Associations between adolescent perceived loneliness and hair cortisol concentration

Abstract

Introduction

Adolescents experience high levels of loneliness, which is linked to poor health in adulthood. Loneliness may contribute to poor health through chronic dysregulation of the hypothalamic-pituitary-adrenal axis. In this analysis, we examined the associations between survey- and ecological momentary assessment (EMA)-based measures of loneliness and hair cortisol concentrations (HCC) in a sample of 1102 adolescents and assessed sex differences in this relationship.

Methods

Data came from wave 1 of the Adolescent Health and Development in Context study. We conducted a series of multivariable linear regression models to examine the associations between loneliness and HCC. Models were adjusted for adolescent and caregiver demographics, adolescent clinical factors, adolescent hair care practices, and adolescent lifetime mental health diagnosis and current psychotropic medication use. An interaction term between sex and loneliness was added to assess for effect moderation.

Results

In our sample, the mean HCC was 1.35 pg/mg (SD=1.1). The mean for the unstandardized survey loneliness measure was 1.79 (SD=0.79) for the total analytic sample. The unstandardized mean for the EMA loneliness measure was − 0.02 (SD=2.1) for the total analytic sample. In model one testing the bivariate linear relationship between loneliness and HCC, higher loneliness via survey and EMA measures was associated with lower HCC (Survey: b= − 0.10, SE=0.03, p=.004; EMA: b= − 0.09, SE=0.03, p=.005). In model two, higher loneliness remained significantly associated with lower HCC (Survey: b= − 0.07, SE=0.03, p=.023; EMA: b= − 0.07, SE=0.03, p=.037), after controlling for the following covariates: sociodemographic factors, pubertal development and BMI, corticosteroid use, hair care practices, season of collection and assayed hair length. In model 3, youth lifetime mental health diagnosis and current psychotropic medication use were added into the regression model, and higher loneliness remained significantly associated with lower HCC (Survey: b= − 0.07, SE=0.03, p=.029; EMA: b= − 0.07, SE=0.03, p=.039). There was no effect modification by sex (Survey: b=0.04, SE=0.06, p=.552; EMA: b= − 0.01, SE=0.06, p=.843).

Conclusions

In our analysis, both survey- and EMA-reported loneliness measures were associated with lower HCC. No evidence of an interaction between sex and loneliness was observed. Future research is needed to validate these findings and investigate longitudinal relationships among adolescent loneliness, stress physiology, and downstream health sequelae.

1. Introduction

Social connection is a vital part of the human experience and perceived loneliness represents a significant threat to health and well-being. Perceived loneliness refers to a painful experience that develops when individuals perceive social isolation or disconnection from others, or they feel their social relationships do not meet their needs for human intimacy or emotional closeness (1–3). While perceived loneliness can be temporary and/or situational, chronic or persistent feelings of loneliness are of public health concern given the evidence linking loneliness to numerous chronic conditions in adulthood including cardiovascular disease, obesity, hypertension, stroke, poor sleep, depression, dementia, and reduced life expectancy (4–11).

In 2023, the US Surgeon General released an advisory report due to the increasing rates of loneliness in the country over the past several decades (12). Adolescents may be particularly vulnerable to feelings of loneliness given the greater focus on peer relationships and heightened sensitivity to social rejection during this period of development (13, 14). Between 2012 and 2018, reports of adolescent loneliness at school increased nearly two-fold with a higher prevalence reported for adolescent females (15). Moreover, while loneliness increased worldwide during the COVID-19 pandemic, one third to over one half of adolescents continue to report moderate levels of loneliness post-pandemic (16). In a recent longitudinal study in Japan, researchers examined the persistence of loneliness across adolescence (N=3165) and identified four loneliness trajectories: consistently low levels of loneliness, moderate to decreasing levels, moderate to increasing levels and consistently high levels (17). Compared to the consistently low loneliness trajectory, adolescents in the other three trajectories were at increased risk for self-harm and suicidal ideation, particularly adolescents who reported consistently high loneliness followed by moderate to increasing loneliness highlighting the risk of chronic loneliness for adolescent mental health and wellbeing.

In pioneering work, Hawkley and Cacioppo theorized that perceived loneliness and social isolation negatively affect physical and mental health through complex (and potentially reciprocal) psycho-bio-behavioral processes set into motion by feeling unsafe socially and implicit hypervigilance for potential social threats that may occur in the future (18). Over time, this hypervigilance affects how lonely persons view and interact with their social milieu as they anticipate negative or threatening social interactions and often respond with feelings of distress, distrust, and hostility that may elicit an actual negative social interaction and/or self-imposed social isolation as a “protective” behavior. When prolonged, these altered psycho-behavioral processes can contribute to neuroendocrine and immune dysregulation (e.g., inflammation) and subsequent impairment to physical and mental health.

The hypothalamic-pituitary-adrenal (HPA) axis, a key component of stress responsivity, is hypothesized as one pathway through which loneliness contributes to poor health (19–22). In prior research with adolescents and young adult samples, momentary, daily, and “trait” or global loneliness were all linked to irregularities in the salivary cortisol diurnal curve, a biomarker of the HPA axis (23, 24). To date, research linking loneliness and hair cortisol concentration (HCC) - a measure of HPA activity over time (one centimeter of hair growth approximates the cumulative cortisol concentration over the preceding month) is limited, and in one of the only studies available, researchers found HCC declined overall among adults during the COVID-19 pandemic lockdown, but the associations between their perceptions of loneliness, measured as feelings of loneliness overall, and HCC were not statistically significant (25).

In the current analysis, we expand on prior research to examine two measures of perceived loneliness using survey- and ecological momentary assessment (EMA) methods and their associations with HCC in a racially and socioeconomically diverse sample of adolescents residing in a large metropolitan area of the midwestern US. Because some stressful experiences as well as mental health conditions have been associated with higher and others with lower cortisol concentration, we did not impose directional hypotheses on the relationships between perceived loneliness and HCC for this analysis. Researchers hypothesize that in healthy stress response systems, cortisol concentration increases in response to an acute stressor. However, when stressors persist, recur, and/or the response to a stressor remains heightened (e.g., through anticipatory or rumination processes associated with hypervigilance surrounding potentially negative social interactions), cortisol concentration may become dampened over time possibly due to alterations in glucocorticoid receptor sensitivity (26). In addition to the main effects, we explored the potential for sex differences in the relationship between loneliness and HCC. Evidence suggests that sex differences in HPA axis activity begin to emerge during puberty, however, some studies have found a greater cortisol response to stressors for males and others for females (27). To date, the studies examining relationships between perceived loneliness and cortisol response during adolescence have not explored sex differences, and in one of the few studies exploring sex differences during adulthood, older adult males with greater perceived social and family loneliness were found to have an altered diurnal cortisol pattern, but no significant relationships were found for older adult females (28). Thus, we also explore moderation without hypothesized directionality of the relationship.

2. Methods

2.1. Study Design, Sample and Data Collection Procedures

We analyzed data from wave 1 of the Adolescent Health and Development in Context study (AHDC), a prospective cohort study on the effects of activity space exposures on the health and wellbeing of urban and suburban youth aged 11–17 years (N=1405). The AHDC was conducted in 2014–2016 in Columbus, Ohio, including surrounding suburban municipalities. Sampling, recruitment, and data collection were conducted in collaboration with a university-based survey research center with expertise in prospective cohort studies. Detailed study design and sampling procedures have been described in prior reports (29–31). The AHDC study and additional analyses were approved by the university institutional review board. Parental consent and youth assent were obtained prior to data collection.

Briefly, data were collected over a weeklong period using a variety of strategies. At the beginning of the week, trained interviewers conducted a face-to-face interview and self-administered survey with the youth and their primary caregiver in the home setting. The youth was then provided with a study-based smartphone for a seven-day smartphone-based Global Positioning System tracking and EMA with the latter consisting of the administration of five randomly timed mini-surveys per day over the course of the week excluding school hours on school days. The EMA averaged approximately 4.5 minutes in length and youth were asked about their current location, activities, and mood, including a question on whether the respondent felt lonely in the moment. Youth were given 20 minutes to acknowledge the prompt and an additional 20 minutes to complete the questionnaire. Youth were asked to report on their experiences at the initial time of the prompt rather than those experiences occurring before the prompt or later in the 20-minute window. Youth received a monetary incentive based on the amount of EMA survey prompts they completed; the EMA response rate was 53%. At the end of the weeklong period, the interviewer returned to the home to conduct a face-to-face exit interview with the youth and collect the hair sample while the primary caregiver completed a self-administered survey. Interviewers had been trained in the hair biomarker collection procedures. After the hair collection, specimens were stored in a locked cabinet at room temperature prior to assay. Youth were excluded from hair cortisol collection if the primary caregiver reported corticosteroid use on the initial phone survey screening for eligibility (n=27). However, in querying caregivers on youth medication use during the in-home survey, corticosteroid use (oral, nasal, inhaled and topical routes of administration) was reported for 67 youth and responses to use were missing for 35 youth. As the hair samples for these youth were assayed for cortisol, we conducted separate statistical analyses with these youth excluded and included (corticosteroid use as covariate).

2.2. Measures

2.2.1. Dependent variable: Hair cortisol concentration

Aligned with our published techniques (28), trained interviewers collected approximately 25–75 mg of hair (0.4–1 cm in diameter) from the posterior vertex region of the scalp cutting as close to the scalp as possible with thinning shears. Cortisol is leached out of hair at more distal lengths (> 3cm) due to repeated washing and environmental exposures (32–34). Thus, consistent with prior research we focused the analysis on the first 3 cm of hair growth. Most youth (80%) had 3 centimeters (cm) of hair for assay. We also included youth with less than 3 cm of hair to minimize bias due to shorter hairstyles. Less than 4% of the hair samples did not have the root end identified to cut the hair to 3 cm in length, thus we assayed the entire strand of hair. Hair specimens were washed with high performance liquid chromatography-grade isopropanol and dried for 1–3 days before being ground into powder. Methanol was added to the ground sample to extract cortisol from the pulverized hair and incubated for 18–24 hours at room temperature with constant agitation. The tubes were centrifuged to pellet the powdered hair, and the supernatant (methanol containing cortisol) was transferred to a clean microcentrifuge tube. Methanol was removed using centrifugation/evaporation for 4–6 hours at room temperature. The cortisol extract was then reconstituted in 100ul of Salimetric immunoassay cortisol analysis diluent buffer. Samples were assayed in duplicate and inter- and intra-assay coefficients of variation were 16.2% and 9.9%, respectively. The HCC was natural log transformed due to the skewed distribution and then winsorized for values three standard deviations above and below the mean (n=7). Results are reported in pg/mg.

2.2.2. Primary Independent Variables of Interest: Loneliness

Two measures of loneliness were examined for this analysis. First, an in-home survey-based loneliness composite measure was created using the 9-item Loneliness Questionnaire-short version (35). This short version of the scale was adapted from a longer questionnaire (36) to enhance the validity and reliability of differentiating youth on the trait scale of loneliness. Youth rated how often they felt the following: I have nobody to talk to, It’s hard for me to make friends, It’s hard to get kids to like me, I don’t have anyone to hang out with, I feel left out of things, There’s no other kids I can go to when I need help, I don’t get along with other kids, I’m lonely, and I have no friends. Response options included always true, true most of the time, true sometimes, hardly ever true, or not true at all and positive items were reverse coded, so higher scores represented greater loneliness. The Cronbach alpha of the scale items in this analysis was .92. Results of exploratory factor analysis (scree plot, Eigen values and factor scores) also supported a single construct. The scores of the 9-items were summed and averaged to create a mean total score for youth who responded to 50% or more of the items; youth who did not respond to at least 50% of the items were set to missing.

Second, an EMA-based loneliness measure of everyday perceptions of loneliness was created using the repeated responses over the weeklong period to a single item asked to youth at each EMA. Specifically, each EMA included an adapted version of the Positive and Negative Affect Schedule (PANAS), in which youth were asked to look at a set of words, which included “lonely”. The youth were then asked, “How are you feeling right now? Please rate how well each word best fits your mood”; response options included, “not at all”, “a little”, “somewhat”, “quite a bit” and “extremely”. The EMA-loneliness measure was created using the estimated youth-level random effect from a multilevel ordinal logit model of EMA-level in-situ loneliness reports preserving the five response categories (nesting EMAs within youth). The variable values can be interpreted as the youth-specific deviation from the average cumulative log odds of responding to the loneliness question at any given category or above versus below. Predicted values incorporate an empirical Bayes (EB) adjustment, “shrinking” values toward the overall mean as a function of their level of unreliability. A detailed discussion of the method used to create the EMA loneliness measure was previously published (37).

2.2.3. Covariates

Covariates in the analysis were selected based on theory and to be consistent with prior research recommendations and analyses (23, 38–40): sociodemographic characteristics, youth clinical factors, youth hair care practices, and study procedures (season of collection and assayed hair length). Sociodemographic factors included youth race and ethnicity, which was categorized as (1) non-Hispanic Black/African American or mixed-race Black/African American; (2) “other” race and ethnicity (grouped youth who identified as non-Black/African American mixed race, Hispanic (any race), Asian, or “other” due to the small sample sizes in individual categories); and (3) non-Hispanic White youth (reference category). Youth sex was categorized as male or female (reference category). Caregiver-reported household income was a measure of prior year household income collapsed into categories of $0 to $30,000, $30,001 to $60,000, and $60,001 and greater (reference category). Caregiver marital status was a dichotomous measure of whether they were currently married or not (yes=1). Caregiver education was a measure of the primary caregiver’s level of educational attainment that was categorized as less than a high school degree, high school degree or GED, some college, bachelor’s degree, or graduate/professional degree (reference category).

Youth clinical factors included pubertal development, which was measured using sex-specific scales of perceived pubertal development (41). Both sexes reported on growth spurt in height, growth of pubic hair, and skin changes; males reported on voice change and growth of facial hair while females reported on breast growth and menstruation onset. Item response options ranged from 1 (no development) to 4 (development complete). Item scores were summed and averaged to create a mean total score for youth who responded to 50% or more of the items; youth who did not respond to at least 50% of the items were set to missing on the total score. Body mass index (BMI) was calculated according to the updated Centers for Disease Control and Prevention child and adolescent BMI guidelines (42). Objective height and weight were both measured twice by a trained interviewer, and a third time if the results were discrepant; the mean of the height and weight measures were calculated prior to the BMI calculation. Based on prior research examining relationships between loneliness and cortisol, if the caregiver reported the youth had ever been diagnosed with any depressive or anxiety disorder; attention deficit disorder/attention deficit hyperactivity disorder (ADD/ADHD); any behavioral, oppositional or conduct disorder; or autism, Asperger’s, pervasive development disorder or other autism spectrum disorder, youth were categorized as affirmative (yes=1) to having a lifetime history of a mental health diagnosis. Youth who were reported by their caregiver to take prescription medication, such as antidepressants or anti-anxiety medications (e.g., Zoloft, Prozac or Celexa) or ADD/ADHD medications (e.g., Ritalin or Adderall) were categorized as affirmative to current psychotropic medication use. An open-ended question was also asked of caregivers regarding other youth prescription medication use; responses were evaluated by a registered nurse. Responses indicating use of prescribed selective serotonin reuptake inhibitors, other antidepressants, anxiolytics, antipsychotics, stimulants, or other psychotropic medications were also coded as affirmative (yes=1). Last, caregivers reported on youth corticosteroid use via four items querying on oral, nasal, inhaled and/or topical use. Youth who were reported to use one or more corticosteroids on these four questions or on the open-ended question as described above were coded as affirmative (yes=1).

Youth hair care practices that can affect HCC included chemical treatments (e.g., hair dye, bleach, chemical straighteners, or perms) in the 3 months prior to data collection (yes=1) and frequency of hair washing (daily washing yes=1). Additionally, HCC can be affected by season of collection and assayed hair length. Season of collection was categorized as autumn, winter, spring, and summer (reference category). Hair length in centimeters used for the cortisol assay was measured by the laboratory technicians and natural log transformed due to the skewed distribution.

2.4. Analytic Sample and Strategy

A total of 1102 youth had a reliable hair sample and assay result available for statistical analysis out of the total 1405 youth in the first wave of the AHDC study. Reasons for non-inclusion included: caregiver and/or youth did not want to provide a hair sample (n=94); initial ineligibility due to reported corticosteroid use on prescreening survey (n=27) or insufficient hair to cut (n=49); assay issues, including sample results outside of the standard curve (n=62) or high variability in results between the duplicate samples (n=15); the cut hair sample was not of sufficient volume to assay (n=21); youth did not complete the second visit that included the biomarker collection (n=14); id issues matching sample to survey, duplicate ids, or missing id (n=12); and reported in the survey to have provided a hair sample, but no sample was returned to the lab (n=9).

In the statistical analysis, participant sociodemographic characteristics were analyzed using descriptive statistics and stratified by sex (Table 1). We then conducted a series of linear regression analyses to examine the associations between the two measures of loneliness (standardized for analysis) and logged HCC. Four models for each of the loneliness measures were examined: (1) bivariate associations between loneliness and HCC; (2) associations between loneliness and HCC, controlling for youth age, youth race and ethnicity, youth sex, household income, caregiver education, caregiver marital status, youth pubertal development, youth BMI, season of collection, corticosteroid use, hair care practices (chemical use and daily hair washing), and length of hair assayed; (3) associations between loneliness and HCC, controlling for the covariates in model two along with youth lifetime mental health diagnosis and current psychotropic medication use; and (4) associations between loneliness and HCC, controlling for the covariates in model 3, and the interaction term between male sex and loneliness. Analyses were conducted using SAS, version 9.4 (Cary, NC); missing responses to the independent variables were examined and multiple imputation of independent variables was conducted using Proc MI (25 multiple imputations used for all analyses). As noted, we conducted the above analysis on youth who were reported on the survey to be taking corticosteroids and also with these youth removed from the analysis. The coefficients, standard errors, and p-values were consistent between both approaches thus we present the results that include the youth reported to be taking corticosteroids, controlling for their use in the multivariable analysis (N=1102), and include the results with youth taking corticosteroids excluded from analysis in the appendix (N=1000).

Table 1.

Characteristics of analytic sample and stratified by sex

	Total (N=1102)		Female (n=584)		Male (n=518)
	n	M (SD)/%	n	M (SD)/%	n	M (SD)/%
Logged hair cortisol concentration	1102	1.35 (1.09)	584	1.30 (1.13)	518	1.40 (1.04)
Youth perceived loneliness (survey)	1071	1.79 (0.79)	574	1.88 (0.82)	497	1.68 (0.74)
Youth perceived loneliness (EMA)	1071	−0.02 (2.1)	571	0.21 (2.14)	500	−0.29 (2.02)
Youth age	1102	14.3 (1.87)	584	14.30 (1.90)	518	14.2 (1.83)
Youth race and ethnicity
Non-Hispanic White	547	49.7%	289	49.5%	258	49.8%
Non-Hispanic Black or Black mixed race	439	39.8%	238	40.7%	201	38.8%
“Other” race and ethnicity	116	10.5%	57	9.8%	59	11.4%
Household income
Under $30,000	362	32.9%	190	32.5%	172	33.2%
$30,0001-$60,000	253	23.0%	133	22.8%	120	23.2%
$60,000 and higher	401	38.3%	171	32.0%	201	38.8%
Missing	64	5.8%	39	6.7%	25	4.8%
Caregiver education
Less than High School	57	5.2%	30	5.1%	27	5.2%
High School Degree or GED	163	14.8%	84	14.4%	79	15.3%
Some College	382	34.6%	204	34.9%	178	34.4%
Bachelor's Degree	281	25.5%	155	26.5%	126	24.3%
Graduate Degree	208	18.9%	106	18.2%	102	19.7%
Missing	11	1.0%	5	0.9%	6	1.1%
Caregiver married
Yes	600	54.5%	309	52.9%	291	56.2%
No	482	43.7%	265	45.4%	217	41.9%
Missing	20	1.8%	10	1.7%	10	1.9%
Youth lifetime mental health diagnosis
Yes	295	26.8%	156	26.7%	139	26.8%
No	792	71.9%	420	71.9%	372	71.8%
Missing	15	1.3%	8	1.4%	7	1.4%
Youth current psychotropic medication use
Yes	169	15.3%	81	13.9%	88	17.0%
No	910	82.6%	492	84.2%	418	80.7%
Missing	23	2.1%	11	1.9%	12	2.3%
Youth current corticosteroid use
Yes	67	6.1%	33	5.6%	34	6.6
No	1000	90.7%	533	91.3%	467	90.1
Missing	35	3.2%	18	3.1%	17	3.3
Youth pubertal development	1026	3.0 (0.73)	542	3.26 (0.65)	484	2.73 (0.66)
Youth body mass index	1072	23.8 (6.43)	569	24.0 (6.36)	503	23.5 (6.49)
Logged hair length	1063	1.0 (0.48)	562	1.15 (0.40)	501	0.86 (0.51)
Season of data collection
Summer	306	30.6%	155	29.1%	161	31.1%
Autumn	255	25.5%	140	26.2%	129	24.9%
Winter	222	22.2%	115	21.6%	117	22.6%
Spring	217	21.7%	123	23.1%	111	21.4%
Youth washes hair daily
Yes	377	34.2%	159	27.2%	218	42.1%
No	678	61.5%	406	69.5%	272	52.5%
Missing	47	4.3%	19	3.3%	28	5.4%
Youth hair chemical use past 3 months
Yes	163	17.1%	166	28.4%	22	4.2%
No	883	80.1%	404	69.2%	479	92.5%
Missing	31	2.8%	14	2.4%	17	3.3%

3. Results

3.1. Descriptive Characteristics

Table 1 presents all descriptive characteristics for the total analytic sample and stratified by sex. The adolescents in the sample for this analysis (N=1102) had a mean age of 14.3 years (SD=1.9), 53% were female, and most identified as either non-Hispanic White (49.7%) or non-Hispanic Black or Black mixed race (39.8%). A total of 32.9% of the adolescents lived in a household with an annual income of less than $30,000, 23% with an annual income between $30,001 – $60,000, and 38.3% with an annual income of $60,001 and higher. Approximately 54.5% of the caregivers were married and 5.2% reported having less than a high school degree, 14.8% had a high school degree or GED, 34.6% reported completion of some college coursework, 25.5% reported having a bachelor’s degree and 18.9% reported having a graduate or professional degree.

The mean logged HCC was 1.35 pg/mg (SD=1.1) for the total analytic sample, 1.3 pg/mg for females (SD=1.13), and 1.4 pg/mg (SD=1.04) for males. Independent-samples t test revealed no significant differences in HCC between males and females (t(1100)= −1.57; p=.116). The mean for the unstandardized survey loneliness measure was 1.79 (SD=0.79, range 1–5) for the total analytic sample, 1.88 (SD=0.82, range 1–5) for females and 1.68 (SD=0.74, range 1–5) for males. The unstandardized mean for the EMA loneliness measure was − 0.02 (SD=2.1, range −2.33–8.36) for the total analytic sample, 0.21 (SD=2.14, range −2.11–8.36) for females and −0.29 (SD=2.02, range −2.33–6.63) for males. Independent-samples t test revealed significant differences in both the standardized survey and the EMA loneliness measures between males and females (survey: t(1067.9)=4.34; p<.0001; EMA: t(1063.4) = 3.96; p<.0001 with males reporting less loneliness than females.

3.2. Linear Regression Results

Results of the linear regression analyses for the standardized survey and EMA loneliness measures are presented in Table 2. In model one testing the bivariate linear relationship between loneliness and HCC, higher loneliness via survey and EMA measures was associated with lower HCC (Survey: b= − 0.10, SE=0.03, p=.004; EMA: b= − 0.09, SE=0.03, p=.005). In model two, higher loneliness remained significantly associated with lower HCC (Survey: b= − 0.07, SE=0.03, p=.023; EMA: b= − 0.07, SE=0.03, p=.037), after controlling for the following covariates: sociodemographic factors, pubertal development and BMI, corticosteroid use, hair care practices, season of collection and assayed hair length. In model 3, in which youth lifetime mental health diagnosis and current psychotropic medication use were added into the regression model, higher loneliness remained significantly associated with lower HCC (Survey: b= − 0.07, SE=0.03, p=.029; EMA: b= − 0.07, SE=0.03, p=.039). As both loneliness measures were standardized, the results can be interpreted as a one standard deviation in increase in loneliness is associated with a 7% decrease in HCC. Last, the interaction between loneliness and sex was examined; no significant differences in the relationship between loneliness and HCC were found between male and female adolescents, accounting for all covariates (Survey: b=0.04, SE=0.06, p=.552; EMA: b= − 0.01, SE=0.06, p=.843).

Table 2.

Associations between loneliness and hair cortisol concentration among adolescents aged 11–17 years (N=1102)

	Model 1	Model 2	Model 3	Model 4
	b (SE) p value	b (SE) p value	b (SE) p value	b (SE) p value
Associations between Survey-Measured Loneliness and HCC
Main Effects
Loneliness (survey)	− 0.10 (.03) .004	− 0.07 (0.03) .023	− 0.07 (0.03) .029	−0.09 (0.04) .032
Male sex				− 0.05 (0.08) .527
Interaction
Loneliness (survey)*Male				0.04 (0.06) .552
Intercept	1.35 (0.03) <.0001	2.10 (0.29) <.0001	2.10 (0.29) <.0001	2.09 (0.29) <.0001
Associations between EMA-Measured Loneliness and HCC
Main Effects
Loneliness (EMA)	− 0.09 (0.03) .005	− 0.07 (0.03) .037	− 0.07 (0.03) .039	− 0.06 (0.04) .150
Male sex				− 0.05(0.08) .476
Interaction
Loneliness (EMA)*Male				− 0.01 (0.06) .843
Intercept	1.35 (0.03) <.0001	2.04 (0.29) <.0001	2.05 (0.29) <.0001	2.05 (0.29) <.0001

Notes:

Loneliness measures are standardized for bivariate and multivariable analysis, mean =0 and sd= 1

Model 1=bivariate analysis

Model 2= multivariable analysis with loneliness and covariates including sociodemographic characteristics (youth sex, age, race and ethnicity, household income, caregiver educational attainment and marital status), clinical factors (pubertal development, BMI, corticosteroid use), hair care practices (chemical use, daily hair washing), and study procedures (assayed hair length, season of collection)

Model 3= multivariable analysis including loneliness and the covariates in model 2 along with youth lifetime mental health diagnosis and current psychotropic medication use

Model 4= multivariable analysis with loneliness and covariates in model 3 along with the interaction term loneliness*male sex

4. Discussion

This analysis is one of the first investigations of the relationships between perceived loneliness and HCC in adolescents. Findings demonstrate that greater loneliness measured via survey (perceived loneliness in general) and EMA methods (average perceived loneliness over five-day EMA collection) were both associated with lower HCC in linear analysis, and not moderated by sex. As noted, cortisol concentration generally increases in response to an acute stressor and then declines when the threat dissipates. However, ample research has also indicated that some stressors and health conditions are associated with lower cortisol concentration (26, 43). For example, child maltreatment, post-traumatic stress disorder, and functional somatic disorders (e.g. fibromyalgia, chronic fatigue syndrome, myalgic encephalomyelitis) have been linked to lower cortisol concentration, including lower HCC in child, adolescent, and adult samples (39, 44). Researchers hypothesize that exposure to stressors that are prolonged or recurrent or when the response to a stressor remains heightened (e.g., anticipatory, rumination), cortisol concentration may become dampened over time possibly due to alterations in glucocorticoid receptor sensitivity (26, 43). Longitudinal research is needed to test these hypotheses, including heterogeneity by contextual (e.g., prior and current adverse exposures, social support) as well as individual-level (e.g., coping mechanisms, pubertal timing, sex hormone) factors. Although our study did not find the relationship between loneliness and HCC differed for male and females, examination of sex hormones as well as pubertal timing (e.g., early or late versus on time developmentally for age) as potential moderators or as statistical controls would advance understanding of these complex relationships compared to only the male/female dichotomy (45, 46).

As noted, there is a dearth of research investigating perceived loneliness and HCC, and the one study we are aware of, found null results (25). Others have examined relationships between “trait” loneliness (measured as perceived loneliness in general) and patterns of the cortisol diurnal curve using saliva. In Doane and Adam’s study of 108 adolescents 17 to 19 years of age, trait loneliness was associated with flatter diurnal cortisol slopes (40) whereas Jopling et al. reported greater loneliness was associated with higher cortisol at waking and a blunted cortisol awakening response (CAR) in a sample of 52 adolescents aged 13 to 14 years with collection taking place during the initial Covid–19 lockdown (23). Although our findings cannot be directly compared to these studies due to the differences in cortisol measures and timing, combined they suggest perceived loneliness may be associated with HPA dysregulation in adolescents. In several small validation studies (47–49), researchers found cortisol concentration in one centimeter of hair was moderately correlated with the averaged cortisol area-under-the curve measures (AUC_dayc, AUC_ground) but not the averaged CAR nor diurnal slope in saliva collected daily for 30 days. Thus, the researchers posit that HCC may be more comparative to AUC measures of the total cortisol secretion throughout the day versus diurnal variation (50). While the Jopling et al study (23) examined relationships between perceived loneliness and salivary cortisol AUC, they did not find a significant relationship. However, they only collected salivary cortisol measures over two days and in a small sample (N=52). Thus, if feasible, future studies should consider examining the effects of loneliness on HCC and on salivary AUC measured longitudinally and in larger samples to facilitate comparison and advance understanding of these complex relationships.

Several additional limitations of this analysis warrant further discussion. First, we conducted a cross-sectional analysis, thus causality cannot be inferred. Moving forward, longitudinal investigations are needed to better assess the chronicity of loneliness and the potential of long-term effects on HPA activity and health. Moreover, a longitudinal design will enable examination of reciprocal relationships as cortisol dysregulation and downstream neuroinflammation may increase sickness feelings, including anhedonia and self-isolation (51), that may contribute to perceived loneliness. Second, sex data were collected without information accounting for gender identity, which could have implications for socialized perceptions of loneliness, stress experiences, sex hormones, and cortisol. Finally, this sample of adolescents reported low levels of loneliness overall on both the 9-item questionnaire and the EMA loneliness measure which limits our understanding of the effects of more severe loneliness.

Despite these limitations, our study has numerous strengths including the large sample size, the inclusion of adolescents spanning 11 to 17 years of age, near equal representation of male and female adolescents, and racial (Black and White) and socioeconomic diversity in the sample. Our findings that perceived overall loneliness using two different measurement strategies was associated with lower HCC highlights the need for further research of loneliness in adolescence. The 2023 Surgeon General’s Advisory (12) on loneliness emphasized the need for further mechanistic research to identify the pathways through which loneliness affects health across the life span. For adolescents, evidence is robust on the links between loneliness and mental health conditions (16, 52), and further research exploring HPA dysregulation as a mediator and/or a moderator (e.g., adolescents with dysregulated stress responses may be more sensitive to predictors and more likely to develop anxiety or depression) of these relationships. Longitudinal research on the predictors of loneliness trajectories (e.g., transient versus increasing or chronic), potential stress related mechanisms, and their subsequent effects on mental health, including potential heterogeneity across these pathways will facilitate understanding of these complex relationships and elucidate avenues for future intervention.

Highlights.

• Adolescents experience high levels of loneliness, which is linked to poor health in adulthood

• Survey and ecological momentary assessment measures of perceived loneliness were associated with lower hair cortisol concentration (HCC) in adjusted models

• Sex did not moderate the relationship between perceived loneliness and HCC

• HCC may be an efficient method for future investigation of chronic loneliness and stressors associated with HPA-axis dysregulation in adolescence

Appendix I.

We conducted sensitivity analysis to examine the relationships between the standardized survey and EMA loneliness measures and HCC excluding youth whose caregiver reported they were taking corticosteroid medications.

Associations between loneliness and hair cortisol concentration among adolescents aged 11–17 years (excluding youth reported to be taking corticosteroids) (N=1,000)

	Model 1	Model 2	Model 3	Model 4
	b (SE) p value	b (SE) p value	b (SE) p value	b (SE) p value
Associations between Survey-Measured Loneliness and HCC
Main Effects
Loneliness (survey)	− 0.09 (.04) .007	− 0.07 (0.03) .041	− 0.07 (0.04) .044	−0.08 (0.04) .067
Male sex				− 0.07 (0.08) .340
Interaction
Loneliness (survey)*Male				0.02 (0.06) .756
Intercept	1.35 (0.03) <.0001	2.09 (0.30) <.0001	2.09 (0.30) <.0001	2.08 (0.29) <.0001
Associations between EMA-Measured Loneliness and HCC
Main Effects
Loneliness (EMA)	− 0.10 (0.03)  .003	− 0.08 (0.03) .015	− 0.08 (0.03) .015	− 0.08, (0.05) .084
Male sex				− 0.09 (0.08) .281
Interaction
Loneliness (EMA)*Male				− 0.01 (0.07) .849
Intercept	1.35 (0.03) <.0001	2.06 (0.30) <.0001	2.07 (0.30) <.0001	2.00 (0.31) <.0001

Notes:

Loneliness measures are standardized for bivariate and multivariable analysis, mean =0 and sd= 1

Model 1=bivariate analysis

Model 2= multivariable analysis with loneliness and covariates including sociodemographic characteristics (youth sex, age, race and ethnicity, household income, caregiver educational attainment and marital status), clinical factors (pubertal development, BMI), hair care practices (chemical use, daily hair washing), and study procedures (assayed hair length, season of collection)

Model 3= multivariable analysis including loneliness and the covariates in model 2 along with youth lifetime mental health diagnosis and current psychotropic medication use

Model 4= multivariable analysis with loneliness and covariates in model 3 along with the interaction term loneliness*male sex