Gender roles are related to cortisol habituation to repeated social evaluative stressors in adults: secondary analyses from a randomized controlled trial

Participants

One-hundred and thirty-eight individuals were recruited from an American Midwestern university to participate in a randomized control trial to test the effects of Mindfulness Based Stress Reduction (MBSR) and Cognitive Behavioral Therapy (CBT) on HPA axis habituation (Manigault et al., Citation2019). The majority of the final sample was female (72.1%) and had completed at least some college education (91.9%). Exclusionary criteria include the following: those who were younger than 18 or older than 50 years of age, were pregnant, endorsed a major endocrine or psychiatric disorder, were on steroid medication, had experience with an MBSR or CBT intervention, were not usually awake by 10:00 AM (excluding weekends), did not have evening availability to participate in interventions, or were associated with the psychology department. Eligible individuals also needed to report experiencing at least moderate levels of stress over the last month (i.e. scored a 4 or above on a 1–10 scale where 1 = little to no stress, and 10 = a great deal of stress). The stress screener item and 4-point cutoff were derived from the annual Stress in America^TM surveys (American Psychological Association, Citation2018) and are fully described in a previous publication of this data (Manigault et al., Citation2019). These exclusion criteria served to minimize inter-individual differences in diurnal cortisol rhythms as well as maximize stress reduction intervention effects.

Procedure

Trier social stress test

Following the 6-week intervention period, participants underwent modified versions of the Trier Social Stress Test (Kirschbaum et al., Citation1993) (TSST) on both laboratory visits (where speech preparation lasted five minutes instead of 10 min). The TSST consisted of a five minute period to anticipate and prepare for a mock job interview speech. This was followed by a five minute period to give the speech and five minutes to perform a difficult arithmetic task in front of an evaluative panel. The evaluative panel was composed of two research assistants wearing white lab coats and trained to maintain stoic expressions. The evaluative panel had no interaction with the participants prior to the TSST. The two TSSTs differed with respect to the starting value of the arithmetic tasks to minimize practice effects but speech instructions remained unchanged, consistent with prior work examining habituation (Thoma et al., Citation2017). The TSST produces a robust cortisol response because this laboratory stressors is characterized by uncontrollability and social evaluative threat (Dickerson & Kemeny, Citation2004). For the first visit, 57% of evaluative panels were “mixed” and 43% were “all female”. For the second visit, 66% of evaluative panels were “mixed”, 30% were “all female”, and 4% were “all male”. Gender role categories were independent of evaluative panel gender composition (all ps > .39).

Mindfulness based stress reduction

Those assigned to the MBSR condition underwent a 6-week version of MBSR training (Klatt et al., Citation2008; Lengacher et al., Citation2009) adapted from standardized treatment manuals (Kabat-Zinn, Citation1990; Kabat-Zinn & Santorelli, Citation1999). Participants were asked to attend group sessions for two hours weekly for a 6-week period in addition to participation in 45–60 min of formal and informal meditation practice at home each day. Those who participated received information on didactic mindfulness, mindfulness meditation practices, and were asked to engage in group discussion.

Cognitive behavioral therapy

A 6-week version of group CBT was developed and implemented from CBT treatments shown to reduce stress (Cully & Teten, Citation2008). Weekly two hour meetings lasted for 6 weeks, along with instructions to practice CBT techniques for 45–60 min at home daily. Topics covered included didactic information on stress and CBT techniques to reduce stress, supplemented by group discussion.

Waitlist

Those assigned to the Waitlist control could participate in the CBT stress management intervention at the end of the study but did not receive any intervention during the study.

Salivary cortisol

Salivary cortisol was collected four times per lab visit, including samples taken immediately before the TSST and 25, 35, and 60 min after the TSST. Participants were given Salivettes (Sarstedt, Inc., Newton, N.C.) and instructed to saturate the swab with saliva for up to 3 min. Instructions were given to ensure that swabs were not touched by hand. Samples were immediately stored at −20 °C after collection and then transferred to a −80 °C freezer until processed. All samples were centrifuged and assayed in duplicate (values averaged) at Ohio University facilities with standard enzyme-linked immunoassay procedures (Salimetrics, LLC, State College, PA). The assay sensitivity was less than 0.007 ug/dL, inter-assay coefficients of variation were less than 11.0%, and intra-assay coefficients of variation were less than 7.0%.

Gender roles

The 60-item Bem Sex Role Inventory (Bem, Citation1974) was used to assess participants’ perceptions of their gender roles as part of the baseline survey. This instrument uses a 7-point Likert scale ranging from 1 (never or almost never true) to 7 (always or almost always true) for stereotypically masculine (n = 20; e.g. ambitious, dominant) and feminine (n = 20; e.g. affectionate, gentle) descriptors, plus neutral filler items (n = 20; e.g. sincere, conscientious). Both the masculinity (α = .77) and femininity (α = .78) subscales showed high internal consistency. Masculinity and femininity ratings did not differ between MBSR, CBT and the waitlist (all ps > .57).

Masculinity and femininity subscales were median split such that all individuals were categorized as either high (above the median) or low (less than or equal to the median) masculinity and femininity. Consistent with prior work (Bem, Citation1981), median split variables were used to categorize participants as either androgynous (high femininity/high masculinity), feminine (high femininity/low masculinity), masculine (low femininity/high masculinity), or undifferentiated (low femininity/low masculinity). As shown on Table 1, assignment to study condition was independent from gender role category.

This median-based coding scheme was preferred over treating masculinity and femininity as continuous scales (and modeling their interaction) for important reasons. More specifically, the present study aimed to detect any difference between the four gender role categories hypothesized to exist under Bem’s model (Bem, Citation1981). To this end, it was necessary to derive four gender categories using median splits (and model gender category as a single factor with 4 levels). In contrast, modeling the interaction of masculinity and femininity as continuous variables would shift the scope of this project toward a different research question (i.e. we would now test if the association between masculinity and cortisol habituation changes as a function of femininity, and vice versa).

Additional measures

Participants self-reported biological sex at birth and waketime, whereas experiment start time was recorded by research assistants.

Treatment of missing and outlying data

Final analyses included 86 participants who completed the study up to the first laboratory visit. Among these individuals, 83 completed the second visit for a total of 676 possible cortisol samples. However, 1 sample contained too little saliva to be assayed and 4 were below assay threshold. Among the remaining 671 valid cortisol samples, 4 were outliers (more than 3 SD from mean). Excluding these outlying values had no influence on the present analyses. Accordingly, outlying cortisol values were retained in final analyses. Individuals who withdrew prior to the first visits did not differ from participants included in the final analyses on the basis of gender role category, age, sex, education, race, stress or ethnicity (all ps > .05).

Modeling total cortisol output

Maximum likelihood multilevel models were used to examine the change in total cortisol as a function of visit (0 = visit 1; 1 = visit 2), and dummy-coded gender role category (contrasting all four gender role categories: androgynous, masculine, feminine, and undifferentiated). These models were run using PROC MIXED on SAS 9.4/STAT 13.1 (SAS Institute Inc., Cary, N.C.). Three-level models were used as sampling occasions (level 1) were nested within laboratory visits (level 2) and visits were nested within individuals (level 3). Consistent with the following equation, visit-level and individual-level intercepts were included as random effects: $$\text{ln}{\left(\text{Cortisol}\right)}_{\text{ijk}}={\pi }_{0\text{jk}}+{\text{ e}}_{\text{ijk}}$$ $${\pi }_{0\text{jk}}={\beta }_{00\text{k}}+{\beta }_{0\text{1k}}\left({\text{Visit}}_{\text{jk}}\right)+{\text{ r}}_{0\text{jk}}$$ $${\beta }_{00\text{k}}={\gamma }_{000}+{\gamma }_{00\text{1}}{\left({\text{dummy1}}_{\text{k}}\right)}_{}+{\gamma }_{00\text{2}}\left({\text{dummy2}}_{\text{k}}\right)+{\gamma }_{00\text{3}}{\left({\text{dummy3}}_{\text{k}}\right)}_{}+{\text{ u}}_{00\text{k}}$$ $${\beta }_{0\text{1k}}={\gamma }_{0\text{1}0}+{\gamma }_{0\text{11}}{\left({\text{dummy1}}_{\text{k}}\right)}_{}+{\gamma }_{0\text{12}}\left({\text{dummy2}}_{\text{k}}\right)+{\gamma }_{0\text{13}}\left({\text{dummy3}}_{\text{k}}\right)$$

Habituation was indexed as a change in total cortisol output from visit 1 to visit 2 (i.e. where a greater decrease in total output implies greater habituation). This modeling approach yields insight comparable to testing a change in Area under the Curve relative to Ground or AUCg (Pruessner et al., Citation2003) but has the advantage of making use of partially missing cortisol data. It is worth noting that HPA axis habituation has been successfully examined and linked to impactful health outcomes using measures of total cortisol output (Kudielka et al., Citation2006; Wüst et al., Citation2005), however, other work also indexes habituation using measures of cortisol reactivity (i.e. peak - baseline) (Breines et al., Citation2014; Roos et al., Citation2019; Thoma et al., Citation2017). Accordingly, analyses of cortisol AUCg and reactivity scores are presented in online supplemental materials to facilitate comparison with other work. Associations between gender role categories and total cortisol output were also examined separately for each visit to provide a frame of reference for habituation effects (i.e. was habituation in a given group driven by low cortisol output during the second visit or high cortisol output during the first visit?). P-values below .05 were considered significant; p-values between .05 and .10 were considered marginally significant; p-values greater than .10 were considered non-significant.

Treatment of covariates

Consistent with prior analyses of the present data (Manigault et al., Citation2019), sex, wake time, and experiment start time were examined as covariates (i.e. tests of primary hypotheses were repeated while including/excluding covariate measures).

Preliminary results

As shown on Table 1, the final analytic sample (n = 86) was primarily comprised of young, non-Hispanic White women with some college education. Comparisons of gender categories revealed that the androgynous, feminine, masculine, and undifferentiated individuals differed with respect to age and education level (marginally significant differences), but not sex, ethnicity, or assignment to study condition. Accordingly, education level and age were also included as covariates (in addition to sex, wake time, and experiment start time).

Habituation

The interaction of gender role category x visit was significant when predicting total natural log transformed cortisol (F(3,502) = 2.65, p = .048), suggesting that the degree to which cortisol changed across visits differed between at least two gender role categories. As illustrated in Figures 1 and 2, the undifferentiated group exhibited a larger decrease in total cortisol from visit 1 to visit 2 than the feminine group (b = −.385, t(502) = 2.58, p = .010) and the masculine group (b = −.287, t(502) = 1.90, p = .057), but not the androgynous group (b = −.113, t(502) = 0.76, p = .44). In contrast, the androgynous group exhibited a marginally larger decrease in total cortisol from visit 1 to visit 2 than the feminine group (b = −.271, t(502) = 1.75, p = .081) but not the masculine group (b = −.174, t(502) = 1.11, p = .26). Finally, masculine and feminine individuals did not differ with respect to decrease in total cortisol from visit 1 to visit 2 (b = 0.097, t(502) = 0.62, p = .53).

Figure 1. Salivary cortisol as a function of study visit and gender role category. Undifferentiated individuals scored below the median on both masculinity and femininity; feminine individuals scored above the median on femininity and below the median on masculinity; masculine individuals scored above the median on masculinity and below the median on femininity; androgynous individuals scored above the median on both masculinity and femininity. Data shown are unadjusted marginal means estimates of salivary cortisol. Error bars shown correspond to standard errors of mean estimates.

Display full size

Figure 2. Salivary cortisol as a function of time since stressor onset, laboratory visit, and gender role category. Undifferentiated individuals (panel a) scored below the median on both masculinity and femininity; feminine individuals (panel b) scored above the median on femininity and below the median on masculinity; masculine individuals (panel c) scored above the median on masculinity and below the median on femininity; androgynous individuals (panel d) scored above the median on both masculinity and femininity. Data shown are unadjusted marginal means estimates of salivary cortisol. Error bars shown correspond to standard errors of mean estimates.

Display full size

Relevant to primary hypotheses, both the undifferentiated and androgynous groups showed a significant decrease in total cortisol from visit 1 to visit 2 (b = −.381, t(79) = 3.79, p < .001, and b = −.267, t(79) = 2.44, p = .016, respectively), whereas the feminine and masculine groups did not (b = .004, t(79) = 0.04, p = .97, and b = −0.093, t(79) = 0.83, p = .40, respectively). The interaction of gender role category x visit remained significant after controlling for sex, wake time, experiment start time, age, education level, and study conditions (F(3,489) = 4.65, p = .003). Moreover, the interaction of sex x gender role category x visit was non-significant (F(3,502) = 1.11, p = .34) suggesting that biological sex did not moderate the effect of gender role categories on cortisol habituation.

Examining sex effects independently revealed a non-significant interaction of sex x visit (F(1, 502) = 0.88, p = .35) and a significant main effect across visits (b = 0.39, t(502) = 2.43, p = .015), such that males exhibited higher total cortisol output across visits relative to females.

First visit

Total natural log transformed cortisol during the first visit was not significantly predicted by gender role category (p = .66).

Second visit

Total natural log transformed cortisol during the first visit was not significantly predicted by gender role category (p = .79).