Curvilinear association between loneliness and hair cortisol concentrations in community-dwelling older adults at elevated cardiovascular risk Authors: Xinyi Zhu, Xiaomei Liu, Boyu Zhai, Xuliang Hou, Lijuan Huo, Juan Li Published: 2026 ABSTRACT Loneliness, a distressing subjective experience that may become a chronic psychosocial stress exposure when persistent, has been linked to dysregulation of the hypothalamic-pituitary-adrenal (HPA) axis, yet empirical findings linking loneliness to cortisol regulation remain inconsistent. One largely unexplored possibility is that loneliness may exert non-linear association on HPA-axis activity. This study examined the curvilinear association between loneliness and hair cortisol concentration (HCC) in community-dwelling older adults. Participants were 126 older adults aged ≥60 years at elevated cardiovascular risk, drawn from the baseline assessment of a randomized controlled trial in China. Loneliness was assessed using the 6-item De Jong Gierveld Loneliness Scale. HCC was quantified using liquid chromatography-tandem mass spectrometry. Linear regression models tested quadratic associations between loneliness and log-transformed HCC, adjusting for demographic, methodological, psychological, social, and health-related covariates. Loneliness showed a significant curvilinear association with HCC, with HCC appearing higher at low-to-moderate loneliness and lower at higher loneliness. The quadratic term remained significant after full covariate adjustment, but sensitivity analyses indicated that statistical significance was sensitive to the exclusion of high-influence cases. These preliminary findings suggest that loneliness may be nonlinearly associated with long-term cortisol output among older adults at elevated cardiovascular risk. The findings may help contextualize prior inconsistencies in the loneliness-cortisol literature and highlight the need for longitudinal studies to clarify temporal and mechanistic pathways. Introduction Loneliness is commonly defined as the distressing subjective experience that arises when perceived social needs are not met by the quantity or quality of social relationships [1]. When persistent, this perceived social disconnection may constitute a chronic psychosocial stress exposure with implications for physiological dysregulation and health risk in later life [2]. This issue is particularly relevant for community-dwelling older adults, for whom changes in social roles, bereavement, and health limitations may increase vulnerability to loneliness and stress-related health risk. While accumulating evidence links loneliness to dysregulation of the hypothalamic-pituitary-adrenal (HPA) axis, empirical results empirical results from salivary cortisol studies have been inconsistent. For example, some studies report heightened cortisol awakening responses (CAR) among lonelier individuals [3], whereas others observe blunted CAR [4,5] or null associations [6]. These discrepancies are often attributed to differences in study characteristics, including age, sex, measurement protocols, and contextual factors. An additional, largely untested possibility is that loneliness does not exert a uniform linear influence on HPA-axis function. From an allostatic load perspective [7], mild-to-moderate loneliness may be accompanied by heightened social vigilance and physiological arousal related to social reconnection, potentially reflected in increased cortisol output or reactivity [8]. Conversely, higher or more persistent loneliness may be associated with HPA-axis downregulation or hypoactivation, a pattern described in some forms of chronic stress adaptation [9]. Such non-linear patterns may reflect a transition from active social monitoring to biological wear-and-tear, yet curvilinear associations between loneliness and cortisol regulation have rarely been examined. Hair cortisol concentration (HCC) is a cumulative biomarker that captures long-term glucocorticoid output over weeks to months, providing a stable marker of HPA-axis activity [10]. This cumulative index is particularly suited for capturing the sustained biological embedding of chronic psychosocial distress. Although loneliness has been linked to altered diurnal cortisol rhythms and cortisol reactivity, evidence regarding its association with HCC remains sparse. In older adults, recent evidence suggests that loneliness may be positively associated with HCC in men but not women [11], whereas studies in younger samples or pandemic contexts have reported lower HCC among lonelier individuals or null associations [12,13]. These mixed findings suggest that loneliness-HCC associations may depend on sample characteristics and may not follow a simple linear pattern. The present study examined whether loneliness exhibits a quadratic association with HCC in a sample of community-dwelling older adults at elevated cardiovascular risk. We hypothesized a curvilinear relationship, with hair cortisol, whereby HCC would be higher at low-to-moderate levels of loneliness and lower at higher loneliness levels. Methods Study design and participants Data were drawn from the baseline assessment of the randomized controlled trial Ameliorate Cardiovascular Condition through Positive Psychology and Lifestyle Intervention in China (ACCOMPLI-CH; Chinese Clinical Trial Registry: ChiCTR2200062929). Full protocol details are reported elsewhere [14]. Briefly, the trial enrolled 1047 community-dwelling older adults ≥60 years at elevated cardiovascular risk from five metropolitan cities across China. Because of budgetary and logistical constraints, hair cortisol assessment was conducted in a subsample of the parent trial as an ancillary assessment of chronic neuroendocrine markers, mainly at the Beijing and Nanchang sites, with a small number of additional samples from Chongqing. Hair collection was not conducted for individuals currently using glucocorticoids or with recent hair perming or dyeing. Details on eligibility criteria, participant flow, invalid samples, and reasons for nonparticipation in hair collection are provided in Supplementary Materials A1 and A2. The analytic sample comprised 126 older adults who provided valid hair samples. To evaluate potential selection into the hair subsample, we compared participants with valid hair samples with participants without hair samples both in the full parent trial sample and within the two main hair-collection sites. Results of these comparisons are presented in Supplementary Materials A3 and Table S1. In the full parent trial sample, participants with valid hair samples differed from those without hair samples on several demographic and psychosocial characteristics. Within the two main hair-collection sites, the hair subsample was older, more likely to be female, reported better self-rated health, and was less likely to have diabetes, but did not differ in loneliness or key psychological and social variables as well as other health-related variables. The study protocol was approved by the Institutional Review Board of the Institute of Psychology, Chinese Academy of Sciences (H22070). Written informed consent was obtained from all participants. Measures Independent variable: loneliness Loneliness was conceptualized as a subjective discrepancy between desired and actual social relationships and was assessed using the 6-item De Jong Gierveld Loneliness Scale [15]. The scale includes items capturing emotional and social loneliness. Items were scored according to the scale guidelines to yield a total score ranging from 0 to 6, with higher scores indicating greater loneliness. Given the restricted range of the two 3-item subscales and the modest sample size, subscale-specific analyses were not conducted. Internal consistency in the present sample was limited (Cronbach's alpha = 0.61). Dependent variable: hair cortisol concentration Hair samples were collected from the posterior vertex region of the scalp, as close to the scalp as possible. The proximal 1 cm hair segment was used for cortisol analysis; for a small number of participants, proximal segments of up to 2–3 cm were analyzed. Hair samples were processed following established protocols [16]. Briefly, hair samples were washed, dried, and extracted in methanol, and cortisol concentrations were quantified using liquid chromatography-tandem mass spectrometry (LC-MS/MS) at the Laboratory of Prof. Huihua Deng, Southeast University. The analytical method showed good performance, with intra- and inter-day coefficients of variation below 15% and recovery rates ranging from approximately 85% to 115%. To account for variation in sample characteristics, cortisol values were corrected for hair weight and hair segment length. The target hair weight for analysis was 20 mg, and cortisol concentrations were normalized by sample weight and expressed as pg/mg. Hair length correction was applied to account for potential leaching effects associated with differences in hair segment length. In addition, corrections for limits of detection (LOD) and quantification (LOQ) were applied: values below the LOD were set to zero, and values between the LOD and LOQ were set to the LOD value. HCC values were right-skewed and therefore natural log-transformed prior to statistical analyses. After log transformation, no values exceeded ±3 standard deviations (SDs) from the mean, and no observations were excluded on this basis. Covariates Covariates were selected based on prior research on determinants of hair cortisol and loneliness [10,12]. Hair sample-related covariates included season of collection (spring/winter vs. summer) and study site (Nanchang/Chongqing vs. Beijing). Demographic factors included age, sex, and years of education. Psychological covariates included depressive symptoms assessed with the Patient Health Questionnaire-9 and perceived stress assessed with the 14-item Perceived Stress Scale, as both are closely related to loneliness and HPA-axis regulation. Health-related factors previously linked to hair cortisol included body mass index (BMI) and sleep quality (Pittsburgh Sleep Quality Index). Social connection variables included social network size (Lubben Social Network Scale 6 items) and perceived social support (Multidimensional Scale of Perceived Social Support). Current smoking status was assessed by self-report and coded as current smoker versus non-current smoker. Further details of these measures are provided in Supplementary Materials A4. Statistical analysis Linear regression models were conducted to examine the curvilinear association between loneliness and hair cortisol. Loneliness scores were mean-centered prior to analysis, and a quadratic term was computed to test non-linear effects. Log-transformed hair cortisol was the dependent variable. Three hierarchical models were specified. Model 1 adjusted for basic demographic and methodological covariates, including age, sex, study site, and season of sampling. Model 2 added loneliness (linear term) and its quadratic term. Model 3 further adjusted for psychological covariates, social connection measures and health-related factors, including depressive symptoms, perceived stress, social network size, perceived social support, sleep quality, body mass index, current smoking status, and years of education. Sex-by-loneliness interaction terms were additionally tested to examine potential sex differences. Model assumptions were checked using residual diagnostics, influence statistics, and multicollinearity diagnostics. Primary analyses were conducted in a restricted sample excluding participants with hair samples weighing less than 5 mg (n = 120). Sensitivity analyses included repeating the models in the full sample (n = 126) and excluding high-influence cases identified by Cook's distance. In addition, the robustness of regression coefficients was examined using bias-corrected and accelerated (BCa) bootstrap confidence intervals with 2000 resamples. Statistical analyses were conducted using IBM SPSS version 22 (IBM, Corp., Armonk, NY). Figures were generated using the ggplot2 package in R [17]. Results Descriptive results Participants (71.63 ± 4.33 years; 84.1% female) showed a mean loneliness score of 1.82 (SD = 1.50). Raw HCC ranged from 0.86 to 85.10 pg/mg (20.90 ± 17.47 pg/mg), with a log-transformed mean of 2.69 pg/mg (SD = 0.89). Descriptive statistics for psychological, sleep, and social variables are also presented in Table S2. Overall, the sample showed relatively low depressive symptoms and perceived stress. Preliminary correlations showed that loneliness was positively correlated with depressive symptoms (r = 0.43, p < 0.001), supporting its overlap with psychological distress, whereas its linear correlation with logged HCC was not significant (Table S3). Regression results Table 1 presents the regression results examining the association between loneliness and HCC. As shown in Fig. 1, loneliness showed a significant curvilinear association with HCC, indicated by a negative quadratic term after adjustment for demographic and methodological covariates (b = −0.063, p = 0.045), while the linear term was not significant. Adjustment for depressive symptoms, perceived stress, and social connection variables did not attenuate the effect (quadratic term: b = −0.074, p = 0.019). The estimated peak occurred at low-to-moderate levels of loneliness (approximately 1.5–1.7 on the 0–6 scale). No significant interactions with sex were observed. Model diagnostics indicated no serious violations of regression assumptions, including multicollinearity, with all VIF values < 1.82 and tolerance values > 0.55 (Supplementary Materials A5).Table 1Regression results of the relationship between loneliness and hair cortisol concentration.Table 1PredictorModel 1 (basic adjustment)Model 2 (loneliness model)Model 3 (full adjustment)bSEpbSEpbSEpAge0.0120.0170.5020.0130.0170.4560.0230.0170.191Sex (female = 0)−0.2080.2220.350−0.2390.2190.276−0.2970.2430.224Site (NC = 0)0.3450.1760.0530.3280.1730.0610.2580.2010.202Season (spring = 0)0.8690.172<0.0010.8890.169<0.0010.8540.175<0.001Loneliness−0.0240.0550.664−0.0130.0640.842Loneliness (quadratic term)−0.0630.0310.045−0.0740.0310.019Depressive symptoms0.0350.0410.396Perceived stress−0.0160.0130.220Social network size0.0040.0130.742Perceived social support0.0070.0070.335Sleep quality index0.0410.0210.053BMI0.0490.0220.026Current smoker0.3410.3850.377Education years−0.00010.0310.996Adjusted R20.1590.1890.235Notes: b = unstandardized coefficients, SE = standardized error. BMI, body mass index.Fig. 1Curvilinear association between loneliness and natural log-transformed hair cortisol concentration in older adults at elevated cardiovascular risk in the primary analytic sample (n = 120). The solid line represents the fitted quadratic regression, and the shaded area indicates the 95% confidence interval. Dots represent individual observations and are horizontally jittered for visualization only. Table 1Regression results of the relationship between loneliness and hair cortisol concentration.Table 1PredictorModel 1 (basic adjustment)Model 2 (loneliness model)Model 3 (full adjustment)bSEpbSEpbSEpAge0.0120.0170.5020.0130.0170.4560.0230.0170.191Sex (female = 0)−0.2080.2220.350−0.2390.2190.276−0.2970.2430.224Site (NC = 0)0.3450.1760.0530.3280.1730.0610.2580.2010.202Season (spring = 0)0.8690.172<0.0010.8890.169<0.0010.8540.175<0.001Loneliness−0.0240.0550.664−0.0130.0640.842Loneliness (quadratic term)−0.0630.0310.045−0.0740.0310.019Depressive symptoms0.0350.0410.396Perceived stress−0.0160.0130.220Social network size0.0040.0130.742Perceived social support0.0070.0070.335Sleep quality index0.0410.0210.053BMI0.0490.0220.026Current smoker0.3410.3850.377Education years−0.00010.0310.996Adjusted R20.1590.1890.235Notes: b = unstandardized coefficients, SE = standardized error. BMI, body mass index. Regression results of the relationship between loneliness and hair cortisol concentration. Notes: b = unstandardized coefficients, SE = standardized error. BMI, body mass index. Curvilinear association between loneliness and natural log-transformed hair cortisol concentration in older adults at elevated cardiovascular risk in the primary analytic sample (n = 120). The solid line represents the fitted quadratic regression, and the shaded area indicates the 95% confidence interval. Dots represent individual observations and are horizontally jittered for visualization only. Sensitivity analysis The quadratic estimate was generally consistent in direction across sensitivity analyses. Including participants with hair samples below 5 mg yielded a comparable effect (b = −0.066, p = 0.031). After excluding high-influence cases, the estimate remained similar in magnitude and direction (b = −0.060), but was no longer statistically significant (p = 0.074). Detailed sensitivity and bootstrap results are presented in Supplementary Materials B1, Table S4 and Fig. S1. Discussion The present study found a quadratic association between loneliness and hair cortisol concentration in community-dwelling older adults at elevated cardiovascular risk. HCC appeared higher at low-to-moderate levels of loneliness and lower at higher levels, but the upper end of the curve was estimated from relatively few highly lonely participants and should therefore be interpreted cautiously. The quadratic estimate was generally consistent in direction across covariate-adjusted and sensitivity analyses, although it did not reach statistical significance after excluding high-influence cases. To our knowledge, this is among the first studies to examine a nonlinear association between loneliness and long-term cortisol output in an aging sample. This pattern may help contextualize prior mixed findings linking loneliness to cortisol regulation. Mild-to-moderate loneliness may co-occur with heightened social vigilance, negative social expectations, and sustained monitoring of social cues, processes that have been linked to higher cortisol output [8]. At higher loneliness levels, lower HCC may be consistent with blunted or downregulated HPA-axis activity, as described in chronic stress adaptation [9]. However, the present data cannot determine whether the pattern reflects increased HCC at low-to-moderate loneliness followed by downregulation, or primarily lower cumulative cortisol output among those with higher loneliness. Given the cross-sectional design and the limited number of highly lonely participants, these interpretations remain speculative. Cortisol regulates inflammation through glucocorticoid signaling; thus, blunted HPA-axis activity or altered glucocorticoid sensitivity may impair inflammatory control [7], a pathway particularly relevant to older adults at elevated cardiovascular risk. Clinically, these findings support loneliness assessment as part of psychosocial risk profiling and, if replicated, suggest that loneliness-related biological dysregulation may not follow a simple linear pattern. Notably, the quadratic association between loneliness and HCC was not attenuated after adjustment for depressive symptoms, perceived stress, sleep quality, social network size, and perceived social support. This suggests that the observed pattern may reflect a specific social-affective dimension of loneliness rather than generalized psychological distress, sleep-related problems, or objective social disconnectedness. Consistent with this distinction, prior work has shown that perceived emotional closeness to social ties, but not objective network size, was associated with HCC in older adults [18], highlighting the importance of subjective social experience in shaping long-term HPA-axis activity. At the same time, depressive symptoms, perceived stress, and sleep disturbance may still overlap with loneliness or represent related processes in loneliness-related stress physiology. These possibilities require longitudinal testing. Several limitations should be acknowledged. First, the cross-sectional design precludes causal inference and does not allow examination of transitions across phases of HPA-axis activation as loneliness persists or intensifies over time. Second, the modest sample size limited the precision of the quadratic estimate. Although sensitivity and bootstrap analyses yielded generally consistent negative estimates, statistical significance was sensitive to analytic specification, and the BCa confidence interval marginally included zero. Relatedly, few participants reported very high loneliness, limiting precision in estimating the upper end of the curve. The small sample size and limited representation of men also constrained analyses of sex differences. Moreover, participants were drawn from an RCT targeting older adults at elevated cardiovascular risk, which may limit generalizability to healthier older adults and should be considered when interpreting the HCC pattern. HCC was assessed in a subsample that was not fully representative of the parent trial. Although site-restricted comparisons showed no differences in loneliness, depressive symptoms, perceived stress, sleep quality, or social network size between participants with and without HCC data, differences in age, sex, self-rated health, and diabetes status were observed. Individual-level reasons for nonparticipation in hair collection were also not systematically recorded. Thus, potential selection bias cannot be ruled out, and the findings should be generalized with caution. Finally, the restricted score range and limited internal consistency of the 6-item DJGLS, possibly related to its short length and low loneliness levels in this subsample, may have increased measurement error and reduced sensitivity to detect fine-grained nonlinear patterns, particularly at extreme levels of loneliness. Overall, these findings suggest that loneliness may be nonlinearly associated with long-term cortisol output among older adults at elevated cardiovascular risk. This preliminary evidence may help contextualize mixed findings in the loneliness–cortisol literature and highlights the need for longitudinal studies to clarify temporal and mechanistic pathways. Declaration of generative AI and AI-assisted technologies in the manuscript preparation process During the preparation of this work the authors used ChatGPT 5.4 and Gemini 3 to refine the language and improve the overall readability of the manuscript. After using these tools, the authors reviewed and edited the content as needed and took full responsibility for the content of the published article. Funding sources This project was supported by the 10.13039/501100012166National Key Research and Development Program of China (2024YFC3606800), Scientific Foundation of the State Key Laboratory of Cognitive Science and Mental Health (E5CX1201GZ), 10.13039/501100014558Institute of Psychology, Chinese Academy of Sciences (E4JY451266), and the Philosophy and Social Science Project of Guangdong Province (GD22YXL03). CRediT authorship contribution statement Xinyi Zhu: Conceptualization, Formal analysis, Funding acquisition, Methodology, Writing – original draft, Writing – review & editing. Xiaomei Liu: Conceptualization, Data curation, Project administration, Writing – review & editing. Boyu Zhai: Data curation, Project administration, Writing – review & editing. Xuliang Hou: Methodology, Writing – review & editing. Lijuan Huo: Funding acquisition, Writing – review & editing. Juan Li: Funding acquisition, Supervision, Writing – review & editing. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.