ABSTRACT
There is growing interest in studying how habitual sleep disturbance affects biological risk factors that may underscore adverse health outcomes. This study examined associations between hair cortisol concentrations and self‐reported sleep quality and objectively measured sleep metrics derived using actigraphy. Data were collected from 306 female and 177 male adults, aged 18–79 years. Hair cortisol was analysed from 3‐cm proximal hair segments from the head to represent cortisol accumulation over approximately 90 days. Sleep quality measures included Pittsburgh Sleep Quality Index (PSQI) scores and five actigraphy‐derived metrics: sleep latency, total sleep time, wake after sleep onset, sleep efficiency and awakening bouts. In the fully adjusted multiple regression model, higher hair cortisol concentrations were associated with poor self‐reported sleep quality (i.e., PSQI > 5; p = 0.020), and higher mean PSQI scores (p = 0.027). No significant relationships were observed with actigraphy‐derived sleep measures. The findings support hair cortisol as a promising biomarker for evaluating chronic stress that often coincides with self‐reported sleep disturbance. The results suggest the importance of aligning time reference periods for biomarker and self‐reported outcomes and highlight the need for further research to reconcile discrepancies between subjective and objective sleep measures.
Keywords: actigraphy, hair cortisol, Pittsburgh Sleep Quality Index, sleep disturbance, stress, subjective sleep quality
1. Introduction
Sleep disturbance of a significant severity can result in biological changes involved in the progression toward long‐term adverse health effects (Chattu et al. 2018; Balbo et al. 2010). These biological changes can include stress reactions, evidenced through alterations that include cortisol, which can become dysregulated and promote pathophysiology through allostatic load (McEwen and Stellar 1993; McEwen 2006). Aligning with the large evidence base linking stressful life events and impaired sleep quality is the recent finding from a national survey among Canadian adults that stress/anxiety/worrying about something was by far the most prevalent factor attributed to sleep disturbance (Michaud et al. 2023).
A great deal of knowledge regarding objectively measured biological changes associated with sleep disturbance has been gleaned through studies conducted in laboratory settings where restricting sleep up to several hours has been a common intervention (Juster and McEwen 2015; Sivakumaran et al. 2023). However, mild sleep restriction (Benasi et al. 2024) or fragmented sleep that may occur more regularly can also increase measures of allostatic load and promote disease even if total sleep duration is not reduced Stamatakis and Punjabi (2010). Consistent with laboratory findings, self‐reported measures of sleep quality using the Pittsburgh Sleep Quality Index (PSQI), Insomnia Severity Index (ISI) and reported habitual sleep duration have also been associated with changes in biological markers of cardiovascular and metabolic health (Suarez 2008; Passos et al. 2023; Knutson 2010). With evidence linking sleep to biological stress responses, a noninvasive biomarker that reliably reflects long‐term stress would be highly valuable for investigating the sleep‐stress relationship.
A candidate biomarker in this context may be hair cortisol, which when secreted from the adrenal cortex following stressor‐induced hypothalamic–pituitary stimulation is reported to integrate and remain in hair with growth (Russell et al. 2012; Stalder and Kirschbaum 2012). As human scalp hair has a predictable average growth rate of ~1 cm/month (Wennig 2000), cortisol in hair can be noninvasively measured and used to retrospectively characterise cortisol levels over several months, particularly in situations where the stressor is sustained (Michaud et al. 2022). These advantages make hair cortisol analysis a promising methodology for evaluating the impact that long‐term sleep disturbance may have on the human stress response and associated illnesses, such as cardiovascular diseases (Juster and McEwen 2015; Sivakumaran et al. 2023; Jaspan et al. 2024).
The potential value of using hair cortisol as an indicator of biological risk from sleep disturbance is supported by recent research showing its positive association with insomnia and daytime sleepiness (Ahabrach et al. 2023; Colledge et al. 2017) and with depression among pregnant women who reported poor sleep quality on the PSQI prior to conception (Abdul Jafar et al. 2023). Fractured sleep among shift workers has been associated with elevated hair cortisol concentrations (Zhang et al. 2020) and positive correlations between hair cortisol concentrations and PSQI scores have been reported in a cross‐sectional study (Mazgelyte et al. 2023) and a longitudinal study among firefighters reporting trauma (Sopp et al. 2021). However, findings across studies are mixed. For example, although hair cortisol was related to perceived stress severity among nurses responding to the COVID‐19 pandemic, it was unrelated to their self‐reported sleep quality on the PSQI (Rajcani et al. 2021). Zhu et al. (2022) reported a statistically positive association between elevated PSQI scores and the cortisol metabolite, cortisone, but not with cortisol itself. Feller et al. (2014) reported higher hair cortisol levels among older individuals (mean age 65.8 years) who slept during the day, but this was no longer significant after adjusting for age and sex, and Lanfear et al. (2020) found no relation between day sleep and hair cortisol. Among children, Eythorsdottir et al. (2020) found no association between hair cortisol concentrations and objectively measured sleep using actigraphy.
Although some trends are beginning to emerge in the research, inconsistencies remain in the evidence linking hair cortisol and sleep quality. These discrepancies may be owing to how sleep was measured and variation in the degree to which studies could account for confounding, mediating, and/or moderating variables. It is also possible that the association between hair cortisol and sleep may be influenced by the time reference period reflected by hair cortisol, which is typically several weeks, and the period covered by sleep assessments. These uncertainties demonstrate that more research is required in this area.
1.1. Objectives
The objective of the current study was to evaluate hair cortisol concentrations in a population of adults that reported their subjective sleep quality on the PSQI and had multiple measures of sleep objectively determined using actigraphy.
2. Methods
2.1. Sample Design and Study Participants
Data collection took place through in‐person interviews between May and September 2013 from adults living in the Canadian provinces Ontario (ON) and Prince Edward Island (PE). One adult between the ages of 18 and 79 years was randomly selected from each household. The questionnaire instrument included modules on basic demographics, health effects, quality of life, sleep quality, diagnosed sleep disorders, perceived stress, lifestyle behaviours and prevalent chronic diseases. The final sample size in the ON and PE was 1011 and 227, respectively. Participants were not compensated in any way for their participation.
As part of the questionnaire, participants completed the PSQI, which provided an assessment of self‐reported sleep quality over the previous 30 days. The seven components of the PSQI are scored on a scale from 0 (better) to 3 (worse); resulting in a global PSQI score ranging between 0 and 21, where a value of greater than 5 is considered to represent poor sleep quality (Buysse et al. 1989).
2.2. Hair Sample Collection and Treatment
Michaud et al. (2016) provide information on the methodology for the collection and laboratory analysis of the hair samples. Briefly, when feasible and upon participant consent, hair samples were obtained from the vertex posterior of the scalp using scissors, cutting as close to the scalp as possible. In participants from whom a length of 3 cm or more of hair was collected, the 3 cm portion most proximal to the scalp was analysed. Hair sample collection and cortisol analysis were conducted in accordance with a previously established protocol described by Pereg et al. (2013). Resulting hair cortisol concentrations were obtained in nanograms of cortisol per gram of hair (ng/g).
2.3. Objectively Measured Sleep
All consenting and eligible participants aged 18–79 years who were expected to sleep at their current address for a minimum of three of the seven nights following the interview were provided with an Actiwatch2 (Philips Healthcare, Andover, MA) wrist‐worn actimeter. Instructions were to wear the device continuously for a consecutive 7‐day period following the interview.
Data analysis was conducted using Actiware version 5.70 with the software set to default settings. The devices were configured to continuously record a data point every 60‐s for the entire 7‐day period. The Actiwatch2 provides key information on sleep patterns based on the sleeper's movement, including timing and duration of sleep as well as awakenings.
Since automatic determinations of rest intervals and sleep onset/offset by the Actiware software can be inaccurate due to several factors including noncompliance with instructions on properly wearing the device, actigraphy data validation was performed for all sleep nights so that rest intervals could be manually corrected whenever there was clear evidence of inaccuracy. Two blinded evaluators independently reviewed all sleep nights using a previously established methodology to determine inaccurate rest intervals. There was 100% agreement on which rest intervals needed to be adjusted. Only inaccurate rest intervals were then manually scored by the evaluators based on a prearranged procedure utilising auxiliary information available from the Actiwatch2 such as movement score, light level and event marker. Interrater reliability was tested by selecting a random sample of sleep nights (n = 244). Agreement was defined as both raters reaching an agreement on the timing of the start and stop of the rest interval for a particular night (±5 min). An overall agreement rate of 88% was achieved with the consensus reached across all sleep nights following further scrutiny of the timing of the rest interval for sleep nights where there were initial differences.
There was no minimum number of sleep nights with valid data; all sleep nights with valid data were retained for analysis. Additional details on the instructions given to participants and how data from the Actiwatch2 was analysed can be found in Michaud et al. (2021).
2.4. Statistical Methods
The main objective was to assess the association between sleep quality and hair cortisol concentrations in Canadian adults using objective and subjective measures of sleep. Participants with invalid or missing measures for PSQI, sleep actimetry, or hair cortisol were excluded from the analysis.
2.4.1. Independent Variables
Five objective measures of sleep quality obtained from wrist actigraphy were included in the analysis. Sleep latency (LATO, how long it took to fall asleep), wake duration after sleep latency (WASO) and total sleep time (TST, time in bed minus sleep latency minus WASO minus snooze time) were recorded in minutes. Sleep efficiency was measured as TST divided by measured time in bed and reported as a percentage. Awakening bouts (WABT) during a sleep period were analysed as a rate per 60‐min in bed to allow for standardised comparisons across participants.
The PSQI was analysed in two ways: first using the continuous score and then dichotomising the score at the established cut‐off of 5, where values greater than 5 indicated poor sleep.
Only a limited list of factors considered by the authors as potential confounders of the association between hair cortisol and sleep, based on conceptual knowledge or previous research, were considered in this analysis. This was the preferred approach since the main objective was to describe the association between hair cortisol and sleep, rather than re‐evaluate the predictors of hair cortisol concentrations published by Michaud et al. (2016). Sociodemographic factors included age group, sex, body mass index (BMI) group, educational attainment and household annual income. Cosmetic hair treatment, diagnosed sleep disorder and use of sleep medication in the past month were the personal factors retained.
2.4.2. Data Analysis and Modelling
Data were analysed using SAS software, version 9.4 (SAS Institute Inc., Cary, NC, USA). A 5% statistical significance level was implemented throughout. As a first step, descriptive statistics of the hair cortisol concentration, sleep measures and sample characteristics were generated. Spearman's rank correlation coefficients were calculated as an initial assessment of the association between hair cortisol and objective measures of sleep. Hair cortisol concentrations showed a clear log‐normal distribution and regression models using it as a dependent variable revealed signs of violation of the assumptions of normality of errors and homoscedasticity. Therefore, all regression models were based on the log‐transformed hair cortisol variable.
In the first modelling stage, separate simple linear regression models using log hair cortisol concentration as the dependent variable and each of the identified variables as a predictor variable were fitted. The second step, considered as base models, consisted of the same models adjusted for age and sex. For the final step, seven multiple linear regression models with the log hair cortisol concentration as the dependent variable were implemented: one for each of the five objective measures of sleep, one for continuous PSQI and a last one for dichotomised PSQI. Final regression models were adjusted for all the sociodemographic and personal factors identified as potential confounders. Type I error was not adjusted for in the current analysis as each regression model in the final step tested a distinct a priori hypothesis. Traditional methods to adjust for multiple testing, such as Bonferroni and Benjamini‐Hochberg, are most appropriate in exploratory analyses or when testing a common null hypothesis across many comparisons. These methods can be overly conservative or inefficient when tests are not independent, as was the case in the present analysis due to dependencies among sleep‐related predictors.
2.4.3. Model Validation and Sensitivity Analyses
Visual examination of the residuals from the regression models did not reveal notable departures from normality and homogeneity of variance assumptions. Outliers and influential observations were also investigated. The variables of sleep efficiency and sleep latency had a few extreme values potentially exerting undue influence on the regression models. Therefore, models using the log‐transformed variables of sleep efficiency and sleep latency were also fitted for comparison purposes. Educational attainment and income are markers of socioeconomic status; including both variables in multiple regression models might introduce collinearity. Hence, income was excluded from the multiple regression models due to a higher proportion of missingness.
Finally, in order to confirm the robustness of the findings and in an attempt to fully capture the patterns in the data, a sensitivity analysis was performed. Quantile regression models were investigated to see if the relationship between hair cortisol concentrations and sleep quality measures changed across the quantiles of hair cortisol. It enabled the evaluation of changes in the magnitude and intensity of the sleep quality measure coefficients across the quantiles. Additionally, quantile regression is more robust to outliers and does not assume normality and constant variance.
3. Results
There were 1238 participants in total, 195 were not able to participate in the hair cortisol portion (not enough hair, too short), leaving 1043 participants of which 917 consented to participate. Of the 917 participants, 214 samples were removed due to insufficient mass, nine were removed because they exceeded the upper limit of quantification and 19 were removed because they were below the lower limit of quantification, leaving a total of 675 valid hair cortisol measures of which 173 were excluded because of invalid sleep actimetry or PSQI. Respondent characteristics from the available respondents are described in Table 1. A total of 502 respondents had valid hair cortisol concentrations and objective and subjective sleep measurements. Observations with missing values for covariates were excluded, with 19 respondents removed from further analysis mainly owing to missing BMI information (n = 18). The majority of respondents were between 45 and 64 years old (48.7%), female (63.4%), had a high school diploma or less (52.2%) and had a household income of less than $60,000 (46.6%).
TABLE 1.
Description of sample characteristics.
| Variable | n | Frequency (%) |
|---|---|---|
| Sex | ||
| Male | 177 | 36.6 |
| Female | 306 | 63.4 |
| Age group, year | ||
| 18–24 | 22 | 4.6 |
| 25–45 | 109 | 22.6 |
| 45–64 | 235 | 48.7 |
| ≥ 65 | 117 | 24.2 |
| BMI group | ||
| < 25 underweight‐normal | 157 | 32.5 |
| 25–30 overweight | 175 | 36.2 |
| ≥ 30 obese | 151 | 31.3 |
| Level of education | ||
| ≤ High school | 252 | 52.2 |
| Trade/certificate/college | 194 | 40.2 |
| University | 37 | 7.7 |
| Income a | ||
| < $60,000 | 225 | 46.6 |
| $60,000–$100,000 | 119 | 24.6 |
| ≥ $100,000 | 87 | 18.0 |
| Unknown | 52 | 10.8 |
| Cosmetic hair treatment | ||
| Yes | 194 | 40.2 |
| No | 289 | 59.8 |
| Diagnosed sleep disorder | ||
| Yes | 49 | 10.1 |
| No | 434 | 89.9 |
| Sleep medication, past month | ||
| Yes | 114 | 23.6 |
| No | 369 | 76.4 |
| PSQI | ||
| ≤ 5 good sleep quality | 246 | 50.9 |
| > 5 poor sleep quality | 237 | 49.1 |
Abbreviations: BMI, body mass index; PSQI, Pittsburgh Sleep Quality Index.
Income in CAD; 1 CAD ~0.9815 USD for the study reference period.
Descriptive statistics for sleep measures are reported in Table 2. The average TST was 406 min, which corresponds to 6 h and 46 min of sleep. The latter number excludes minutes scored as WASO, which represented a mean of 52 min per night in this sample.
TABLE 2.
Descriptive statistics for self‐reported and objective measures of sleep.
| Variable | n | Mean | Median | SD |
|---|---|---|---|---|
| PSQI score (0–21) | 483 | 6.3 | 5.0 | 4.2 |
| Actigraph | ||||
| Sleep efficiency (%) | 483 | 84.8 | 85.8 | 6.3 |
| LATO (min) | 483 | 11.9 | 8.9 | 12.8 |
| TST (min) | 483 | 406.4 | 408.0 | 56.0 |
| WASO (min) | 483 | 52.0 | 48.6 | 23.0 |
| WABT (rate per 60‐min in bed) | 483 | 3.0 | 2.9 | 0.8 |
Abbreviations: LATO, sleep latency; PSQI, Pittsburgh Sleep Quality Index; SD, standard deviation; TST, total sleep time; WABT, awakening bouts; WASO, wake after sleep onset.
A considerable difference between the median hair cortisol concentration of 130.9 ng/g and the mean of 274.7 ng/g was observed, emphasising the right‐skewed distribution of this variable. Thus, the geometric mean for hair cortisol by respondent's characteristics are presented in Table 3. Unadjusted concentrations highlighted higher levels in males, those with an educational attainment of high school or less, those with a BMI of 30 or more, and those that did not report having a cosmetic hair treatment within the previous 3 months.
TABLE 3.
Geometric mean of hair cortisol concentration by respondent characteristics.
| Variable | n | Geometric mean (95% CI) |
|---|---|---|
| Sex | ||
| Male | 177 | 187.2 (164.4–213.1) |
| Female | 306 | 123.1 (109.6–138.3) |
| Age group, year | ||
| 18–24 | 22 | 130.4 (95.1–178.9) |
| 25–45 | 109 | 135.1 (114.9–158.8) |
| 45–64 | 235 | 141.1 (123.7–161.0) |
| ≥ 65 | 117 | 160.1 (131.1–195.5) |
| BMI group | ||
| < 25 underweight‐normal | 157 | 118.8 (103.2–136.7) |
| 25–30 overweight | 175 | 141.8 (121.1–166.1) |
| ≥ 30 obese | 151 | 177.3 (151.3–207.9) |
| Level of education | ||
| ≤ High school | 252 | 154.8 (136.6–175.5) |
| Trade/certificate/college | 194 | 139.6 (120.8–161.4) |
| University | 37 | 99.2 (80.1–122.9) |
| Income | ||
| < $60,000 | 225 | 152.0 (132.9–173.8) |
| $60,000–$100,000 | 119 | 129.2 (110.4–151.3) |
| ≥ $100,000 | 87 | 138.1 (110.2–173.0) |
| Unknown | 52 | 152.2 (114.4–202.7) |
| Cosmetic hair treatment | ||
| Yes | 194 | 109.8 (95.4–126.4) |
| No | 289 | 171.9 (153.7–192.1) |
| Diagnosed sleep disorder | ||
| Yes | 49 | 167.7 (128.2–219.3) |
| No | 434 | 141.1 (128.3–155.1) |
| Sleep medication, past month | ||
| Yes | 114 | 159.5 (140.4–181.2) |
| No | 369 | 129.7 (114.6–146.9) |
| PSQI | ||
| ≤ 5 good sleep quality | 246 | 144.5 (121.1–172.3) |
| > 5 poor sleep quality | 237 | 143.3 (129.2–158.9) |
Abbreviations: BMI, body mass index; PSQI, Pittsburgh Sleep Quality Index.
Spearman correlation coefficients between the five sleep actigraphy measures and hair cortisol concentrations showed no sign of association (Table 4). All correlations were relatively weak and nonsignificant. Negative coefficients for sleep efficiency and TST indicated that lower values of these measures were associated with higher hair cortisol concentrations. As expected, sleep latency exhibited a different pattern, with higher sleep latency values corresponding to higher hair cortisol concentrations.
TABLE 4.
Spearman rank correlation coefficients between sleep actigraphy measures and hair cortisol concentrations.
| Sleep actigraphy measure | Correlation coefficient with hair cortisol | p |
|---|---|---|
| Sleep efficiency | −0.04 | 0.389 |
| LATO | 0.07 | 0.145 |
| TST | −0.06 | 0.227 |
| WASO | −0.01 | 0.810 |
| WABT | −0.05 | 0.240 |
Abbreviations: LATO, sleep latency; TST, total sleep time; WABT, awakening bouts; WASO, wake after sleep onset.
Results from the simple regression models and the same models adjusted for age and sex are reported in Table 5. Overall, results were similar for the simple regression and the base models. Males, participants with a BMI of 30 or higher, those with an education level of high school or less, and those without cosmetic hair treatments had statistically significant higher hair cortisol levels. None of the five actigraphy objective measures of sleep were associated with hair cortisol concentration, indicating insufficient evidence to conclude that an effect exists at the population level. However, the dichotomised PSQI variable was significant, suggesting that the data provide evidence to reject the null hypothesis of no association. In these models, PSQI scores greater than 5, representing poor sleep, were associated with higher hair cortisol concentrations.
TABLE 5.
Simple regression models a and base models a adjusted for age and sex for hair cortisol concentration.
| Simple regression models | Models adjusted for age and sex | |||||
|---|---|---|---|---|---|---|
| Beta | SE | p | Beta | SE | p | |
| Sex | ||||||
| Male | 0.419 | 0.093 | < 0.0001 | 0.420 | 0.093 | < 0.0001 |
| Female | Reference | Reference | ||||
| Age group, year | 0.561 | 0.539 | ||||
| 18–24 | −0.205 | 0.233 | 0.380 | −0.226 | 0.229 | 0.323 |
| 25–45 | −0.170 | 0.134 | 0.204 | −0.150 | 0.131 | 0.252 |
| 45–64 | −0.126 | 0.114 | 0.266 | −0.140 | 0.111 | 0.209 |
| ≥ 65 | Reference | Reference | ||||
| BMI group | 0.002 | 0.005 | ||||
| < 25 underweight‐normal | −0.401 | 0.113 | 0.000 | −0.359 | 0.112 | 0.002 |
| 25–30 overweight | −0.223 | 0.110 | 0.043 | −0.250 | 0.109 | 0.022 |
| ≥ 30 obese | Reference | Reference | ||||
| Level of education | 0.037 | 0.103 | ||||
| ≤ High school | 0.445 | 0.176 | 0.012 | 0.372 | 0.174 | 0.033 |
| Trade/certificate/college | 0.342 | 0.179 | 0.057 | 0.305 | 0.177 | 0.085 |
| University | Reference | Reference | ||||
| Income | 0.504 | < 0.0001 | ||||
| < $60,000 | −0.002 | 0.154 | 0.992 | −0.009 | 0.153 | 0.955 |
| $60,000–$100,000 | −0.164 | 0.167 | 0.326 | −0.124 | 0.165 | 0.454 |
| ≥ $100,000 | −0.098 | 0.176 | 0.579 | −0.071 | 0.175 | 0.684 |
| Unknown | Reference | Reference | ||||
| Cosmetic hair treatment | ||||||
| Yes | −0.448 | 0.091 | < 0.0001 | −0.311 | 0.109 | 0.005 |
| No | Reference | Reference | ||||
| Diagnosed sleep disorder | ||||||
| Yes | 0.173 | 0.151 | 0.253 | 0.084 | 0.151 | 0.578 |
| No | Reference | Reference | ||||
| Sleep medication, past month | ||||||
| Yes | 0.008 | 0.108 | 0.939 | 0.042 | 0.106 | 0.692 |
| No | Reference | Reference | ||||
| PSQI | ||||||
| > 5 poor sleep quality | 0.206 | 0.091 | 0.024 | 0.210 | 0.090 | 0.020 |
| ≤ 5 good sleep quality | Reference | Reference | ||||
| Actigraphy | ||||||
| Sleep efficiency | −0.011 | 0.007 | 0.145 | −0.003 | 0.007 | 0.704 |
| LATO | 0.006 | 0.004 | 0.113 | 0.003 | 0.004 | 0.369 |
| TST | −0.001 | 0.001 | 0.141 | −0.001 | 0.001 | 0.413 |
| WASO | −0.001 | 0.002 | 0.743 | −0.002 | 0.002 | 0.425 |
| WABT | −0.062 | 0.054 | 0.253 | −0.086 | 0.055 | 0.121 |
Abbreviations: BMI, body mass index; LATO, sleep latency; PSQI, Pittsburgh Sleep Quality Index; SE, standard error; TST, total sleep time; WABT, awakening bouts; WASO, wake after sleep onset.
Dependent variable is log‐transformed hair cortisol concentration.
Results from multiple regression models adjusted for potential confounders were similar to the unadjusted models (Table 6). The only sleep measure significantly associated with hair cortisol concentrations was PSQI. When transformed back to the original scale, a regression coefficient of 0.21 for the dichotomised PSQI on the log‐transformed scale can be interpreted as an expected increase of 23% in the geometric mean of hair cortisol levels when transitioning from a PSQI score below 5 to one above 5.
TABLE 6.
| Variable | Beta | SE | p |
|---|---|---|---|
| Sleep efficiency | −0.005 | 0.0074 | 0.502 |
| Sleep efficiency (log‐transformed) | −0.2739 | 0.5517 | 0.620 |
| LATO | 0.0039 | 0.0036 | 0.277 |
| LATO (log‐transformed) | 0.0262 | 0.0574 | 0.648 |
| TST | −0.0006 | 0.0008 | 0.427 |
| WASO | −0.0011 | 0.002 | 0.578 |
| WABT | −0.0641 | 0.0559 | 0.252 |
| PSQI | |||
| > 5 poor sleep quality | 0.2084 | 0.0893 | 0.020 |
| ≤ 5 good sleep quality | Reference | ||
| PSQI continuous | 0.0236 | 0.0106 | 0.027 |
Abbreviations: LATO, sleep latency; PSQI, Pittsburgh Sleep Quality Index; SE, standard error; TST, total sleep time; WABT, awakening bouts; WASO, wake after sleep onset.
Models adjusted for sex, age group, BMI group, cosmetic hair treatment, diagnosed sleep disorder and education.
Dependent variable is log‐transformed hair cortisol concentration.
In the sensitivity analyses, models using log‐transformed sleep efficiency and sleep latency variables yielded results very similar to those obtained from the untransformed models (Table 6).
Quantile regression models for the five sleep actimetry variables did not reveal substantial changes in the regression coefficients for the quantiles across the distribution of hair cortisol concentrations. For PSQI, at levels of hair cortisol above the 70th percentile, regression coefficients were somewhat smaller (data not shown). This indicates that, at high levels of hair cortisol, sleep quality as assessed by the PSQI has a lesser impact on hair cortisol concentration relative to that observed at lower levels of hair cortisol. Overall, results from the regression analyses using the continuous and dichotomised PSQI were consistent, and both showed a significant positive association, linking high PSQI with elevated hair cortisol concentrations.
4. Discussion
The current study provides evidence that hair cortisol levels may be a sensitive biological indicator of cumulative stress that can be associated with sleep disturbance when measured through self‐report. While comparisons to other studies are limited because this research area is still developing, our findings are consistent with previous research in this area where positive associations between hair cortisol concentrations and insomnia and perceived sleepiness have been reported (Ahabrach et al. 2023). Moreover, elevated hair cortisol concentrations have been observed among healthy middle‐aged women with elevated scores on the PSQI sleep latency domain (Mazgelyte et al. 2023). In contrast, these authors found no association between hair cortisol levels and global PSQI scores above the ‘poor sleep’ threshold, even though they reported a weak statistically significant positive correlation with the overall PSQI score.
Accounts of sleep quality from self‐report and objective measures are not always consistent (Kirsch 2012; McCall and Edinger 1992; Patel et al. 2009; Jackowska et al. 2016). This disparity was affirmed and extended to biomarkers of inflammation in a subsample of the Cleveland Family Study cohort (Patel et al. 2009). The authors reported statistically positive associations between self‐reported habitual sleep duration and the proinflammatory biomarkers, C‐reactive protein (CRP) and interleukin‐6 (IL‐6), and tumour necrosis factor‐alpha (TNF‐α, p = 0.057). However, neither CRP nor IL‐6 was related to PSG‐measured sleep duration, and reduced sleep duration measured through PSG was associated with elevated TNF‐α. In another study, psychological and biological measures were compared to sleep quality evaluated with PSQI and actigraphy (Jackowska et al. 2016). The authors reported a significantly lower cortisol awakening response with poor sleep measured with the PSQI, but not with actigraphy. On the other hand, diastolic blood pressure was only statistically elevated among sleepers with actigraphy‐measured longer sleep latency. In the current study, hair cortisol concentration was associated with self‐reported, but not actigraphy‐measured sleep. It is noteworthy that Eythorsdottir et al. (2020) also showed no association with actigraphy‐measured sleep patterns over 5 days among a younger sample of children, aged 2–6 years. One possible explanation may be the mismatch between hair cortisol levels and the time reference period reflected by the various sleep outcomes evaluated in the current study. Cortisol was quantified from a 3‐cm hair sample taken from the apex of the scalp, which more or less reflects changes in cortisol integrated into the hair shaft over the previous 90 days (Russell et al. 2012; Wennig 2000). This time period is more aligned to the previous 30 days evaluated with the PSQI than to the 7 days of actigraphy. The strength of the association for both self‐reported and objectively measured sleep might increase by minimising the differences in time reference periods. More research is needed to test this possibility because Zhu et al. (2022) sampled hair that reflected approximately 30 days of growth and found that poor sleep quality on the PSQI was associated with the cortisol metabolite, cortisone, but not cortisol. It should also be considered that self‐reported sleep disturbance may provide a better indicator of underlying stress reactions, and that hair cortisol levels may be a more sensitive indicator of these changes when compared to those measured using sleep actigraphy. Some support for this comes from Michaud et al. (2016), which showed that hair cortisol concentrations were statistically positively correlated with self‐reported stress over a 30‐day period, but not with measured resting heart rate, diastolic or systolic blood pressure averaged over five measures taken at 1‐min intervals.
We also conducted a quantile analysis to look more closely at the relationship between hair cortisol levels and sleep. At the highest concentrations, the observed correlation between hair cortisol and the PSQI tended to decrease. This is somewhat counterintuitive insofar as elevated hair cortisol typically reflects an elevated activation of the hypothalamic pituitary adrenal (HPA) axis, which itself has been associated with reduced sleep quality (El Mlili et al. 2021). The relationship between sleep disruption and the HPA axis as a primary nexus to developing metabolic and psychiatric illnesses has been reviewed by Balbo et al. (2010). Indeed, sleep exerts a direct impact on the secretion of cortisol from the adrenal glands, characterised by inhibition during sleep onset and excitation following awakening (i.e., the well‐established cortisol awakening response). The reliability of our findings undoubtedly requires more research; however, it is possible that at the highest concentrations, hair cortisol is driven by factors that are more strongly linked to the activation of the HPA response than PSQI‐measured sleep quality. It should also be considered that because the HPA axis activity is self‐regulating through negative feedback, sustained elevations in cortisol may suppress hypothalamic corticotropin‐releasing hormone (CRH) in the brain (Anisman 2015). Insofar as CRH in the brain is considered to be involved in arousal and awakening independent of stressor exposure (Chang and Opp 2001), it could be speculated that its reduction through elevated cortisol is somehow related to improved sleep. The pattern at the highest hair cortisol levels notwithstanding, the association between hair cortisol levels and PSQI was reliable, observed both when poor sleep was defined using the established threshold of scores above five (Buysse et al. 1989) and when using mean PSQI scores. Furthermore, this relationship was maintained after adjusting for several variables known to influence stress reactions and/or sleep quality.
While this area of research is still in its infancy, our findings add to a growing evidence base that suggests hair cortisol may provide a promising biomarker for research that aims to evaluate biological responses toward sustained stressors that could influence both sleep quality and stress reactions. In their review of the potential utility of hair cortisol in relation to sleep disorders, El Mlili et al. (2021) highlighted the benefits of using hair cortisol as a biological indicator of the accumulated costs to sleep disorders wherein they noted several research needs, including consideration for how hair cortisol may be sensitive to the use of pharmaceutical treatment for sleep disorders. Our results add some insight in this regard insofar as the associations with PSQI remained after adjusting for being diagnosed with a sleep disorder.
5. Conclusion
Our study has the strength of a large randomly selected sample, where hair cortisol concentrations were evaluated in relation to both a validated measure of self‐reported sleep quality and a 7‐day period of objectively measured sleep outcomes. Controlling for several variables that could confound the association between hair cortisol and sleep also adds strength to the validity of the multiple regression models. Despite its strengths, the current study is limited by the time mismatch between the hair cortisol reference period (e.g., ~90‐days) and both measures used to evaluate sleep.
Another limitation of this study is that its design does not allow for a thorough assessment of the bidirectional relationship between HPA axis activity and sleep (Steiger 2002). While we implicitly suggest that disturbed sleep drives cortisol changes, our observed correlation cannot capture these complex temporal dynamics. Multiple lines of evidence from both human and animal studies show that the inverse is also evident, as discussed in detail by Balbo et al. (2010). There is also an interesting body of research showing that high sleep reactivity (i.e., one's predisposition to experiencing sleep disturbance in response to stressors) is related to stress reactivity, which describes the magnitude of one's biophysiological reaction to stressors. The two concepts are closely related, whereby high stressor reactivity can augment sleep reactivity and vice versa, creating a cycle that predisposes one to adverse health (Reffi et al. 2023). Furthermore, although the current study accounted for the use of sleep medication, future research should also consider other types of medication known to affect both sleep and HPA axis function, such as benzodiazepines and other psychotropic drugs.
As previously reviewed by Michaud et al. (2023), studies that may especially benefit from the use of hair cortisol as a biomarker of long‐term stress include those evaluating environmental noise exposure and its potential association with adverse health effects. Chronic exposure to environmental noise has a long history of being studied as a risk factor for adverse health outcomes where the putative causal pathways are through noise‐induced activation of stress reactions and noise‐induced sleep disturbance (Babisch 2002). Although some studies have found positive associations between noise exposure and salivary cortisol (Selander et al. 2009; Lefèvre et al. 2017), such short‐term measures are subject to considerable variability, which includes, but is not limited to, diurnal rhythms in endogenous cortisol activity (Edwards et al. 2001; Hennig et al. 2000; Legler et al. 1982). Assessing chronic changes in cortisol using saliva, blood or urine requires multiple sampling points, which adds to research design costs and increases the probability of noncompliance with prescribed sampling regimens that are more burdensome relative to hair sampling. More research in this and similar areas that aim to evaluate objective measures of stress from sustained environmental stressors, including habitual sleep disturbance, is warranted.
Author Contributions
David S. Michaud: conceptualization, methodology, supervision, investigation, funding acquisition, project administration, writing – original draft, writing – review and editing. Mireille Guay: methodology, data curation, formal analysis, writing – review and editing.
Conflicts of Interest
The authors declare no conflicts of interest.
Acknowledgement
Open Access funding provided by the Health Canada library.
Michaud, D. S. , and Guay M.. 2026. “Elevated Hair Cortisol Concentrations Are Associated With Poor Sleep Quality Evaluated Using the Pittsburgh Sleep Quality Index but Not With Actigraphy.” Journal of Sleep Research 35, no. 1: e70105. 10.1111/jsr.70105.
Funding: This research was funded internally entirely by Health Canada.
Data Availability Statement
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.
References
- Abdul Jafar, N. K. , Tham E. K. H., Eng D. Z. H., et al. 2023. “Preconception Sleep Quality Moderates the Association Between Preconception Hair Cortisol Levels and Mental Health in Pregnant Women.” Journal of Affective Disorders 334: 187–196. 10.1016/j.jad.2023.04.129. [DOI] [PubMed] [Google Scholar]
- Ahabrach, H. , El Mlili N., Mafla‐España M. A., and Cauli O.. 2023. “Hair Cortisol Concentration Associates With Insomnia and Stress Symptoms in Breast Cancer Survivors.” International Journal of Psychophysiology 181: 49–56. 10.1016/j.ijpsycho.2023.07.006. [DOI] [PubMed] [Google Scholar]
- Anisman, H. 2015. Stress and Your Health: From Vulnerability to Resilience. John Wiley & Sons, Ltd. [Google Scholar]
- Babisch, W. 2002. “The Noise/Stress Concept, Risk Assessment and Research Needs.” Noise & Health 4, no. 16: 1–11. [PubMed] [Google Scholar]
- Balbo, M. , Leproult R., and Van Cauter E.. 2010. “Impact of Sleep and Its Disturbances on Hypothalamo‐Pituitary‐Adrenal Axis Activity.” International Journal of Endocrinology 2010, no. 1: 1–16. 10.1155/2010/759234. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Benasi, G. , Cheng B., Aggarwal B., and St‐Onge M.‐P.. 2024. “The Effects of Sustained Mild Sleep Restriction on Stress and Distress Among Healthy Adults: Findings From Two Randomized Crossover Studies.” Sleep Medicine 115: 83–87. 10.1016/j.sleep.2024.02.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Buysse, D. J. , Reynolds C. F., Monk T. H., Berman S. R., and Kupfer D. J.. 1989. “The Pittsburgh Sleepquality Index: A New Instrument for Psychiatric Practice and Research.” Psychiatry Research 28, no. 2: 193–213. 10.1016/0165-1781(89)90047-4. [DOI] [PubMed] [Google Scholar]
- Chang, F.‐C. , and Opp M. R.. 2001. “Corticotropin‐Releasing Hormone (CRH) as a Regulator of Waking.” Neuroscience & Biobehavoral Reviews 25: 445–453. [DOI] [PubMed] [Google Scholar]
- Chattu, V. K. , Manzar M. D., Kumary S., et al. 2018. “The Global Problem of Insufficient Sleep and Its Serious Public Health Implications.” Healthcare (Basel) 7, no. 1: 1. 10.3390/healthcare7010001. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Colledge, F. , Brand S., Zimmerer S., et al. 2017. “Individuals Following Aneurysmal Subarachnoid Haemorrhage, Hair Cortisol Concentrations Are Higher and More Strongly Associated With Psychological Functioning and Sleep Complaints than in Healthy Controls.” Neuropsychobiology 75, no. 1: 12–20. 10.1159/000477966. [DOI] [PubMed] [Google Scholar]
- Edwards, S. , Evans P., Hucklebridge F., and Clow A.. 2001. “Association Between Time of Awakening and Diurnal Cortisol Secretory Activity.” Psychoneuroendocrinology 26, no. 6: 613–622. 10.1016/S03064530(01)00015-4. [DOI] [PubMed] [Google Scholar]
- El Mlili, N. , Ahabrach H., and Cauli O.. 2021. “Hair Cortisol Concentration as a Biomarker of Sleep Quality and Related Disorders.” Lifestyles 11, no. 2: 81. 10.3390/life11020081. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Eythorsdottir, D. Y. , Frederiksen P., Larsen S. C., Olsen N. J., and Heitmann B. I.. 2020. “Associations Between Objective Measures of Physical Activity, Sleep and Stress Levels Among Preschool Children.” BMC Pediatrics 20: 258–265. 10.1186/s12887-020-02108-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Feller, S. , Vigl M., Bergmann M. M., Boeing H., Kirschbaum C., and Stadler T.. 2014. “Predictors of Hair Cortisol Concentrations in Older Adults.” Psychoneuroendocrinology 39: 132–140. 10.1016/j.psyneuen.2013.10.007. [DOI] [PubMed] [Google Scholar]
- Hennig, J. , Friebe J., Ryl I., et al. 2000. “Upright Posture Influences Salivary Cortisol.” Psychoneuroendocrinology 25: 69–83. 10.1016/S0306-4530(99)00037-2. [DOI] [PubMed] [Google Scholar]
- Jackowska, M. , Ronaldson A., Brown J., and Steptoe A.. 2016. “Biological and Psychological Correlates of Self‐Reported and Objective Sleep Measures.” Journal of Psychosomatic Research 84: 52–55. 10.1016/j.jpsychores.2016.03.017. [DOI] [PubMed] [Google Scholar]
- Jaspan, V. N. , Greenberg G. S., Parihar S., et al. 2024. “The Role of Sleep in Cardiovascular Disease.” Current Atherosclerosis Reports 26: 249–262. 10.1007/s11883-024-01207-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Juster, R. P. , and McEwen B. S.. 2015. “Sleep and Chronic Stress: New Directions for Allostatic Load Research.” Sleep Medicine 16: 7–8. 10.1016/j.sleep.2014.07.029. [DOI] [PubMed] [Google Scholar]
- Kirsch, D. B. 2012. “A Neurologist's Guide to Common Subjective and Objective Sleep Assessments.” Neurologic Clinics 30: 987–1006. 10.1016/j.ncl.2012.08.009. [DOI] [PubMed] [Google Scholar]
- Knutson, K. L. 2010. “Sleep Duration and Cardiometabolic Risk: A Review of the Epidemiologic Evidence.” Best Practice & Research. Clinical Endocrinology & Metabolism 24: 731–743. 10.1016/j.beem.2010.07.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Lanfear, J. , Voegal C. D., Binz T. M., and Paul R. A.. 2020. “Hair Cortisol Measurement in Older Adults: Influence of Demographic and Physiological Factors and Correlation With Perceived Stress.” Steroids 163: 108712. 10.1016/j.steroids.2020.108712. [DOI] [PubMed] [Google Scholar]
- Lefèvre, M. , Carlier M.‐C., Champelovier P., et al. 2017. “Effects of Aircraft Noise Exposure on Saliva Cortisol Near Airports in France.” Occupational and Environmental Medicine 74, no. 8: 612–618. 10.1136/oemed-2016-104208. [DOI] [PubMed] [Google Scholar]
- Legler, M. , Brandenberger G., Hietter B., Siméoni M., and Reinhardt B.. 1982. “Diurnal Cortisol Peaks and Their Relationships to Meals.” Journal of Clinical Endocrinology & Metabolism 55, no. 4: 757–761. 10.1210/jcem-55-4-757. [DOI] [PubMed] [Google Scholar]
- Mazgelyte, E. , Walatkevičiūté A., Songailiené J., et al. 2023. “Association of Hair Glucocorticoid Levels With Sleep Quality Indicators: A Pilot Study in Apparently Healthy Perimenopausal and Menopausal Women.” Frontiers in Endocrinology 14. 10.3389/fendo.2023.1186014. [DOI] [PMC free article] [PubMed] [Google Scholar]
- McCall, W. V. , and Edinger J. D.. 1992. “Subjective Total Insomnia: An Example of Sleep State Misperception.” Sleep 15: 71–73. [PubMed] [Google Scholar]
- McEwen, B. S. 2006. “Sleep Deprivation as a Neurobiologic and Physiologic Stressor: Allostasis and Allostatic Load.” Metabolism 55: S20–S23. 10.1016/j.metabol.2006.07.008. [DOI] [PubMed] [Google Scholar]
- McEwen, B. S. , and Stellar E.. 1993. “Stress and the Individual: Mechanisms Leading to Disease.” Archives of Internal Medicine 153: 2093–2101. [PubMed] [Google Scholar]
- Michaud, D. S. , Feder K., Keith S. E., et al. 2016. “Self‐Reported and Measured Stressrelated Responses Associated With Exposure to Wind Turbine Noise.” Journal of the Acoustical Society of America 39, no. 3: 1467–1479. 10.1121/1.4942402. [DOI] [PubMed] [Google Scholar]
- Michaud, D. S. , Keith S. E., Guay M., Voicescu S., Denning A., and McNamee J. P.. 2021. “Sleep Actigraphy Time‐Synchronized With Wind Turbine Output.” Sleep 44, no. 9: 1–12. 10.1093/sleep/zsab070. [DOI] [PubMed] [Google Scholar]
- Michaud, D. S. , Marro L., Denning A., Shackleton S., Toutant N., and McNamee J. P.. 2023. “Chronic Sleep Disturbance Among Adult Canadians: Associations and Implications to the Evaluation of Noise Impacts Under Canada's Impact Assessment Act.” Environmental Impact Assessment Review 101: 107109. 10.1016/j.eiar.2023.107109. [DOI] [Google Scholar]
- Michaud, D. S. , Thomson E. M., van Oosterhout P., and McNamee J. P.. 2022. “Hair Cortisol as a Viable Tool for the Assessment of an Association Between Environmental Noise Exposure and Chronic Stress.” Journal of the Acoustical Society of America 152: 866–876. 10.1121/10.0012887. [DOI] [PubMed] [Google Scholar]
- Passos, G. S. , Youngstedt S. D., Rodrigues A. A., et al. 2023. “Insomnia Severity Is Associated With Morning Cortisol and Psychological Health.” Sleep Science 16: 92–96. 10.1055/s-0043-1767754. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Patel, S. R. , Xhu X., Storfer‐Isser A., et al. 2009. “Sleep Duration and Biomarkers of Inflammation.” Sleep 32, no. 2: 200–204. 10.1093/sleep.32.2.200. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Pereg, D. , Chan J., Russell E., et al. 2013. “Cortisol and Testosterone in Hair as Biological Markers of Systolic Heart Failure.” Psychoneuroendocrinology 38, no. 12: 2875–2882. 10.1016/j.psyneuen.2013.07.015. [DOI] [PubMed] [Google Scholar]
- Rajcani, J. , Vytykacova S., Solarikova P., and Brezina I. 2021. “Stress and Hair Cortisol Concentrations in Nurses During the First Wave of the COVID‐19 Pandemic.” Psychoneuroendocrinology 129: 105245. 10.1016/j.psyneuen.2021.105245. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Reffi, A. N. , Kalmbach D. A., Cheng P., and Drake C. L.. 2023. “The Sleep Response to Stress: How Sleep Reactivity Can Help Us Prevent Insomnia and Promote Resilience to Trauma.” Journal of Sleep Research 32: e13892. 10.1111/jsr.13892. [DOI] [PubMed] [Google Scholar]
- Russell, E. , Koren G., Rieder M., and van Uum S. H.. 2012. “Hair Cortisol as a Biological Marker of Chronic Stress: Current Status, Future Directions and Unanswered Questions.” Psychoneuroendocrinology 37: 589–601. 10.1016/j.psyneuen.2011.09.009. [DOI] [PubMed] [Google Scholar]
- Selander, J. , Bluhm G., Theorell T., et al. 2009. “Salivacortisol and Exposure to Aircraft Noise in Six European Countries.” Environmental Health Perspectives 117, no. 11: 1713–1717. 10.1289/ehp.0900933. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Sivakumaran, K. , Ritonja J. A., Palmer N., et al. 2023. “Effect of Sleep Disturbance on Biomarkers Related to the Development of Adverse Health Outcomes: A Systematic Review of the Human Literature.” Journal of Sleep Research 32, no. 3: e13775. 10.1111/jsr.13775. [DOI] [PubMed] [Google Scholar]
- Sopp, M. R. , Micheal T., Lass‐Hennermann J., Haim‐Nachum S., and Lommen M. J. J.. 2021. “Longitudinal Associations Between Hair Cortisol, PTSD Symptoms, and Sleep Disturbances in a Sample of Firefighters With Duty‐Related Trauma Exposure.” Psychoneuroendocrinology 134: 105449. 10.1016/j.psyneuen.2021.105449. [DOI] [PubMed] [Google Scholar]
- Stalder, T. , and Kirschbaum C.. 2012. “Analysis of Cortisol in Hair—State of the Art and Future Directions.” Brain, Behavior, and Immunity 26: 1019–1029. 10.1016/j.bbi.2012.02.002. [DOI] [PubMed] [Google Scholar]
- Stamatakis, K. A. , and Punjabi N. M.. 2010. “Effects of Sleep Fragmentation on Glucose Metabolism in Normal Subjects.” Chest 137, no. 1: 95–101. 10.1378/chest.09-0791. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Steiger, A. 2002. “Sleep and the Hypothalamo‐Pituitary‐Adrenocortical System.” Sleep Medicine Reviews 6: 125–138. 10.1053/smrv.2001.0159. [DOI] [PubMed] [Google Scholar]
- Suarez, E. C. 2008. “Self‐Reported Symptoms of Sleep Disturbance and Inflammation, Coagulation, Insulin Resistance and Psychosocial Distress: Evidence for Gender Disparity.” Brain, Behavior, and Immunity 22, no. 6: 960–968. 10.1016/j/bbi.2008.01.011. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Wennig, R. 2000. “Potential Problems With the Interpretation of Hair Analysis Results.” Forensic Science International 107: 5–12. 10.1016/S0379-0738(99)00146-2. [DOI] [PubMed] [Google Scholar]
- Zhang, Y. , Shen J., Zhou Z., et al. 2020. “Relationships Among Shift Work, Hair Cortisol Concentration and Sleep Disorders: A Cross‐Sectional Study in China.” BMJ Open 10: e038786. 10.1136/bmjopen-2020-038786. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Zhu, M. , Yuan L., Wu Y., et al. 2022. “Simultaneous LC‐MS/MS Quantification of Glucocorticoids, Melatonin and Its Metabolites in Hair.” Journal of Chromatography B 1196: 123217. 10.1016/j.jchromb.2022.123217. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.