.cdc.gov/nchs/data/nhis/tecdoc.pdf: Accessed. November 30, 2012.

51. Thiebaut AC, Benichou J. Choice of time-scale in Cox’s model analysis of epidemiologic cohort data: a simulation study. Stat Med 2004;23: 3803Y20.

52. Cologne J, Hsu WL, Abbott RD, Ohishi W, Grant EJ, Fujiwara S, Cullings HM. Proportional hazards regression in epidemiologic follow-up studies: an intuitive consideration of primary time scale. Epidemiology 2012;23: 565Y73.

53. Korn EL, Graubard BI, Midthune D. Time-to-event analysis of longitu dinal follow-up of a survey: choice of the time-scale. Am J Epidemiol 1997;145:72Y80.

54. Long JS, Freese J. Regression Models for Categorical Outcomes Using Stata. 2nd ed. College Station, TX: Stata Press; 2003.

55. Musick MA, House JS, Williams DR. Attendance at religious services and mortality in a national sample. J Health Soc Behav 2004;45:198Y213.

56. Hummer RA, Rogers RG, Nam CB, Ellison CG. Religious involvement and U.S. adult mortality. Demography 1999;36:273Y85.

57. Berkman LF, Syme SL. Social networks, host resistance, and mortality: a nine-year follow-up study of Alameda County residents. Am J Epidemiol 1979;109:186Y204.

58. Croezen S, Haveman-Nies A, Picavet HS, Smid EA, de Groot CP, Van’t Veer P, Verschuren WM. Positive and negative experiences of social support and long-term mortality among middle-aged Dutch people. Am J Epidemiol 2010;172:173Y9.

59. Eng PM, Rimm EB, Fitzmaurice G, Kawachi I. Social ties and change in social ties in relation to subsequent total and cause-specific mortality and coronary heart disease incidence in men. Am J Epidemiol 2002;155:700Y9.

60. Fox AJ, Goldblatt PO, Jones DR. Social class mortality differentials: ar tefact, selection or life circumstances? J Epidemiol Community Health 1985;39:1Y8.

61. Barger SD, Donoho CJ, Wayment HA. The relative contributions of race/ ethnicity, socioeconomic status, health, and social relationships to life satisfaction in the United States. Qual Life Res 2009;18:179Y89.

62. Barger SD. Perceived emotional support and frequent social contacts are associated with greater knowledge of stroke warning signs: evidence from two cross-sectional US population surveys. J Health Psychol 2012;17: 169Y78.

63. Strine TW, Chapman DP, Balluz L, Mokdad AH. Health-related quality of life and health behaviors by social and emotional support. Their relevance to psychiatry and medicine. Soc Psychiatry Psychiatr Epidemiol 2008; 43:151Y9.

64. Hays JC, Landerman LR, George LK, Flint EP, Koenig HG, Land KC, Blazer DG. Social correlates of the dimensions of depression in the elderly. J Gerontol B Psychol Sci Soc Sci 1998;53:P31Y9.

65. Cornwell EY, Waite LJ. Social disconnectedness, perceived isolation, and health among older adults. J Health Soc Behav 2009;50:31Y48.

66. George LK, Blazer DG, Hughes DC, Fowler N. Social support and the outcome of major depression. Br J Psychiatry 1989;154:478Y85.

67. Benson K, Hartz AJ. A comparison of observational studies and ran domized, controlled trials. N Engl J Med 2000;342:1878Y86.

68. Concato J, Shah N, Horwitz RI. Randomized, controlled trials, observa tional studies, and the hierarchy of research designs. N Engl J Med 2000;342:1887Y92.

69. Marmot M. The Status Syndrome: How Social Standing Affects Our Health and Longevity. New York: Times Books; 2004.

Psychosomatic Medicine 75:510Y517 (2013)

517

http://www.cdc.gov/nchs/data/nhis/tecdoc.pdf