Social Integration and Diurnal Cortisol Decline: The Role of Psychosocial and Behavioral Pathways

Abstract

Objective

A growing number of studies have associated various measures of social integration, the diversity of social roles in which one participates, with alterations in hypothalamic-pituitary-adrenocortical (HPA) functioning. The pathways through which social integration may be linked to HPA functioning, however, are as yet unknown. The present study examined whether daily social interactions, affective responses, health behaviors, and personality help explain the association between social integration and diurnal cortisol slope.

Methods

A sample of 456 healthy, employed adults (53.9% female, 82.0% White, 72.2% BA or greater, mean age of 42.86) completed a 4-day ecological momentary assessment protocol that measured cortisol, social interactions, affect, sleep, and physical activity at frequent intervals throughout the day. Social integration was measured at baseline.

Results

Regression results controlling for age, sex, race, and education indicated that more socially integrated individuals showed steeper cortisol slopes (B = −0.00253, p = .006). Exploratory analyses suggested that the consistency (i.e. reduced variability) in nightly sleep midpoint partially explained this association (B = −0.00042, 95% CI: −0.00095, −0.00001). Personality, mood, social interaction patterns, and non-sleep health behavior differences did not account for the association between social integration and HPA activity.

Conclusion

This study replicates previous findings linking social integration and HPA functioning, and it examines patterns of nightly sleep as possible pathways through which the association may operate. Results have implications for understanding mechanisms for health risk and for development of future interventions.

Introduction

An impressive literature links social integration (SI), defined as the diversity of social roles in which one participates, with reduced risk for premature morbidity and mortality(1), but the mechanisms accounting for these effects are not well understood. It is possible that SI and the regular social activity that accompanies it may assist in the regulation of biological systems linked with health maintenance(2), an effect which could plausibly contribute to some of its putative salubrious effects.

The hypothalamic-pituitary-adrenocortical (HPA) axis is one candidate biological system that appears to be linked with health risk, on the one hand, and with social behavior, on the other. The HPA axis is an important component of a number of biological processes involved in disease pathogenesis, including blood pressure regulation, weight distribution, inflammation, gluconeogenesis, and thrombosis and hemostasis(3–7). Circulating cortisol, a major marker of HPA axis function in humans, shows a marked diurnal pattern, peaking during the morning hours and declining throughout the afternoon and evening(8). Accumulating evidence suggests that people with a flatter diurnal cortisol decline may be at increased risk for subclinical and clinical cardiovascular disease. Specifically, previous work has demonstrated that individuals with flatter diurnal cortisol slopes, measured through sampling during waking hours, show higher levels of preclinical atherosclerosis and coronary calcification as well as higher rates of cardiovascular disease mortality (9–11). Cumulative (area-under-the-curve, AUC) measures of cortisol secretion have likewise been associated with preclinical atherosclerosis (12), however results are less consistent (9, 10).

The HPA axis is associated not only with biological processes relevant to health and survival, but with social ties as well. In humans, the relationship between SI and HPA activity has been assessed in at least four previous studies. Three studies related indices of SI to diurnal cortisol assessed over a single day: As part of a 15-year follow-up in the CARDIA cohort(13), a 3-item measure of social role diversity (any close friends, any close relatives, any group memberships) was associated with a steeper diurnal cortisol slope, reductions in total diurnal cortisol secretion, and lowest bedtime cortisol values. Likewise, in a study of 2229 older British adults, individuals newly living alone and those recently widowed showed higher nighttime cortisol readings and flatter diurnal slopes, calculated from three samples collected over the course of a 24-hour period, compared to those living with others or currently married, respectively(14). In a subsample of 238 healthy, middle-aged adults from the Whitehall II study cohort, social isolation (3-item scale asking if the participant lives alone and has monthly contact with friends and family) was associated with greater cortisol output over one working day(15). Although social isolation was not associated with diurnal slope in this sample, more than one collection day may be needed to establish reliable salivary diurnal slope estimates(16). Most recently, Chin, Murphy, and Cohen(17) reported an association between SI, as measured through the Social Network Index(18), and cortisol AUC and slope derived from three days of monitoring among older individuals, where more socially integrated individuals showed a steeper diurnal cortisol slope and greater AUC. In sum, the current literature show associations between social integration and a number of HPA parameters, with the most consistent patterns of findings shown, once again, for measures of diurnal cortisol slope. Less is known about possible mechanisms accounting for these effects.

Possible Indirect Effects

Daily social interactions are one factor that may explain associations between SI and diurnal cortisol slope. Conceivably, more socially integrated individuals may have more consistent, more positive, or larger numbers of daily social interactions, all of which have been related to cortisol patterns. Among non-depressed individuals, days that involve more regular social activities are associated with a steeper cortisol decline compared to days involving fewer social activities(19). Furthermore, intimate spousal interactions have been associated with reduced AUC(20), and days with arguments have been linked with elevated AUC(21). Additionally, Stetler and colleagues (2004) showed that, among non-depressed individuals, days with more regular engagement in daily activities were associated with healthier cortisol patterns (22). Number, positivity/negativity, and timing of daily social interactions may account for the hypothesized relationship between SI and diurnal cortisol slope.

Daily mood may also explain links between SI and cortisol. Momentary negative affect has been positively associated with cortisol responses(23). Positive affect has been shown to be inversely associated with AUC(24, 25), and trait positive affect has been linked to steeper cortisol slope(26). Daily experiences of loneliness have also been related to blunted cortisol activity(27). To the extent that SI is linked with affect(24, 28), such effects could plausibly explain associations between SI and diurnal cortisol dynamics.

In addition to daily social and affective experiences, health behaviors may explain the association between SI and HPA axis activity. A greater number of social roles may provide positive social motivation towards engagement in healthier behaviors(29). Health behaviors, in turn, have been associated with HPA axis activity. Current smoking and shorter self-reported sleep duration have been linked with increased total cortisol secretion(30, 31). Likewise, self-reported and actigraphy-measured short sleep duration and variable sleep duration, as well as long-term, excessive alcohol consumption are associated with a flatter diurnal slope(32–35).

Possible Confounders

While examining possible pathways linking SI and cortisol slope, this study also examines possible third-variable confounds that may explain this relationship. Prior work has linked trait neuroticism with elevated bedtime cortisol(36) and higher mean daily AUC(24) but not with slope(37). Higher conscientiousness has been associated with steeper slope(37) but not with AUC(24). Given that personality differences are linked with patterns of diurnal cortisol secretion, and that select personality traits may lead individuals to engage in more diverse social roles(38), such factors may explain any observed associations between SI and cortisol dynamics. However, to date, no study has accounted for personality when investigating the link between SI and cortisol slope.

Previous Investigation of Explanatory Pathways

To our knowledge, only one study has investigated possible pathways explaining the relationship between SI and HPA axis activity. The previously cited Chin and colleagues(17) examined the role of stress, health behaviors (smoking, drinking, physical activity), and affect (positive and negative) in accounting for this relationship, and showed an effect for older adults only. Mediation analyses revealed no significant indirect effects explaining the relationship between SI and cortisol in the older adults in this sample.

The current investigation explores pathways that may account for links between SI and cortisol, including circadian factors such as sleep, and patterns of daily social interactions. Additionally, the present report examines whether the association between social integration and cortisol measures persist after adjustment for personality traits. The present report utilizes ambulatory measures of sleep, social behavior, affective states, and physical activity as potential triggers of neurohormonal activity in the context of individuals’ daily lives.

Methods

Participants

Participants were drawn from the Adult Health and Behavior-II (AHAB-II) cohort from the University of Pittsburgh, a study of risk factors and subclinical CVD in healthy midlife adults which recruited individuals between March 2008 and October 2011. To be eligible to participate, participants had to be between 30 and 54 years old and working at least 25 hours per week outside the home. Full exclusionary criteria have been reported elsewhere(39). The study was approved by the University of Pittsburgh Institutional Review Board. Participants provided informed consent and received compensation up to $410.

Procedure

Participants completed four visits relevant to this study. Demographics and social network information were assessed at Visit 1. Participants received extensive training and practice on the use of an electronic diary and ambulatory monitoring equipment at Visit 2 before initiating the ambulatory monitoring phase. During this 7-day monitoring period, participants wore an Actiwatch-16 actigraphy device on the wrist continuously (Mini-Mitter Inc., Bend, OR) for estimating sleep patterns, and a SenseWear armband monitor during waking hours (Body Media, Pittsburgh, PA; SenseWear Pro3) for measuring daily energy expenditure. Participants also carried an electronic diary (Palm Z22) and collected salivary cortisol for four days in the midst of the week (3 work days and 1 non-work day). Salivary cortisol was collected at waking, at 30 minutes, 4 hours, and 9 hours after waking, and at bedtime each these four days, with sampling timing prompted by the electronic diary. To collect the sample, participants were instructed to gently chew on a supplied cotton swab for 2 minutes, place it into a provided salivette, and then store the salivette in their refrigerator until returning to the lab. For compliance purposes, the electronic diary prompt displayed a unique 4-digit code that participants were instructed to write on the salivette at the time of sampling. Participants were given a 5-minute window to complete each prompt. Additional information on collection and compliance procedures is provided elsewhere(39).

Measures

Social Integration (SI)

Social integration was measured using the Social Network Index (SNI)(18). This questionnaire assesses engagement in 12 different types of social relationships, including relationships with family members, friends, workmates, and social/religious groups, and it provides a count of the number of social roles in which participants have participated in at least once in 2 weeks. This latter measure served as our index of SI (range = 1–12).

Cortisol

Cortisol samples were assayed in duplicate in the laboratory of Dr. Clemens Kirschbaum (Dresden, German) using a commercially available chemiluminescence immunoassay (IBL-International) with a cortisol-biotin conjugate as a tracer with a sensitivity of 0.43 nmol/L. Both the intra and interassay variance coefficients were below 8% (26). Cortisol samples with values below the lowest reliably detected levels (0.3 nmol/L) or above 60 nmol/L (outliers determined from examination of preliminary distributions) were excluded from all analyses. After exclusion of invalid data (missing samples, samples with unverifiable timing, out-of-range cortisol values) an average of 95.1% of each participant’s samples were available for analysis. To be consistent with previous work utilizing this data and to eliminate the influence of the cortisol awakening response on AUC and slope calculations, the second cortisol value of the day (30 minutes post-awakening) was excluded from calculations (26, 39). There is some suggestion that this awakening response may be regulated by different neurobiological mechanisms than the rest of the underlying diurnal cortisol curve (40).

In order to account for variation in day length, slope calculations regressed each participant’s log-transformed cortisol values on number of minutes since awakening on each monitoring day(9). Slope values were calculated for days that had non-missing data for beginning of day and either 9^th hour or bedtime cortisol samples. The resulting slope values were averaged by individual across the available monitoring days. Individuals with less than two days of slope calculations were excluded from analyses. Area under the curve (with respect to ground) was estimated from raw cortisol values by trapezoidal integration, as described by Pruessner et al.(41). AUC values were averaged across the available monitoring days for each participant to yield a trait-like measure of cortisol AUC. Individuals with less than two days of complete AUC data were excluded from AUC analyses.

Variances associated with each of the relevant cortisol indices were estimated across days and participants (Proc Var in SAS). These were then used to estimate within-person reliability (equation #5 from Cranford et al., 2006). Diurnal cortisol slope and AUC values were reasonably consistent within-people (α = 0.84, α = 0.69; respectively). We also conducted separate analyses of the cortisol samples collected during each of the five diurnal periods, averaged across days.

EMA

Social interactions during daily life were measured by electronic diary using a 41-item questionnaire administered hourly during the waking hours of each of the four monitoring days. As part of each questionnaire, participants were asked about the timing, participants, and quality of their most recent social interaction. For assessment of quality, four Likert scale items were used with responses [NO! No no yes Yes YES!] coded into a 1–6 scale. Two items assessed positive features (i.e., “agreeable interaction?” and “pleasant interaction”) and two assessed negative features (i.e., “someone in conflict with you?” and “someone treated you badly?”)(42). From these hourly EMA assessments, seven interaction scores were derived. The relative frequency of total social interactions was calculated by dividing the number of observations spent in a social interaction (in the last 10 minutes or currently) by the total number of hourly observations. Relative frequency of positive interactions involved counting the number of positive interactions (score of “4: yes” or more on one or more of the positive features above) and dividing by the frequency of recent social interactions. Relative frequency of negative interactions was defined similarly, using scores of at least “4: yes” on the reverse-coded negative features. Social interaction timing was assessed using four variables. Social interaction “window length” was conceptualized as the difference in time between the first and the last social interaction of the day. Social interaction midpoint was defined as the time of the first social interaction of the day plus half of the window length. Variability in social interaction window length and social interaction midpoint were scored as the standard deviation of the analogous metrics. All measures were aggregated for each individual. Variability measures were log-transformed to reduce the effect of outliers.

In addition to social interactions, participants were queried about their current mood state via the hourly questionnaire. Participants responded to two items (happy, cheerful) assessing current positive affect and six items (upset, hostile, nervous, afraid, lonely, and sad) assessing current negative affect. Item responses were converted to a 1- to 6-point rating and averaged by person to derive trait-level positive and negative affect scores. In an analogous manner, anxiety (nervous, afraid), hostility (hostile, angry), depression (lonely, sad), and loneliness (lonely) scores were calculated, using individual items from the negative affect scale.

Health Behaviors

Substances

Smoking status (nonsmoker, ex-smoker, current smoker, and other tobacco user) was reported at Visit 1 and was coded dichotomously (1=current smoker or tobacco user, 0=non-smoker or ex-smoker). At Visit 1, participants reported the number of alcoholic beverages consumed over the course of a week. This value was dichotomized into little to none (0–1 alcoholic drinks peer week) and 2+ alcoholic drinks per week.

Physical Activity

Energy expenditure data was calculated from the SenseWear armband monitor in 1-minute epochs using data from a biaxial accelerometer, a heat flux sensor, a galvanic skin response, a skin temperature sensor, and near body temperature sensor included in this device(39, 43). Objective energy expenditure was operationalized as the average level of metabolic expenditure (METs units) per day. For this measure, only days on which salivary cortisol samples were collected were included(39).

Sleep

Both self-reported sleep quality and actigraphy-derived sleep time were collected. Participants completed the Pittsburgh Sleep Quality Index (PSQI)(44). The PSQI global sleep score encapsulates sleep quality, duration, efficiency, disturbance, use of sleeping medication, and daytime dysfunction over the past month, where higher scores indicate poorer sleep quality. This measure has been shown to discriminate between good sleepers and sleep disordered individuals(44–46). Actigraphy sleep pattern data were collected in 1-minute epochs across the night using the Actigraph-16 (Bend, OR: Philips Electronics) using automated, standard medium thresholds to derive estimates of sleep. Four sleep variables were used in analyses: sleep midpoint, total sleep time, variability in sleep midpoint, and variability in total sleep time. Sleep midpoint is calculated as the midpoint between sleep onset and wake onset, where sleep onset is defined as a period lasting at least 10 consecutive minutes with activity counts <40 per epoch and wake onset is defined as 10 consecutive minutes of ≥40 activity counts per epoch. Total sleep time is the total duration of time elapsed between sleep onset and wake onset, excluding periods of waking activity during the sleep interval(39). Nightly sleep midpoint and total sleep time measurements were averaged across days to derive person-level means. The intraindividual variations, or standard deviations, around these means were calculated to derive person-level variability measures for each of the sleep metrics. Variability measures were log-transformed to reduce the effect of outliers.

Trait Personality

At Visit 4, participants completed the 240-item Revised NEO Personality Inventory (NEO-PI-R)(47) to assess trait openness, conscientiousness, neuroticism, agreeableness, and extraversion. Participants responded on a 5-point scale to each item (0=Strongly Disagree, 4=Strongly Agree). These scales have been previously found to have adequate validity and internal consistency (Cronbach’s α > .85) in middle-aged adults (47, 48).

A priori covariates

Four covariates known to associate with diurnal cortisol slope were identified prior to statistical analyses. These covariates include self-reported gender (male vs. female), age, race (White vs. non-White), and highest level of education (1 = high school diploma or less, 2 = associate or technical degree, 3 = bachelor’s degree, 4 = graduate degree). All information was gathered via the demographic questionnaire administered at visit 1.

Statistical Analyses

Regression and correlation analyses were conducted using R statistical software(49). Linear regression models were used to assess the relationship between SI and cortisol measures, controlling for the a-priori covariates. A series of partial correlation models were then run to assess the association between potential explanatory variables (both potential third-variable confounds and potential mediators) and cortisol measures found to associate with social integration, again, controlling for the a-priori covariates and adjusting for multiple comparisons using the Benjamini and Hochberg method with the false discovery rate set at 5% (50). Correlations with social interaction timing variables also controlled for analogous day parameters (day length, window, and variability). The potential confounds found to associate with diurnal cortisol slope were then added to the initial linear regression models assessing the relationship between SI and cortisol measures.

Potential mediators that were found to be significantly associated with diurnal cortisol slope were next examined in relation to social integration through linear regressions. Analyses of indirect effects were run for potential mediators that had demonstrated significant associations with both diurnal cortisol slope and social integration. Analyses were run using the Mediation statistical package (v4.6.6) for causal mediation analysis in R software 3.5.1(51). Significance criterion for the a-pathways, b-pathways, and indirect effect pathways was set at p < .05. For each of the indirect effects models, a-priori covariates were included, and 10,000 bootstrapping samples were utilized to generate unstandardized beta coefficients, standard errors, and confidence intervals.

Results

Sample Characteristics

Of the total sample of N = 494, 38 participants were removed due to missing or invalid data. Excluded participants did not differ from included participants in terms of SI, cortisol slope, AUC, age, education, or race, but they were more likely to be male (p = .044). We removed 30 participants who had fewer than 4 nights of sleep monitoring, 3 participants who had missing SNI data, and 4 participants with missing cortisol data. One additional participant was removed who produced cortisol slope values more than 3 standard deviations above the study mean. The analytic sample was composed of 456 healthy, employed adults (53.9% female, 82.0% white, 72.1% bachelor’s degree or greater, mean age of 42.86). Analyses of the individual diurnal cortisol collection periods varied in sample size (see Table 1). SI showed a relatively normal distribution in this population, with a mean score of 6.28 social roles (range of 1–11). Cortisol slope had a mean value of −0.12 Δnmol/L/min (range of −0.01 to −0.26). See Table 1 for additional sample characteristics.

Table 1.

Descriptive Statistics

Variable	n	Mean or %	SD
Salivary cortisol measures
Sample 1 (nmol/L)	456	17.86	6.34
Sample 2 (nmol/L)	444	24.42	7.58
Sample 3 (nmol/L)	454	8.17	3.24
Sample 4 (nmol/L)	453	5.72	2.55
Sample 5 (nmol/L)	455	3.11	2.73
Area under the curve (AUC) (nmol/L)	446	121.15	37.53
Diurnal cortisol slope (Δnmol/L/min)	456	−0.12	0.04
Demographics
Age	456	42.86	7.32
Sex (female)	456	53.9%
Race (white)	456	82.0%
Education	456
Less than HS		5.9%
HS degree		21.9%
College degree		38.4%
Graduate degree		33.8%
Health Behaviors
Energy expenditure (average METs)	445	1.63	0.29
Chronotype	448	39.40	7.12
Sleep midpoint (hours)	456	3:05	1:04
Total sleep time (hours)	456	5.94	0.87
Variability in sleep midpoint (hours)	456	0:56	0.37
Variability in total sleep time (hours)	456	1:06	0:39
Sleep quality	450	4.99	2.62
Smoking status (smoker)	455	14.5%
Alcohol use (2+ drinks/week)	451	45.5%
Social integration (# social roles)	456	6.28	2.03

Social Integration and Cortisol

As reported in Table 2, mean cortisol slope was associated with age, sex, and race, with older individuals, women, and nonwhite individuals showing flatter slopes. After covariate adjustment for age, sex, race, and education, greater SI was associated with a steeper cortisol decline as predicted (see Table 2). We investigated the relationship between SI and cortisol at specific times of the day. In these analyses, SI was associated with cortisol only for the bedtime sample, with more socially integrated individuals showing lower cortisol levels for the final cortisol reading of the day (see Figure 1). The model in which bedtime cortisol was regressed on SI also controlled for time since awakening to adjust for those with longer waking days. This association at the end of day remained significant before and after controlling for hours since awakening. Consistent with much of the previous literature, SI was not significantly associated with AUC cortisol (see Table 2), and thus all additional analyses were performed only for cortisol slope.

Table 2.

Diurnal Cortisol Slope and AUC Regressed on Social Integration

	Cortisol Slope			Cortisol AUC
Variable	B	SE	p	B	SE	p
Step 1
Intercept	−0.16186	0.01488	<.001***	131.58	14.37	<.001***
Age	0.00081	0.00026	.002**	0.2159	0.2472	.38
Sex	0.00895	0.00377	.018*	−6.0502	3.6346	.097†
Race	0.02374	0.00499	<.001***	−7.9995	4.8214	.098†
Education	0.00118	0.00216	.59	−0.8641	2.0834	.68
Social Integration	−0.00253	0.00092	.006**	−1.0180	0.8832	.25
Multiple R2 = .09
Step 2
Intercept	−0.1309	0.01916	<.001***
Age	0.00079	0.00026	.002**
Sex	0.0109	0.00384	.005**
Race	0.02590	0.00508	<.001***
Education	0.00211	0.00222	.34
Social Integration	−0.00209	0.00094	.027*
Conscientiousness	−0.00014	0.00010	.17
Extroversion	−0.00018	0.00009	.061†
Multiple R2 = .12

Note.

^†An indicates p < .10

^*an indicates p < 0.050

^**an indicates p < .010, and

^***an indicates p < .001.

Figure 1.

Diurnal Cortisol Slope by Social Integration

Note. For visualization purposes, social integration was divided into “high” and “low” through a median split. The “low” group corresponds to Social Network Index scores of 1–5 and the “high” group represents scores of 6–11.

Partial correlations (controlling for the a priori covariates) demonstrated that total sleep time, variability in total sleep time, variability in sleep midpoint, and proportion of positive interactions were all significantly correlated with diurnal cortisol slope (see Table 3). These four “daily life” variables were identified as potential indirect pathways. Simultaneously, while none of the potential personality variable confounds showed statistical significance after adjustment for multiple comparisons, the associations between extraversion and conscientiousness reached marginal significance. Given the strong theoretical potential for confounding and general dearth of literature controlling for trait-like personality factors, extraversion and conscientiousness were identified as potential confounders.

Table 3.

Correlations Between Potential Explanatory Variables and Diurnal Cortisol Slope

	Diurnal Cortisol Slope
Openness (NEO)	0.04
Extroversion (NEO)	−0.12†
Neuroticism (NEO)	0.07
Agreeableness (NEO)	−0.08
Consciousness (NEO)	−0.12†
Total Sleep Time	−0.24***
Sleep Midpoint	0.01
Variability in Total Sleep Time	0.13*
Variability in Sleep Midpoint	0.23***
Sleep Quality (PSQI)	0.11†
Smoking Status	0.09
Alcohol Use	0.03
Physical Activity (METS)	0.05
Negative Affect	0.03
Positive Affect	0.01
Loneliness	0.03
Depression	0.03
Hostility	0.01
Anxiety	0.04
Proportion of Positive Interactions	−0.13*
Proportion of Negative Interactions	0.04
Frequency of Social Interactions	−0.05
Social Interaction Window	−0.02
Social Interaction Midpoint (hours)	−0.05
Variability in Social Interaction Window (hours)	0.05
Variability in Social Interaction Midpoint (hours)	−0.04

Note: Chart depicts partial correlations between cortisol slope and potential covariates and mediators. All correlations control for age, sex, race, and education. All p-values are adjusted for the false discovery rate (5%).

^***depicts p <.001

^**depicts p <. 01

^*depicts p < .05, and

^†depicts p < .10.

Potential Confounders

After adjusting for extraversion and conscientiousness, SI remained an independent predictor of diurnal cortisol slope (b = −0.00209, p = .027) (Table 2), The observed association between SI and cortisol slope may reflect patterns of ongoing social behavior that are not accounted for by standard personality traits. Because a previous study on this topic had found moderating effects of age(17), we also examined the interactions between age and SI on diurnal cortisol slope and AUC; these effects were not significant (ps > .71). Likewise, in accordance with previous work showing gender differences in the link between SI and health(52), we examined gender as a moderator. We showed no interaction between gender and SI on diurnal cortisol slope and AUC, (ps > .14).

Potential Indirect Effects

We examined several remaining factors, found to be correlated with diurnal cortisol slope, that may help account for the observed association between SI and cortisol slope: total sleep time, variability in total sleep time, variability in sleep midpoint, and proportion of positive interactions.

As demonstrated in Table 4, socially integrated individuals differed from less socially-integrated individuals with respect to their typical sleep patterns, showing somewhat longer sleep times on average, but this association did not meet conventional levels of significance (b = 0.03751, p = .053). In terms of sleep variability, socially integrated people showed less variability in their sleep midpoint than their socially isolated counterparts (b = −0.02547, p = .034), but did not differ from less socially integrated individuals in their variability in total sleep time (see Table 4). Tests of indirect effects demonstrated that variability in sleep midpoint accounted for a portion of the relationship between SI and cortisol slope (b = −0.00042, 95% CI: −0.00095, −0.00001, p = .042). While relative frequency of positive social interactions was related to diurnal cortisol slope, it was not significantly associated with SI (p = .51). Socially integrated individuals in this sample were not more likely to engage in more frequent positive interactions.

Table 4.

Exploratory Regression Results and Tests of Indirect Effects

Model	B	SE	p	95% CI
Total Sleep Time
SI → TST (a)	0.03751	0.01931	.053†	−0.00044 – 0.07545
TST → diurnal cortisol slope (b)	−0.01111	0.00218	<.001***	−0.01539 – −0.00683
SI → TST → diurnal cortisol slope (a×b)	−0.00042		.053†	−0.00091 – 5.6x10−6
Variability in Total Sleep Time
SI → variability in TST (a)	0.00223	0.01088	.84	−0.01916 – 0.02361
Variability in TST → diurnal cortisol slope (b)	0.01116	0.00394	.005**	0.00342 – 0.01890
Variability in Sleep Midpoint
SI → variability in SM (a)	−0.02547	0.01195	.034*	−0.02547 – −0.00199
Variability in SM → diurnal cortisol slope (b)	0.01649	0.00353	<.001***	0.00955 – 0.02344
SI → variability in SM → diurnal cortisol slope (a×b)	−0.00042		.042*	−0.00095 – −0.00001
Proportion of Positive Interactions
SI → Prop. Positive Interactions (a)	0.00090	0.00138	.51	−0.00180 – 0.00361
Prop. Positive Interactions → diurnal cortisol slope (b)	−0.08853	0.03049	.004**	−0.14846 – −0.02860

Note: Analyses included age, sex, race, and education as covariates. SI= Social Integration, TST= Total Sleep Time, SM = Sleep Midpoint. Tests of indirect effects are only reported for variables in which the a-path and b-path are both significant.

^†An indicates p < .10

^*an indicates p < 0.05

^**an indicates p < .01, and

^***an indicates p < .001.

Discussion

The link between SI and mortality and morbidity is well established in the literature(1), and the HPA axis provides one possible mechanism through which this association may operate. The findings of the current study are congruent with earlier work, showing that a higher number of social roles is associated with steeper diurnal slope, with lower cortisol levels being especially apparent at the end of the day for socially integrated individuals(13, 17). The current paper also expanded previous work to adjust for possible third-variable confounds in this relationship. Specifically, extroversion and conscientiousness were adjusted for when examining the relationship between SI and diurnal cortisol slope, as they were shown to predict diurnal cortisol slope in this sample. The present report showed that relationships between SI and cortisol slope are not solely a function of personality differences.

SI was not associated with AUC in this sample. Previous studies have shown mixed results with respect to the link between SI and AUC(13, 15). These discrepant findings may be due to differences in sample characteristics across studies. The present sample, for example, is older than samples in previous studies demonstrating relationships between SI and AUC(13, 17). It is possible that such associations are moderated by age.

In addition to replicating previous links between SI and HPA-axis functioning, this study provides new insight into the role of health behaviors, daily social interactions, and mood in explaining the relationship between SI and diurnal cortisol slope. Similar to Chin et al.(17) the present study did not find evidence that physical activity, alcohol use, or smoking status explain the relationship between SI and diurnal cortisol slope. Daily mood ratings and negative interactions, likewise, were unrelated to diurnal cortisol slope, and therefore were not examined as potential mediators.

The most productive new findings involved actigraphy-derived measures of sleep and sleep variability. Longer sleep duration, less variable sleep midpoint, and less variable sleep duration all predicted a steeper (healthier) diurnal slope pattern in this sample. In turn, more socially integrated individuals demonstrated less variable sleep midpoints than less socially integrated individuals. Most notably, the association between SI and cortisol slope appeared to be accounted for, in part, by sleep midpoint variability. These relationships are in line with previous work demonstrating links between sleep/wake variability and flatter cortisol slope(35), as well as general negative health and wellbeing outcomes(53). As a whole, these results raise the possibility that circadian factors may play a role in the link between SI and health outcomes.

It is possible that the diversity, frequency, or regularity of social contacts associated with a socially integrated network may assist in the entrainment of circadian rhythms, accounting for the associations between SI, evening dips in cortisol, and sleep patterns. A parallel body of research has demonstrated links between social zeitgebers, or recurring daily activities, cultural demands, or social interactions, and programming of daily biological rhythms(19, 54). While the present study did not find that social interaction window length, midpoint, or variability in length and midpoint accounted for differences in cortisol slope, there may be other, more complex patterns of social interaction conditioned by social integration that may have an effect on the sleep-wake cycle. For instance, the presence of family obligations or activities in the morning may prompt individuals to adapt an earlier and more consistent sleep-wake cycle, which may in turn influence cortisol patterning. Alternatively, social patterns that promote productive “unwinding” may lead to increased positive affect or lower anxiety at bedtime, which may facilitate earlier and more consistent sleep behavior. Interestingly, previous work has demonstrated links between daily ratings of happiness and calmness and increased nightly sleep duration and quality(55). The current study is the first, to our knowledge, that has linked SI with characteristic patterns of sleep-wake activity, providing some evidence consistent with the possibility that differences in sleeping patterns may potentially account, in part, for associations between SI and positive biological outcomes. Further research is needed to help us better characterize the mechanisms accounting for the association between SI and sleep.

This study does not come without limitations. First, the cross-sectional design precludes inferences of causality. Further work should replicate these findings in a longitudinal sample. Likewise, by virtue of the study criteria (which excluded individuals with CVD and those currently on autonomic medications), the participants were, by definition, healthy, and therefore somewhat less representative of the midlife population. Similarly, the present sample is more highly educated than the national average, with 72% of individuals possessing a bachelor’s degree. This being said, the large sample size and representative ethnic and gender dispersion does provide an otherwise relatively good estimate of the local population norms. The present study also has many strengths. Most notably, the EMA design provides reliable measures over extended time of several variables, such as mood, social interactions, sleep, and physical activity that might not be as accurately depicted using retrospective survey reports. The study provides the most complete analysis to date of putative factors potentially explaining the relationship between cortisol slope and SI.

The findings from this study are consistent with a growing literature demonstrating a relationship between SI and diurnal cortisol slope, a marker of HPA-axis functioning(13–17). Future research should examine the extent to which such associations play a role in the association between SI and clinical health outcomes. Moreover, subsequent studies should investigate the role that circadian factors, perhaps entrained by patterns of social activity, may play in linking SI with health outcomes. The present study suggests that SI differences in diurnal cortisol may, in part, operate through sleep and sleep variability. The processes by which social role behaviors may shape sleep habits or vice versa remain to be determined.

Conclusion

Overall, our findings strengthen the evidence linking increased social integration with improved HPA axis functioning, and suggest that sleep duration and variability in midpoint may play a role in explaining this relationship. Given the potential link between blunted HPA functioning and cardiovascular disease, results advance our understanding of the biological and behavioral pathways that may account for the observed associations between social integration and health.

Glossary

SI

social integration

HPA

hypothalamic-pituitary-adrenocortical

AUC

area-under-the-curve