Daily Social Interactions, Close Relationships, and Systemic Inflammation in Two Samples: Healthy Middle-Aged and Older Adults

Abstract

Objective:

Systemic inflammation is thought to be a biological mediator between social relationship quality and premature mortality. Empirical work has yielded mixed support for an association of social relationship variables with systemic inflammation, perhaps due to methodological limitations. To date, research in this literature has focused on global perceptions of social relationships, with limited attention to the covariance of characteristics of daily social interactions with inflammation. Here, we examine whether daily interactions, as assessed by ecological momentary assessment (EMA), associate with peripheral markers of inflammation among midlife and older adults.

Methods:

Global social support and integration were measured using the Interpersonal Support Evaluation List (ISEL) and the Social Network Index (SNI), respectively, in older adults from the Pittsburgh Healthy Heart Project (PHHP), and in middle-aged adults from the Adult Health and Behavior Project–II (AHAB-II). Using time-sampled EMA, we assessed the proportion of the day spent in positive and negative social interactions. Systemic markers of inflammation were interleukin (IL)-6 and C-reactive protein (CRP).

Results:

Global measures of support and integration did not associate with inflammation in either sample. In older adults, relative frequency of total positive interactions, those with close others (i.e. spouse, friends, family), and those with coworkers predicted lower concentrations of IL-6 in fully adjusted models, accounting for age, sex, race, education, BMI, smoking and alcohol. In middle-aged adults, relative frequency of positive interactions with close others was also inversely associated with IL-6 level and relative frequency of negative marital interactions was unexpectedly inversely associated with CRP level.

Conclusions:

Characteristics of daily social interactions among midlife and older adults associate with markers of systemic inflammation that are known to predict risk for cardiovascular disease. Ambulatory measures may better capture health-relevant social processes in daily life than retrospective, global self-report measures.

1. Introduction

Social relationship characteristics, or the quality and quantity of one’s social ties, have been shown to be associated with a variety of health outcomes, including reduced risk of cardiovascular disease (CVD) (Berkman et al., 1992; Brummet et al., 2001), lower incidence of cancer (Ell et al., 1992; Hibbard & Pope, 1993; Welin et al., 1992) and infectious diseases (Lee & Rotheram-Borus, 2001; Patterson et al., 1996), and greater longevity (Holt-Lunstad et al., 2010). Although pathways linking social factors to health outcomes remain unclear, it is widely suggested that the immune system may play a role (Kiecolt-Glaser et al., 2010). Indeed, a growing number of studies have examined the association between social connections and inflammation, including circulating levels of pro-inflammatory cytokines, such as interleukin (IL)-6, and acute phase proteins, such as C-reactive protein (CRP) (Mezuk et al., 2010; Ford et al., 2006). This proinflammatory profile is implicated in the pathophysiology of chronic disease (Black & Garbutt, 2002), suggesting that increased inflammation might mediate associations between relational characteristics and morbidity and/or premature mortality. Social relationship characteristics could be associated with markers of inflammation through engagement in risky health behaviors, such as excessive alcohol intake and smoking, both of which are related to greater inflammation (Imhof et al., 2001; Frohlich et al., 2003). Alternatively, relational characteristics could also be associated with inflammation through psychophysiological mechanisms, such as dysregulation of the sympathetic-nervous system (SNS) and the hypothalamic-pituitary-adrenal (HPA) axis, both of which are involved in modulation of inflammation (Eisenberger & Cole, 2012).

Social relationship characteristics can be conceptualized within two domains: social integration and social support. Social integration (SI) is considered a structural index of an individual’s social ties, reflecting social engagement in diverse social roles, whereas social support (SS) is a measure of perceived availability of material aid and emotional support from others. While there is some evidence that low SI and low functional SS associate with chronic inflammation, findings are inconsistent and may differ by age and gender. A recent study reported only a weak inverse association between SI and CRP and an unexpected positive association between perceived SS and CRP (Glei et al., 2012). Three epidemiological studies reported inverse associations between SI and concentration of CRP in men, but not in women (Loucks et al., 2006a; Ford et al., 2006; Loucks et al., 2006b), with some evidence that the association is stronger for older men than younger men (Ford et al., 2006). Similar inverse relationships of perceived SS with CRP among men have been observed in some (Mezuk et al., 2010), but not all (McDade et al., 2006) studies.

In addition to the potential modifying role of sample characteristics, variability in associations across specific relationship domains may also help explain inconsistent findings. Recent evidence suggests that qualitative aspects of social relationships within specific relationship domains may contribute to inconsistent findings across studies (Shor et al., 2013). In this regard, relationship quality may be an important factor to consider given evidence that strain in close relationships (i.e. spouse, friends, family) and total relationship strain associates with higher circulating levels of fibrinogen, an acute phase marker of inflammation, above and beyond the effects of demographic factors, BMI, smoking, and physical activity (Yang et al., 2014).

Similarly, within the context of marriage, lower spousal support, but not general marital strain, associated with higher levels of IL-6 and CRP among women (Donoho et al., 2013). In a separate study, both partner support and partner strain were associated with IL-6 (in the expected directions) in younger women, but not in men, with only partner support emerging as a significant predictor when both measures of marital adjustment were considered simultaneously (Whisman & Sbarra, 2012). Hence, in the context of marriage, positive relationship quality appears to be more consistently associated with inflammatory markers than measures of negative interaction, especially in women.

In addition to demographic and relationship quality issues, limitations in the methodology used to measure social relationship constructs may also contribute to inconsistent association with inflammation. Self-report measures that assess global beliefs about SS and SI behaviors do not assess dynamic and event-specific behavioral mechanisms, such as daily social interactions, that may contribute to disease risk. In this regard, global self-report measures may more strongly reflect beliefs and attributions about relationships than the frequency and quality of social interactions (Conner & Barrett, 2012; Tulving, 1983). It may be the case that actual social events act as triggers of the biological changes that promote inflammation, and that our ratings of the quality of our relationships are only indirect markers of these psychophysiological processes. In response to these limitations, it has been argued that multiple ambulatory assessments in naturalistic settings may provide more accurate representations of event-specific experiences (Conner & Barrett, 2012; Tulving, 1983). Ecological momentary assessment (EMA) is a method designed to capture event-specific information, measuring behaviors, affect, and cognitions in real-time and in the natural environment (Stone & Shiffman, 1994). Thus, EMA may provide greater access to the dynamic biological and behavioral mechanisms that promote disease risk (Shiffman, Stone, Hufford, 2008). To support this theory, a separate body of literature reports consistent associations between characteristics of daily social interactions, measured through EMA or daily diaries, and inflammation. For example, greater interpersonal strain in the domains of family, peers, and school, measured through daily diaries, predicted higher levels of CRP in a sample of adolescents (Fuligni et al., 2009), and negative and competitive interactions in daily life predicted elevated levels of sTNFαRII (type II receptor for pro-inflammatory cytokine TNFα) and elevated levels of IL-6, respectively (Chiang et al., 2012).

Social relationships may also be particularly important for the health and well-being of older adults, given that social roles and prioritization of close relationships (Carstensen et al., 1999) may change with age in parallel with increased risk for chronic inflammatory disease of aging. Psychological and biological processes linking social relationships to inflammation may differ by age and gender, qualitative features of social relationships may be particularly important within specific relationships, and the measurement of specific interactions within daily life may be an important consideration in capturing these effects. Therefore, in the two samples studied below, we test cross-sectional associations of both global measures of SS and SI and of EMA measures of social processes (measured by the relative frequency and quality of social interactions) with circulating markers of systemic inflammation, as measured by CRP and IL-6. EMA measurements were used to examine the nature of social interactions across different domains of social relationships, including those with spouses, close friends, and family. Findings from two separate samples are examined: (1) a sample of older adults from the Pittsburgh Healthy Heart Project (PHHP) and (2) a sample of middle-aged adults from the Adult Health and Behavior Phase II (AHAB-II) project. In both cases, it is hypothesized that lower IL-6 and CRP concentrations will associate with, i) greater SS and SI, assessed by global measures, ii) greater relative frequency of interactions across all observations, iii) greater relative frequency of overall positive interactions, especially with spouse, friends and family, and iv) lower relatively frequency of negative interactions. By examining quantitative and qualitative aspects of social network characteristics and daily social interactions in the natural environment, across relationship domains, we hope to learn about the role of specific relationships and age-related effects that may account for inconsistent findings in this literature examining social relationship characteristics and inflammation.

2. Method

2.1. Participants.

Participants included 306 men and women enrolled in the Pittsburgh Healthy Heart Project (PHHP), a prospective study of healthy, community-dwelling adults aged 50–70 years, and 419 men and women enrolled in the Adult Health and Behavior Project – Phase 2 (AHAB-II), aged 30–54. For both samples, exclusionary criteria included: (a) history of schizophrenia, bipolar disorder, chronic hepatitis, chronic lung disease, hypertension, and heart, renal, or neurological conditions; (b) drinking > 5 portions of alcohol, 3 times or more/week; (c) prescription of insulin or glucocorticoid, anti-arrhythmic, antihypertensive, lipid-lowering, or prescription weight-loss medications; (d) pregnancy or lactation. In AHAB-II, participants were additionally required to be working at least 25 hours per week outside the home, had to refrain from fish-oil supplements (due to requirements for another substudy), had more than 8^th grade reading skills, and could not be shift workers. In AHAB-II, those on any psychotropic medication were excluded, whereas in PHHP, only medications with significant autonomic effects were disallowed, permitting those on SSRIs to participate. In addition, although individuals with a history of chronic disease were generally not recruited in the PHHP sample, people with diabetes who were not taking insulin, those with a history of cancer but no treatment in the past 6 months, and those with mild or moderate rheumatoid arthritis were eligible.

AHAB-II participants were recruited between February 2008 and August 2011 through mass mailings to individuals selected from voter registration and other public domain lists. PHHP participants were similarly recruited between September 1998 and April 2000. Both studies were approved by the University of Pittsburgh Institutional Review Board. Participants in AHAB-II received compensation up to $410, depending on extent of participation in study visits and protocol compliance. Participants in PHHP received $200 for participation in baseline measures.

2.2. Procedure.

In each study, participants completed multiple laboratory visits, some of which are not relevant to this report. In each case, demographic assessments and a fasting blood draw were completed at Visit 1.

In PHHP, measures of global SS and SI were assessed at Visit 2. Ecological momentary assessments were completed every 45 minutes over 6 days, in two 3-day periods, separated by four months. In AHAB-II, measures of global SS and SI were assessed at Visit 4. Ecological momentary assessments were completed on an hourly basis between Visits 2 and 3 over a 4-day monitoring protocol (3 working days and 1 nonworking day). The monitoring protocol consisted of two, 2-day monitoring periods, usually one period at the beginning of the work week and another at the end of the work week, with at least one non-monitoring day in between.

2.3. Instruments

2.3.1. Social Support and Social Integration.

In both samples, perceived SS was measured by the Interpersonal Support Evaluation List (ISEL), assessing tangible support, belonging support, and appraisal support (Cohen et al., 1985). Each item was scored on a 4-point scale and scores were summed and averaged across the 3 subscales. SI was measured by the Social Network Inventory (SNI). The SNI assesses participation in 12 types of relationships; one point is assigned for each role the individual participates in within their social network at least once every 2 weeks (a measure of “social role diversity”). SI was measured as the sum score on the SNI. Both questionnaires have shown adequate validity and test-retest reliabilities (Delistamati et al., 2006; Treadwell et al., 1993; Cohen et al., 2012).

2.3.2. Social interaction measures.

In both samples, the proportion of the day spent in social interactions, and positive and negative interactions was measured through periodic momentary interviews on an electronic diary (ED) (Palm ™ Pilot Professional handheld, Palm, Inc, Santa Clara, CA was used in the PHHP sample, Palm Z22 was used in the AHAB-II sample). Additional ambulatory monitoring devices (not relevant for the current report) were also used in both studies.

In the PHHP sample, participants were trained to use the ED device and practiced using the device for an additional day. Participants then returned for a “Shakedown” visit, where the participant’s data were reviewed by a research assistant. If there were no questions or concerns, the monitoring period began the following day. The ED interviews were administered every 45 minutes and consisted of 41–43 questions regarding mental state, mood, and the physical and social environment. A subset of these items assessed whether the participant was currently in a social interaction, (if not) the interval since the most recent interaction, the duration of interaction, the type of interaction (in person vs. phone, etc.), the quality of interaction, and the partners involved in the interaction. Although this EMA protocol requires considerable engagement and participation from subjects, the protocol from these two samples has provided valid measures of social interaction that associate in the expected direction with existing measures of social relationship characteristics (Janicki et al., 2006; Vella et al., 2008), and with various health outcomes, including measures of carotid artery intima-media thickness (IMT) (Janicki et al., 2005; Joseph et al., 2014).

In AHAB-II, participants were similarly trained to use their ED, with a one day practice and data review during a “Shakedown” phone call. During the monitoring period, participants were prompted by the PDA to complete a 43-item questionnaire on an hourly basis throughout the waking day. The interview consisted of 11 items pertaining to daily social interactions. A subset of the interview questions assessed the same social interaction characteristics as for PHHP (e.g. length, partner, etc.). See Appendix A and B.

There were some notable differences between the protocols of each sample. Firstly, PHHP used 3-item scales to assess positive and negative interactions, whereas AHAB-II used 2-item scales to assess each. Secondly, in PHHP, participants were presented with a visual analogue scale that was converted to an 11 point Likert rating (e.g. Pleasant? Yes ----|----No), whereas in AHAB-II, participants were presented with a 6-point Likert scale (e.g. NO! No no yes Yes YES!) for each item. And lastly, the schedule of monitoring (i.e. frequency of interviews, duration of monitoring periods) differed between the two studies, as described above. Most notably, the PHHP protocol involved 6 days of monitoring, as compared with 4 days in the AHAB-II study.

In both samples, 3 summary scores were derived for each individual based on: 1) relative frequency of total interactions, defined as the proportion of all observations that were spent in social interactions currently or in the 10 min prior to each interview (range = 0–100%), 2) relative frequency of positive interactions, defined as the proportion of observations that were rated as positive (i.e. proportion of interviews in which a) there was a current or recent interaction, and b) in which 1 or more items was rated ≥ the midpoint of the scale on items regarding “pleasant,” “agreeable” or “friendly” interactions in PHHP, or scored as yes, Yes or YES! on the “pleasant” and “agreeable” interactions in AHAB-II), and 3) relative frequency of negative interactions (i.e. proportion of interviews in which a) there was a current or recent interaction and b) in which 1 or more items was rated ≥ the midpoint of the scale on items regarding whether a participant was in “conflict” or “treated badly” or scored as yes, Yes or YES! in AHAB-II). Although previous reports using the social interaction data from AHAB-II sample have measured mean negativity or mean positivity within interactions (Joseph et al., 2014), the relative frequency measure in the current report was adapted here in order to take into consideration temporal exposure as a potentially important dimension of psychosocial risk (Kamarck et al., 2012). We additionally examined the frequency of positive and negative interactions with close others (i.e. spouse, friends, family member) and coworkers. In PHHP, individuals who were not employed full time or part time also reported interactions with coworkers; presumably, these were individuals interpreted to be colleagues in a professional, academic, or volunteering setting.

2.3.3. Inflammation measures.

In PHHP, blood was drawn between 8:30 am and 11:30 am at the baseline screening visit. Participants were instructed to fast and to avoid caffeine for 12 hours prior to these visits. Blood samples, collected in tubes with no additives, were centrifuged within 3 hours of collection to isolate serum. Serum aliquots were frozen at −70 degrees Celsius until the time of assay. Serum samples were sent to the Laboratory for Clinical Biochemistry Research at the University of Vermont. There, IL-6 was measured using ultra-sensitive enzyme-linked immunosorbent assay kits (R&D Systems, Minneapolis, MN), which have a detection range of 0.156–10.0 pg/mL. The interassay coefficient for this method is 6.3% at the University of Vermont. CRP was measured with a BNII nephelometer utilizing a particle-enhanced immunonephelometric assay (Dade Behring, Deerfield, IL). The detection range for this assay is 0.15–1100 mg/L, and the routine interassay coefficient of variation is 5% at the University of Vermont.

In AHAB-II, blood samples were drawn during Visit 1, between the hours of 8:00 am and 10:00 am, for the measurement of circulating concentrations of CRP and IL-6. On this occasion, participants were asked to fast for 8 hours, to avoid exercise for 12 hours, and to avoid alcohol for 24 hours prior to coming to the laboratory. High-sensitivity CRP was measured by the University of Vermont using the same Methods as in the PHHP study. Plasma IL-6 concentrations were determined by the University of Pittsburgh’s Behavioral Immunology Laboratory using the same ELISA kits used in the PHHP study (intra and interassay coefficient of variation < 10%). Both studies excluded participants from the blood draw if diagnosed with autoimmune connective tissue disorders (e.g. rheumatoid arthritis), HIV/AIDS, chronic hepatitis, chronic asthma (especially when using medication ≥ 7 times in 14 days prior to blood draw), if using oral glucocorticoid medication, if experiencing acute viral or bacterial infection, and if they had experienced cold or flu in the past 2 weeks. Participants were asked to refrain from non-steroidal anti-inflammatory medication for 24 hours (e.g. Ibuprofen, aspirin) prior to their visit.

Consistent with recent recommendations (Pearson et al., 2003), individuals with CRP concentrations ≥ 10 mg/L were excluded from analyses, due to assumption of recent acute infection. In both studies, CRP and IL-6 values were log transformed to reduce skewness and the log-transformed values were used for all analyses.

2.3.4. Covariates.

Standard covariates were assessed in both studies, and include: age (in years); sex (0=male, 1=female); race (0=white, 1=non-white); education level (1=high school or less, 2= technical school or some college/Associate’s, 3=Bachelor’s degree, 4=Master’s degree or higher; BMI (weight/height²); smoking status (0=nonsmoker, 1=current smoker), daily alcohol intake (using the quantity-frequency method previously reported in Garg et al., 1993), and history of chronic medical conditions described above.

Regression models that did not adjust for any covariates test the simple association of social interaction characteristics with IL-6 and CRP. Subsequently, two sets of multiple regression models (SAS PROC GLM) were used to test the association of these social interaction characteristics, in addition to global measures of SS and SI, with IL-6 and CRP. The first multivariate regression model adjusts for demographic variables, such as age, sex, race, and education, along with BMI (“partially adjusted” model). The second model additionally adjusts for alcohol intake and smoking status (“fully adjusted” model).

3. Results

In PHHP, the total sample on which analyses were conducted was N=306 for measures of social interactions and global measures of SS and SI, and N=229 for marital interactions in individuals who were married or living together with a significant other. IL-6 values ranged from .44 pg/mL to 12 pg/mL, while CRP values ranged from .15 mg/L to 9.69 mg/L. In AHAB-II, the total sample on which analyses were conducted was N=419 for measures of social interactions and global measures of SS and SI, and N=282 for marital interactions in individuals who were married or had been living with a romantic partner. IL-6 values ranged between 0.063 and 9.832 pg/mL. CRP values were between 0.15 and 9.90 mg/L. See Table 1 for Select Sample Characteristics for PHHP and AHAB-II.

Table 1.

Select Sample Characteristics- Demographic and Clinical Characteristics of the Analytic Sample (PHHP: N=306; AHAB-II: N=419)

Characteristic	Mean (SD) or % (n) PHHP	Mean (SD) or % (n) AHAB-II
% male (n)	50.9 (156)	47.02 (197)
% African American (n)	14.3 (44)	16.9 (71)
% bachelor’s degree or higher (n)	50.3 (154)	71.8 (301)
% current smokers (n)	6.0 (19)	16.0 (67)
Mean age (SD)	60.66 (4.79)	42.90 (7.29)
Mean BMI (SD)	27.54 (4.38)	26.77 (5.03)
Median CRP (Variance) mg/L	1.57 (4.22)	.77 (3.58)
Median IL-6 (Variance) pg/mL	1.46 (2.26)	.86 (.94)

In PHHP, demographic and biological factors, along with health behaviors were positively correlated with CRP (i.e. being female and CRP: r= .18, p=.001; BMI and CRP: r= .31, p<.0001; current smoking status and CRP: r= .21, p=0002), whereas level of education was inversely associated with CRP level (r= −.20, p=.0003). BMI and current smoking status were also positive associated with IL-6 level (BMI and IL-6: r= .30, p<.0001; current smoking status: r= .18, p= .002). CRP level and IL-6 concentrations were significantly correlated with each other (r=.36, p<.0001). Individuals spent an average of 61% of the monitoring period in social interactions (range: 3.8% to 100%), an average of 56% of the time in positive interactions (range: 0% to 97%), and an average of 3% of the time in negative interactions (range: 0% to 79%). It is notable that approximately 25% of the sample reported spending no time in negative social interactions.

In AHAB-II, the analytic sample consisted of 419 individuals with complete data for all predictors (i.e. social interaction characteristics and global measures of SS and SI), inflammatory biomarkers, and covariates. Demographic factors and health behaviors were positively correlated with IL-6 and CRP (age and IL-6: r= .15, p=.002; African American race and IL-6: r= .15, p=.002; African American race and CRP: r= .14, p=.004; Non-African American minority race and CRP: r= .11, p= .02; BMI and IL-6: r=.25, p<.0001; BMI and CRP: r= .31, p<.0001; smoking status and IL-6: r= .12, p=.02; smoking status and CRP: r= .16, p=.001). Education was inversely correlated with IL-6 (r= −.15, p=.002) and CRP (r= −.17, p=.0007). CRP and IL-6 were also positively correlated with each other (r= .34, p<.0001). AHAB-II participants spent an average of 66% of the monitoring period in social interactions (range: 13% to 100%), an average of 63% of the time in positive interactions (range: 13% to 100%), and an average of 4% of the time in negative interactions (range: 0% to 36%).

3.1. Social Support, Social Integration, and Inflammation.

In PHHP, neither SS nor SI was associated with either inflammatory marker in the partially (all bs < .024, all ps > .38) or fully adjusted regression models (all bs < .033, all ps > .22). Further analyses were conducted to test the association of individual subscales of SS with each inflammatory marker. These subscales include measures of appraisal support, tangible support, and belonging support. In PHHP, none of the subdimensions of support were associated with IL-6 (all bs < .012, all ps > .07) or CRP (all bs < .0053, all ps > .58) in either model. See Supplementary Table 1.

Similarly, in AHAB-II, neither SS nor SI was associated with either inflammatory marker in the partially adjusted (all bs < .0063, all ps > .31) or fully adjusted models (all bs < .0064, all ps > .32). Additional analyses of subdimensions of the support scale reported no association of appraisal, tangible, and belonging support with IL-6 (all bs < .−.020, all ps > .19) or CRP (all bs< −.017, all ps > .49) in either model. See Supplementary Table 1.

3.2. Daily Social Interactions, and Inflammation.

First, the results in PHHP are presented. Results from unadjusted analyses examining the simple association of social interactions characteristics with inflammatory markers are presented in Table 2. These unadjusted analyses show that in the PHHP sample, relative frequency of positive interactions with coworkers is associated with lower IL-6 (b = −.33, F(1, 304)=, p= .04), whereas there is no association between other measures of social interaction characteristics and either marker of inflammation.

Table 2.

Daily Social Interactions and Inflammation in PHHP and AHAB-II- Standardized Regression Coefficients from Unadjusted Analyses in Regression Models Predicting log CRP and log IL-6 from EMA-assessed social interaction characteristics in middle-aged (N=419) and older adults (N=306)

Pittsburgh Healthy Heart Project – PHHP (N=306)
Variable	Log CRP		Log IL-6
	B	F	B	F
Frequency of social interactions	−.023	.16	−.090	2.49
Frequency of positive interactions	.051	.81	−.11	3.40
Frequency of positive interactions with close others	.053	.85	−.11	3.46
Frequency of negative interactions	.089	2.40	.0084	.02
Frequency of negative interactions with close others	.088	2.39	.0077	.02
Frequency of positive marital interactions (N=229)	.035	.28	−.015	.05
Frequency of negative marital interactions (N=229)	.043	.42	.026	.15
Frequency of positive interactions with coworkers	.059	1.05	−.12	4.36*
Frequency of negative interactions with coworkers	.084	2.18	.0014	0.00
Adult Health and Behavior – II (N=419)
Variable	Log CRP		Log IL-6
	B	F	B	F
Frequency of social interactions	.058	1.42	.040	.67
Frequency of positive interactions	.089	3.34	.035	.50
Frequency of positive interactions with close others	.029	.35	−.11	5.13*
Frequency of negative interactions	−.057	1.38	−.0066	.02
Frequency of negative interactions with close others	−.042	.73	−.034	.48
Frequency of positive marital interactions (N=282)	−.037	.38	−.102	2.92
Frequency of negative marital interactions (N=282)	−.096	2.62	−.037	.39
Frequency of positive interactions with coworkers	.045	.85	.12	2.39*
Frequency of negative interactions with coworkers	−.056	1.31	.037	.57

Note

^*p<.05,

^**p<.001,

^***p<.001

Results from multivariate regression models show that there are no significant associations between the relative frequency of daily social interactions and circulating concentrations of IL-6 (all bs < .29, ps > .09) or CRP (all bs < .057, ps > .86) in either the partially adjusted or fully adjusted model. Similarly, there was no significant association between the relative frequency of negative interactions and either marker of inflammation (IL6: bs < −.11, ps > .79; CRP: bs < 1.17, ps > .11). However, the relative frequency of positive social interactions in daily life was inversely associated with IL-6 concentrations (b= −.33, F(1,299) = 4.17, p=.042). This inverse association remained significant, above and beyond the effects of alcohol intake and smoking (b= −.32, F(1,297) = 4.07, p=.044). In contrast to the findings for IL-6, no significant associations of CRP with the relative frequency of positive interactions were observed (bs < .19 , ps > .53). See Table 3.

Table 3.

Daily Social Interactions and Inflammation in PHHP – Standardized Regression Coefficients from Regression Models Predicting log CRP and log IL-6 from EMA-assessed social interaction characteristics in older adults (N=306)

	Log CRP				Log IL-6
	Model 1		Model 2		Model 1		Model 2
Variable	B	F	B	F	B	F	B	F
Age	−.0097	.03	−.0080	.02	.054	.90	.055	.98
Sex	.19	11.72**	.21	13.99**	.024	.18	.047	.66
Black Race	.012	.05	−.014	.07	.060	1.16	.039	.50
Education (highest degree)	−.12	4.51*	−.098	3.23	−.054	.88	−.037	.43
BMI	.31	32.93***	.34	39.99***	.30	28.85***	.33	34.36***
Smoking	--	--	.22	16.63***	--	--	.18	10.64*
Alcohol Intake	--	--	.057	1.11	--	--	.075	1.81
Frequency of Interactions	−.0097	.03	−.0041	.01	−.094	2.83	−.091	2.73
	R2 = .157		R2 = .208		R2 = .111		R2 = .150
	Log CRP				Log IL-6
	Model 1		Model 2		Model 1		Model 2
Variable	B	F	B	F	B	F	B	F
Age	−.003	.00	−.002	.00	.056	.96	.057	1.04
Sex	.19	11.70**	.21	13.99**	.020	.12	.042	.55
Black Race	.014	.06	−.013	.06	.056	.99	.035	.39
Education (highest degree)	−.12	4.29*	−.095	3.04	−.053	.86	−.036	.41
BMI	.31	32.91***	.34	39.94***	.30	29.62***	.33	35.17***
Smoking	--	--	.22	16.88	--	--	.18	10.74*
Alcohol Intake	--	--	.056	1.06	--	--	.074	1.74
Frequency of positive interactions	.038	.27	.033	.40	−.11	4.17*	−.11	4.07*
	R2 =.158		R2 = .209		R2 = .115		R2 = .154
	Log CRP				Log IL-6
	Model 1		Model 2		Model 1		Model 2
Variable	B	F	B	F	B	F	B	F
Age	−.0026	0.00	−.0019	.0.00	.055	.95	.056	1.03
Sex	.19	11.68**	.21	13.96**	.020	.13	.043	.56
Black Race	.014	.06	−.013	.06	.056	1.00	.035	.39
Education (highest degree)	−.12	4.28*	−.095	3.03	−.053	.86	−.036	.41
BMI	.31	32.93***	.39	39.97***	.30	29.55***	.33	35.08***
Smoking	--	--	.22	16.89	--	--	.18	10.73**
Alcohol Intake	--	--	.056	1.06	--	--	.07	1.74
Frequency of positive interactions with close others	.030	.30	.035	.44	−.11	4.16*	−.11	4.06*
	R2 = .158		R2 = .209		R2 = .115		R2=.154
	Log CRP				Log IL-6
	Model 1		Model 2		Model 1		Model 2
Variable	B	F	B	F	B	F	B	F
Age	.0061	.01	.0026	0.00	.073	1.66	.071	1.60
Sex	.19	12.32**	.21	14.51**	.020	.12	.041	.49
Black Race	.0065	.01	−.018	.11	.061	1.16	.040	.51
Education (highest degree)	−.11	4.31*	−.096	3.17	−.043	.58	−.027	.23
BMI	.31	33.33***	.34	40.14***	.30	28.91***	.33	34.36***
Smoking	--	--	.21	15.44***	--	--	.18	11.13***
Alcohol Intake	--	--	.058	1.17	--	--	.069	1.52
Frequency of Negative Interactions	.086	2.54	.063	1.40	.0058	.01	−.015	.07
	R2 = .164		R2 = .211		R2 = .103		R2 = .143
	Log CRP				Log IL-6
	Model 1		Model 2		Model 1		Model 2
Variable	B	F	B	F	B	F	B	F
Age	.0062	.01	.0026	.00	.073	1.66	.071	1.60
Sex	.19	12.33**	.21	14.53**	.020	.12	.041	.49
Black Race	.0065	.01	−.018	.11	.061	.16	.040	.51
Education (highest degree)	−.11	4.31*	−.096	3.16	−.043	.58	−.027	.23
BMI	.31	33.36***	.34	40.17***	.30	28.91***	.33	34.36***
Smoking	--	--	.21	15.43***	--	--	.18	11.13***
Alcohol Intake	--	--	.059	1.18	--	--	.069	1.52
Frequency of negative interactions with close others	.086	2.55	.063	1.42	.0054	.01	−.015	.07
	R2 = .164		R2 = .211		R2 = .103		R2 = .143

Model 1 contains age, sex, race, education, and BMI; Model 2 additionally controls for smoking status and alcohol intake.

Note:

^*<.05,

^**<.001,

^***<.0001

To test whether this omnibus, inverse association between total positive interactions and IL-6 was accounted for by specific social roles (i.e. interaction partner), the nature of interactions with close others (i.e. spouse, friends, and family), and with coworkers was tested in association with inflammation. Relative frequency of positive interactions with close others associated with lower IL-6, in the first model (b= −.37, F(1, 299) = 4.16, p= .042), and after additional adjustment for alcohol intake and smoking (b= −.33, F(1,297)= 4.06, p=.0449). See Table 3. Similarly, the relative frequency of positive interactions with coworkers significantly associated with lower IL-6 in the first model (b= −.39, F(1,299)= 6.33, p=.01), and after adjustment for health behaviors (b= −.40, F(1,297)= 6.92, p=.01). See Table 5. These results illustrate that the omnibus, inverse association between relative frequency of total positive interactions and concentrations of IL-6 is robust in nature, and not particularly driven by specific social roles, in older adults, after adjusting for demographic and health behavior covariates.¹

Table 5.

Relational Domains and Inflammation in PHHP - Standardized Regression Coefficients in Regression Models Predicting log CRP and log IL-6 from EMA-assessed interactions with spouse (N= 229) and coworkers in older adults (N=306)

	Log CRP N= 229				Log IL-6 N= 229
	Model 1		Model 2		Model 1		Model 2
Variable	B	F	B	F	B	F	B	F
Age	.0082	.02	.0082	.02	.12	2.94	.12	3.02
Sex	.22	10.90*	.24	13.70**	−.011	.03	.0040	.00
Black Race	−.057	0.87	−.051	.70	.019	.09	.019	.09
Education (highest degree)	−.15	5.58*	−.14	4.53*	−.0017	.00	.016	.05
BMI	.32	24.82***	.35	30.29***	.36	29.68***	.39	33.87**
Smoking	--	--	.20	11.34**	--	--	.17	7.42*
Alcohol Intake	--	--	.059	.92	--	--	.019	.09
Frequency of positive marital interactions	.039	.36	.045	.50	.059	.77	.063	.90
	R2 = .185		R2 = .229		R2 = .131		R2 =.159
	Log CRP N= 229				Log IL-6 N= 229
	Model 1		Model 2		Model 1		Model 2
Variable	B	F	B	F	B	F	B	F
Age	.0018	.00	.0013	.00	.10	2.45	.10	2.50
Sex	.22	11.47**	.25	14.47**	−.0046	.00	.011	.03
Black Race	−.060	.98	−.054	.82	.015	.05	.015	.06
Education (highest degree)	−.15	5.64*	−.14	4.60*	−.002	.00	.015	.05
BMI	.32	24.42***	.34	29.80***	.36	28.84***	.38	32.92***
Smoking	--	--	.20	11.45**	--	--	.17	7.42*
Alcohol Intake	--	--	.061	.98	--	--	.019	.09
Frequency of negative marital interaction	.044	.51	.054	.81	.033	.27	.039	.39
	R2 =.185		R2 = .230		R2 = .129		R2 = .158
	Log CRP N= 306				Log IL-6 N= 306
	Model 1		Model 2		Model 1		Model 2
Variable	B	F	B	F	B	F	B	F
Age	−.012	.05	−.012	.05	.091	2.62	.092	2.80
Sex	.19	11.49**	.21	13.72**	.026	.21	.050	.77
Black Race	.015	.07	−.012	.05	.052	.88	.030	.30
Education (highest degree)	−.11	4.24*	−.095	3.04	−.056	.95	−.039	.49
BMI	.31	32.79***	.34	39.76***	.31	30.26***	.33	36.14***
Smoking	--	--	.22	16.76***	--	--	.18*	11.17**
Alcohol Intake	--	--	.055	1.03	--	--	.078	1.95
Frequency of positive interactions with coworkers	.035	.41	.032	.38	−.14*	6.33*	−.14*	6.92*
	R2 = .158		R2 = .209		R2 = .121		R2 =.162
	Log CRP N= 306				Log IL-6 N= 306
	Model 1		Model 2		Model 1		Model 2
Variable	B	F	B	F	B	F	B	F
Age	−.00096	0.00	−.0032	0.00	.072	1.61	.070	1.60
Sex	.19	11.97**	.21	14.17**	.019	0.12	.041	.50
Black Race	.0046	.01	−.018	.12	.062	1.20	.042	.57
Education (highest degree)	−.12	4.36*	−.097	3.20	−.044	.59	-.027	.23
BMI	.31	33.08***	.34	39.91***	.30	28.92***	.33	34.50***
Smoking	--	--	.21	15.63***	--	--	.19	11.41**
Alcohol Intake	--	--	.057	1.11	--	--	.070	1.54
Frequency of negative interactions with coworkers	.073	1.86	.047	.79	−.0094	.03	−.033	.37
	R2 = .162		R2 = .210		R2 = .103		R2 =.144

Analyses were further conducted to test the association between the relative frequency and quality of marital interactions, defined as dyadic interactions that occurred exclusively with the spouse, and inflammatory markers. In a subsample of married individuals (N=229 individuals who were either legally married or living with romantic partner and had complete data for marital interactions, inflammatory outcomes, and covariates), results showed that the relative frequency of marital interactions was not associated with concentrations of IL-6 or CRP (bs < .29, ps > .22), nor was the relative frequency of positive (bs < .23, ps > .55) or negative marital interactions (bs < 1.63, ps > .37). See Table 5.

In regards to the AHAB-II sample, unadjusted analyses examining the simple association of social interaction characteristics with inflammatory biomarkers are presented in Table 2. These results show that the relative frequency of positive interactions with close others is associated with lower IL-6 (b = −.50, F(1, 417)= 5.13, p= .02), but the relative frequency of positive interactions with coworkers is associated with higher IL-6 (b= .68, F(1,417)= 5.69, p= .02). Multivariate regression analyses show that the relative frequency of total daily social interactions was not associated with either IL-6 or CRP in either model (bs < .023, ps > .90.; bs < .15, ps > .62, respectively). In contrast to PHHP, the relative frequency of total positive interactions was not associated with IL-6 or CRP in either model (all bs < −.024, ps > .90 and all bs < .37, ps > .24, respectively), nor was the relative frequency of negative social interactions (bs < .−.084, ps > .89 and all bs < −1.42, ps > .15, respectively). See Table 4.

Table 4.

Daily Social Interactions and Inflammation in AHAB- II - Standardized Regression Coefficients from Regression Models Predicting log CRP and log IL-6 from EMA-assessed social interaction characteristics in middle-aged adults (N=419)

	Log CRP				Log IL-6
	Model 1		Model 2		Model 1		Model 2
Variable	B	F	B	F	B	F	B	F
Age	.081	3.11	.076	2.78	.14	9.09*	.14	8.66*
Sex	.063	1.87	.065	2.00	.19	.17	.020	.19
Black Race	.047	0.95	.047	.97	.040	.69	.040	.69
Education (highest degree)	−.062	1.63	−.043	.75	−.094	3.71	−.081	2.64
BMI	.37	61.47***	.37	59.80***	.33	46.86***	.32	45.51***
Smoking	--	--	.080	2.93	--	--	.052	1.21
Alcohol Intake	--	--	−.051	1.30	--	--	−.029	.40
Frequency of Interactions	.018	.17	.023	.25	.0034	.01	.0058	.02
	R2 = .186		R2 = .194		R2 = .175		R2 =.178
	Log CRP				Log IL-6
	Model 1		Model 2		Model 1		Model 2
Variable	B	F	B	F	B	F	B	F
Age	.082	3.24	.078	2.90	−.14	9.03*	.14	8.61*
Sex	.061	1.74	.063	1.86	.020	.18	.021	.21
Black Race	.046	.91	.046	.92	.041	.70	.041	.70
Education (highest degree)	−.057	1.41	−.037	.57	−.095	3.79	−.082	2.69
BMI	.37	61.28***	.36	59.56***	.33	46.98***	.32	45.61***
Smoking	--	--	.083	3.13	--	--	.052	1.19
Alcohol Intake	--	--	−.052	1.36	--	--	−.028	.39
Frequency of positive interactions	.047	1.09	.054	1.40	−.006	.02	−.002	.00
	R2 = .188		R2 = .196		R2 = .175		R2 = .178
	Log CRP				Log IL-6
	Model 1		Model 2		Model 1		Model 2
Variable	B	F	B	F	B	F	B	F
Age	.085	3.33	.082	3.13	.12	6.89*	.12	6.61*
Sex	.063	1.86	.065	1.98	.026	.31	.027	.34
Black Race	.049	1.02	.050	1.06	.038	.62	.038	.62
Education (highest degree)	−.061	1.62	−.041	.70	−.10	4.55*	−.092	3.42
BMI	.37	61.64***	.37	60.02***	.33	48.03***	.32	46.71***
Smoking	--	--	.082	3.11	--	--	.044	.89
Alcohol Intake	--	--	−.056	1.56	--	--	−.014	.09
Frequency of positive interactions with close others	.03	.45	.043	.89	−.10	5.21*	−.099	4.64*
	R2 = .187		R2 = .195		R2 = .185		R2 =.187
	Log CRP				Log IL-6
	Model 1		Model 2		Model 1		Model 2
Variable	B	F	B	F	B	F	B	F
Age	.075	2.64	.068	2.22	.14	9.03*	.14	8.48*
Sex	.066	2.03	.069	2.22	.019	0.17	.021	.21
Black Race	.053	1.19	.054	1.25	.041	0.69	.041	.72
Education (highest degree)	−.061	1.63	−.040	.67	−.094	3.78	−.082	2.67
BMI	.37	62.26***	.367	60.51***	.33	46.97***	.32	45.66***
Smoking	--	--	.088	3.50	--	--	.053	1.22
Alcohol Intake	--	--	-.050	1.25	--	--	−.028	.39
Frequency of negative interactions	−.055	1.51	−.065	2.09	−.00046	.00	−.0065	.02
	R2 = .189		R2 = .197		R2 = .175		R2 =.178
	Log CRP				Log IL-6
	Model 1		Model 2		Model 1		Model 2
Variable	B	F	B	F	B	F	B	F
Age	.075	2.68	.069	2.29	.14	8.65*	.13	8.11*
Sex	.068	2.15	.071	2.36	.021	.21	.023	.25
Black Race	.053	1.21	.054	1.26	.044	.82	.045	.84
Education (Highest Degree)	−.059	1.48	−.038	.60	−.090	3.47	−.077	2.37
BMI	.37	62.61***	.37	61.01***	.33	47.39***	.32	46.08***
Smoking	--	--	.085	3.28	--	--	-.028	1.37
Alcohol Intake	--	--	-.05	1.26	--	--	.055	.39
Frequency of negative interactions with close others	−.049	1.19	−.056	1.56	−.034	.55	−.039	.71
	R2 = .188		R2 = .196		R2 = .176		R2 =.179

Model 1 contains age, sex, race, education, and BMI; Model 2 additionally controls for smoking status and alcohol intake.

Note :

^*<.05,

^**<.001,

^***<.0001

We explored the possibility that positive interactions in the context of specific relationships may relate to inflammation in this younger sample in multivariate regression models. There was no association of frequency of positive interactions with close others and CRP in either model (all bs < .33, ps > .35) but relative frequency of positive interactions with close others was inversely associated with IL-6 in the partially adjusted model (b= −.47, F(1,411)= 5.21, p= .02), and after adjusting for alcohol intake and smoking (b= −.45, F(1,409)= 4.64, p = .03). See Table 4. Notably, this finding replicates that reported from the PHHP sample. The relative frequency of negative interactions with close others did not predict either IL-6 or CRP (bs < −.63, ps > .40; bs < −1.54, ps > .21, respectively) in either model.

In contrast with the PHHP sample, the relative frequency of positive interactions with coworkers in the AHAB-II sample was not associated with IL-6 or CRP in either multivariate model (bs =.48, ps > .07, and bs < .18, ps > .69, respectively). Similarly, the relative frequency of negative interactions with coworkers was also not associated with IL-6 or CRP in either model (bs < 1.19, ps > .35, bs < −2.42, ps > .25. See Table 6.

Table 6.

Relational Domains and Inflammation in AHAB-II - Standardized Regression Coefficients in Regression Models Predicting log CRP and log IL-6 from EMA-assessed social interaction with spouse (N=282) and coworkers in middle-aged adults (N=419)

	Log CRP N= 282				Log IL-6 N= 282
	Model 1		Model 2		Model 1		Model 2
Variable	B	F	B	F	B	F	B	F
Age	.089	2.57	.090	2.58	.16	7.76*	.15	7.61*
Sex	.072	1.58	.073	1.61	.06	1.16	.064	1.26
Black Race	−.022	0.14	−.021	.13	−.094	2.63	−.095	2.67
Education (highest degree)	−.042	0.51	−.028	.22	−.12	3.86	−.11	3.28
BMI	.42	50.84***	.41	47.17***	.36	37.12***	.36	35.25***
Smoking	--	--	.075	1.72	--	--	.033	0.32
Alcohol Intake	--	--	−.022	0.16	--	--	.035	0.39
Frequency of positive marital interactions	−.03	.30	−.028	.25	−.089	2.56	−.088	2.49
	R2 = .203		R2 = .208		R2 = .203		R2 =.206
	Log CRP N= 282				Log IL-6 N= 282
	Model 1		Model 2		Model 1		Model 2
Variable	B	F	B	F	B	F	B	F
Age	.086	2.43	.085	2.41	−.17	8.96*	.17	8.81*
Sex	.081	2.06	.081	2.06	.074	1.67	.076	1.76
Black Race	−.015	.06	−.014	.06	−.084	2.10	−.085	2.15
Education (highest degree)	−.027	.22	−.012	.04	−.10	3.36	−.10	2.81
BMI	.43	53.58***	.42	49.86***	.36	36.21***	.35	34.28***
Smoking	--	--	.081	2.04	--	--	.037	.42
Alcohol Intake	--	--	−.034	.38	--	--	.031	.30
Frequency of negative marital interaction	−.11	4.38*	−.12	4.72*	−.035	.41	−.033	.36
	R2 =.214		R2 = .220		R2 = .197		R2 = .199
	Log CRP N= 419				Log IL-6 N= 419
	Model 1		Model 2		Model 1		Model 2
Variable	B	F	B	F	B	F	B	F
Age	.080	3.04	.075	2.69	.14	9.29*	.14	8.80*
Sex	.066	2.02	.068	2.18	.023	.24	.025	.28
Black Race	.047	0.96	.048	.98	.038	.61	.038	.61
Education (highest degree)	−.062	1.62	−.044	.77	−.083	2.94	−.071	2.02
BMI	.37	61.74***	.36	60.08***	.32	47.02***	.32	45.53***
Smoking	--	--	.079	2.89	--	--	.053	1.26
Alcohol Intake	--	--	−.049	1.17	--	--	−.020	.20
Frequency of positive interactions with coworkers	.018	.16	.014	.10	.082	3.28	.081	3.16
	R2 = .186		R2 = .193		R2 = .181		R2 =.184
	Log CRP N= 419				Log IL-6 N= 419
	Model 1		Model 2		Model 1		Model 2
Variable	B	F	B	F	B	F	B	F
Age	.078	2.91	.073	2.54	.14	9.31*	.14	8.85*
Sex	.061	1.75	.064	1.90	.023	0.24	.024	.27
Black Race	.048	1.00	.048	1.02	.040	0.69	.041	.70
Education (highest degree)	−.067	1.94	−.048	.95	−.091	3.56	−.080	2.61
BMI	.37	61.53***	.37	59.73***	.33	47.36***	.32	46.08***
Smoking	--	--	.083	3.17	--	--	.049	1.06
Alcohol Intake	--	--	−.049	1.20	--	--	−.029	.42
Frequency of negative interactions with coworkers	−.046	1.07	−.051	1.30	.042	.86	.039	.74
	R2 = .188		R2 = .196		R2 = .177		R2 =.179

However, characteristics of marital interactions were significantly associated with CRP level in this younger sample in an unexpected direction. Specifically, greater relative frequency of negative marital interactions predicted lower CRP in the partially adjusted (b = −5.76, F(1,274)= 4.38, p = .04) and fully adjusted model (b = −5.99, F(1,272) = 4.72, p = .03). This finding differs from others reported previously in 3 ways: 1) it is in association with CRP level, instead of IL-6, 2) it shows an unexpected inverse association, and 3) it concerns the frequency of negative interactions, rather than positive interactions. In fact, relative frequency of positive marital interactions did not associate with IL-6 or CRP in either model (all bs < −.51, ps > .11). Relative frequency of negative marital interactions was not associated with IL-6 level in either model (all bs < −1.01, ps > .52). See Table 6. The total frequency of marital interactions, overall, was also not associated with IL-6 or CRP in either model (all bs < −.42, ps > .16).

4. Discussion

We examined the association of social relationships and daily social interaction characteristics with circulating levels of inflammation in two samples: middle-aged and older adults. We extended prior literature by examining social interactions as measured in real-time EMA methods and distinguishing among specific interpersonal domains, such as interactions with spouse, friends, and family.

In middle-aged and older adults, we found global measures of social support and integration to be unrelated to inflammatory markers, IL-6 and CRP. Although some studies have reported an association between certain social metrics, particularly social integration, and inflammation (Ford et al., 2006; Loucks et al., 2006), many have failed to relate similarly assessed constructs to inflammation (McDade et al., 2006; Glei et al., 2012). The present results do not rule out associations of other self-report measures of social support and social integration with inflammation. Additionally, it is plausible that global measures of social relationships may moderate the effect of stressors on inflammation as indicated by previous work (Mezuk et al., 2010).

We speculate that inconsistent findings in this literature may be either be due to chance, or, in part, to limitations associated with global, self-report measures and their inability to consistently capture representative samples of social interaction behaviors. As mentioned previously, global measures of support and integration may reflect appraisal of conceptualized beliefs regarding one’s social network, whereas EMA measures may more accurately reflect momentary triggers and characteristics of social interactions by providing time-sampled observations of individuals’ typical social lives, a feature which may yield a more representative sample of event-specific triggers of social interaction behaviors (Conner & Barrett, 2012).

The literature thus far also shows limited exploration of the function of different social roles in explaining the inconsistent findings between global measures of social relationships and inflammation. Therefore, we examined the association of EMA measures of the proportion of daily life spent in total, positive or negative interactions with inflammation. Across the two cohorts, we found that characteristics of daily social interactions were significantly associated with IL-6. In both samples, greater proportion of positive interactions with close others in daily life predicted lower levels of IL-6, above and beyond the effects of demographic characteristics and biobehavioral covariates. In the middle-aged sample, the association of social interaction characteristics with various interaction partners were differentially associated with inflammation, such that frequency of positive interactions with close others was inversely associated with IL-6 level, whereas none of the interactions characteristics with coworkers were associated with markers of inflammation.

We found no associations between the proportion of total negative interactions and inflammatory markers, possibly due to the extreme infrequency of such interactions in both samples (i.e. in both samples, participants spent about 3% of the monitoring period engaged in negative interactions). However, these null findings are in contrast with previous reports of an association between social strain and inflammation (Chiang et al., 2012; Fuligni et al., 2009). In Chiang et al., 2012, there was a positive association between competitive interactions and inflammation and Fuligni and colleagues (2009) reported a positive association between interpersonal stress and inflammation in adolescents. In contrast to the measures used here, it may be that more subtle measures of negative interactions have greater variability across assessments and may provide more meaningful distinctions among individuals, rendering associations with outcomes more visible. This association may also be more consistent in adolescents.

We found no association between marital interaction characteristics and circulating levels of inflammatory biomarkers in the PHHP sample. This may be, in part, due to reduced power, given the smaller sample size of the married subsample. Previous studies that report an association of partner support and strain with inflammation have considerably larger sample sizes than those used here (N=542 in Donoho et al., 2013 and N= 415 in Whisman & Sbarra, 2012). However, in the AHAB-II sample, there was an unexpected, inverse association between negative marital interactions and CRP level. This finding should be interpreted with caution, given that it was not predicted and did not replicate across samples. The limited variability of negative marital interactions in this married subsample (i.e. 63.8% of individuals in the subsample reported no negative marital interactions during the monitoring period) further reduces confidence in this effect. Therefore, more work needs to be done to elucidate the relationship between the quality of marital interactions in daily life and inflammation.

With regard to the positive interaction findings, the differences in findings reported in PHHP and AHAB-II may be partly due to differences in age and differences in the positive interaction measures used between the two samples. The additional 2 days of monitoring (i.e. 6 days in PHHP vs. 4 days in AHAB-II) and the greater frequency of interviews (i.e. every 45 min in PHHP vs. every hour in AHAB-II) resulted in greater number of observations of positive interactions in PHHP compared to AHAB-II. Further, there was an additional item assessing positive interactions in PHHP, which was not present in AHAB-II (i.e. “friendly” interactions), which may have contributed to greater measurement sensitivity in PHHP. In addition, it may be that the effect of frequent positive social interactions overall may be more robust in older adults, such that the effects were not limited to a social role. The larger presence of close social bonds in daily life and frequent positive interactions with these partners may be particularly health protective in older adults.

In both samples, the relative frequency of positive interactions was associated with IL-6, with such effects emerging only among close others in the middle-aged sample. The association between frequency of positive interactions and inflammation is consistent with the growing literature supporting an association of positive affect and daily positive events with circulating inflammatory markers (Stellar et al., 2015; Sin et al., 2015). For example, Sin and colleagues (2015) reported that positive interpersonal events were associated with lower IL-6 and fibrinogen in women. This finding emphasizes the value of positive events in daily life, and especially positive interpersonal events, in association with biological markers of health.

The quality of daily interactions may contribute to systemic inflammation through a variety of pathways, including attenuation of stress responses to acute stress and promotion of healthy behaviors. Positive relationship characteristics, such as positive interactions, may allow an individual to perceive implicit support and interpersonal resources during times of stress. In an acute stress paradigm, it has been shown that individuals exposed to the friendly version of the Trier Social Stress Test (TSST), characterized by friendly and warm behavior of an evaluative committee, showed reduced activation of the hypothalamic-pituitary-adrenal (HPA) axis compared to individuals who faced social evaluative threat in the traditional TSST condition (Wiemers et al., 2013). Under conditions of chronic stress, the over-activation of the HPA axis is often thought to promote elevated inflammation, perhaps due to glucocorticoid receptor resistance (Miller et al., 2002; Cohen et al., 2012). In addition, social evaluative threat has been associated with autonomic arousal, which, in turn, is implicated in the modulation of systemic inflammation (Bosch et al., 2009; Tracey, 2002).

In addition to their effects on psychophysiological responding, social interactions may be linked with health behaviors. When compared with their socially isolated counterparts, socially integrated individuals have been shown to abstain from risky health behaviors (Cohen et al., 2004), such as smoking and excessive alcohol intake, which, in turn, are associated with elevated systemic inflammation (Imhof et al., 2001; Shiels et al., 2014).

5. Limitations and Conclusions

This study does not come without limitations. First, both studies drew from cross-sectional data, and therefore, we are limited in our ability to infer causality. Secondly, participants were subjected to a wide variety of exclusionary criteria, which resulted in two very healthy samples. This may limit generalizeability of these results. For example, the null findings reported in the two samples here may be partly due to the enforcement of strict exclusionary criteria pertaining to chronic health conditions and medication use (e.g. see section 2.1), whereas other studies do not routinely exclude these participants. Thirdly, it is a limitation that inflammatory markers are only assessed at one time point, especially because there is evidence to suggest that there is considerable intra-individual variability when CRP is measured during multiple times (i.e. daily, weekly, monthly, and tri-monthly measurements) with individuals moving from one CRP risk category to another (Bogaty et al., 2013). Nevertheless, even when measured at only one time point, CRP levels have been predictive of negative health outcomes, including future risk of a fatal or nonfatal coronary event (Koenig et al., 1999). Relatedly, in both samples, the time of blood draw preceded monitoring periods of social interaction measures. However, in addition to evidence suggesting stability in measures of biomarkers, there is also evidence reporting substantial test-retest reliability in EMA measures of psychosocial states and social interactions when these measures are aggregated across several days, and repeated several weeks or months later (Kamarck et al., 1998; Janicki et al., 2006). To the extent that these measures are stable over time, the order of assessment should be less critical. Lastly, it is important to note that a number of statistical analyses were run in these analyses and results showed modest effect sizes. Therefore, greater confidence in these results could be achieved by replication of these findings.

There are also notable strengths of this research. First, our use of EMA measures of social interactions in the natural environment may provide a more accurate measure of these social processes as they unfold compared to retrospective global self-report measures. In fact, both samples reported null associations between global measures of social relationships and inflammation, while reporting significant associations between daily social behaviors and inflammation. Second, we examined two samples and partially replicated findings among different age groups and across both genders. And lastly, these samples were large by EMA research standards and the frequency and length of sampling made it possible for us to examine the association of social interactions within specific relationship domains with inflammatory markers. Future research may consider exploring additional mechanistic processes that may account for the association between social constructs and systemic inflammation. These processes may include regulation of daily affect, frequency of engagement in health behaviors, and modulation of biological stress responses, all of which are implicated in chronic, systemic inflammation.

Glossary

EMA

Ecological Momentary Assessment

SS

Social Support

SI

Social Integration

IL-6

Interleukin-6

CRP

C-reactive protein

AHAB-II

Adult Health and Behavior Project Phase II (AHAB=II)

PHHP

Pittsburgh Healthy Heart Project

ISEL

Interpersonal Support Evaluation List

SNI

Social Network Inventory

BMI

Body Mass Index

Appendix A.

EMA Items Regarding Daily Social Interactions for Pittsburgh Healthy Heart Project.

-At time of BP Reading-
“Currently in a social interaction?”	No, Yes
(If yes, skip to the “Think about this most recent interaction…” prompt)
When was your most recent social interaction?	0–10 min before BP, 11–45 min before BP, 45+ min before BP
Think about the most recent interaction…
Length of interaction	Less than 1 min, 1–10 min, 10–20 min 20–45 min, 45+ min
1. Type of Interaction	In person, Telephone, E-mail
2. With how many people?	1 other, 2 others, 3 others, 4 or more
3. Interacting with whom?	Spouse/Partner, Other family or relative(s), Other friend(s), Other coworker(s), Other
(If just “Spouse/Partner”, then skip to “Pleasant interaction?” question. If any other response is chosen, with or without “Spouse/Partner”, then NEXT.)
4. Interacting with a “confidant”?	No, Yes
Pleasant interaction subscale
5. Pleasant interaction?	NO 1 2 3 4 5 6 7 8 9 10 11 YES
6. Agreeable interaction?	NO 1 2 3 4 5 6 7 8 9 10 11 YES
7. Friendly interaction?	NO 1 2 3 4 5 6 7 8 9 10 11 YES
Social conflict subscale
8. Someone treated you badly?	NO 1 2 3 4 5 6 7 8 9 10 11 YES
9. Someone interfered with your efforts?	NO 1 2 3 4 5 6 7 8 9 10 11 YES
10. Someone in conflict with you?	NO 1 2 3 4 5 6 7 8 9 10 11 YES

NOTE: Items 5, 6, and 7 were used to assess positive interactions and items 8 and 10 were used to assess negative interactions.

Appendix B

MA Items Regarding Daily Social Interactions for Adult Health and Behavior.

- At time of BLOOD PRESSURE
“In a social interaction?”	No, Yes
- If yes, skip to “Think about this most recent interaction…” prompt
- At time of BLOOD PRESSURE
“When did your most recent social interaction end?	0–10 min before ALARM, 11–45 min before ALARM, 45+ min before ALARM
- PROMPT SCREEN: Think about this most recent interaction.…
1. Type of interaction?	In person, Telephone, Instant Messaging, Webcam (e.g. Skype)
2. With how many people?	1 other, 2 others, 3 others, 4 or more
3. Interacting with whom?	Spouse/Partner, Co-worker, other friend, Other family or relative(s), Other acquaintances, Stranger
4. Pleasant interaction?	NO! No no yes Yes YES!
5. Agreeable interaction?	NO! No no yes Yes YES!
6. Someone treated you badly?	NO! No no yes Yes YES!
7. Someone in conflict with you?	NO! No no yes Yes YES!
8. I told someone they annoyed me.	NO! No no yes Yes YES!
9. I yelled at someone.	NO! No no yes Yes YES!

Note: Items 4 and 5 used to assess positive interactions and items 6 and 7 are used to assess negative interactions.