Participants

Eligible participants were German-speaking healthy women aged between 18 and 60. We included only women based on the assumption of gender differences in stress responses [25], i.e. that women predominantly engage in a tend and befriend response to stress. Thus, we expected women to readily accept the rational of our intervention, which would facilitate the implementation of our social support stress management. To minimize the interference of medical conditions and behaviors on stress reactivity and to ensure the safety of participants, women were excluded when they reported current or chronic mental or physical disease assessed on self-report, currently received any medication, were in psychological or psychiatric treatment, and when smoking more than five cigarettes a day. Also and to prevent habituation effects to the social stress protocol, women with previous experiences with the Trier Social Stress Test [26] were excluded. Written informed consent was obtained from all participants.

We expected a medium effect size as previous randomized controlled studies on stress management interventions have demonstrated medium to large effect sizes for salivary cortisol stress reactivity [19–21]. Thus, estimating an intermediate to large effect size of f = 0.35, G*Power 3.1 calculated a total of N = 48 participants needed to detect interaction effects between two conditions and three measurements at 85% power using the originally planned MANCOVA for repeated measures [27].

Procedure

In February/March 2015 (trial 1) and September/October 2015 (trial 2) participants were recruited in lectures and via online advertisement at the University of Basel and the University of Applied Sciences and Arts of Northwestern Switzerland. Interested women completed an online screening questionnaire. A total of 53 women fulfilled inclusion criteria and were randomly allocated by two graduate students (SG and FK) in a 1:1 ratio to the intervention or the waitlist control condition, with allocation concealment by opaque sequentially numbered sealed envelopes. As we planned to include more subjects in trial 1 than we could finally engage and opaque envelopes with allocation to condition were prepared in advance, not all prepared envelopes could be used, resulting in a uneven distribution of participants to the two conditions. Participants in the intervention condition received a two-week social support stress management (SSSM) in groups of 5–8 participants (week 2 and 3). In the waitlist control condition, participants received the intervention after study completion (after week 8). Self-report questionnaires were completed online at baseline (week 1), post-intervention (week 4) and follow-up (week 8). Between the post and the follow-up assessment (week 5 to 7) all participants underwent a psychosocial stress test with assessement of neuroendocrine, cardiovascular and psychological stress responses. The two confederates leading the psychosocial stress test and data analysts were blinded to condition assignment. Data collection was to 62% completed by study conductors blinded to condition assignment. Blinding of participants and intervention providers was not possible due to the study design.

Intervention

The SSSM was conceptualized to address the importance of social support for health and wellbeing and to improve interpersonal skills to give and receive social support. Besides providing information on social support effects and its possible mechanisms, the SSSM focused on modeling supportive, non-judgemental communication between therapists and participants and on carrying out exercises in small groups to improve interpersonal communication skills. With this, the aim was to enhance the quality of interpersonal interactions and to allow personal and interpersonal exploration. Participants were encouraged to practise these skills in real life, i.e. between sessions, and to actively provide as well as ask for social support, since bi-directional support seems to be more effective in terms of well-being than just receiving support [28]. The intervention was based on a manual (for details, see below) and adherence to the manual was supervised by a certified psychotherapist (JG).

The intervention consisted of 16 training hours divided into 6 sessions with the first and the final session lasting 4 hours and the other four sessions 2 hours each. Each session focused on a different aspect of social support (see S1 Appendix). All sessions took place within a two-week period. Two graduate students in clinical psychology (SG, SP) and a PhD student (NH) with substantial training and experience in psychotherapy conducted the SSSM in the intervention condition. In the control condition, the intervention was conducted by four graduate students (SG, SP, SA, FK) and a PhD student (NH). Graduate students were constantly supervised by the PhD student and all therapists were supervised weekly by a certified psychotherapist (JG).

Manual of the social support stress management

The goal of the intervention is to impart interpersonal skills to deal with stress and give participants opportunities to train these skills. Most likely these skills are not completely new and at least partially already available. It is important to stress their importance, to implement and practise them out in a safe environment under professional supervision. Therefore, therapists should model social support by communicating openly and empathically and to validate each other and participants. Problems should be approached in a solution-focused way with emphasis on what was, is or could be helpful. Topics should be discussed first on the basis of personal experience and then in the context of scientific studies and models. At the end of each session, participants should be encouraged to share their experiences of the session, verbalize what they take home from the session and what they could try out in the next days.

The goal of the first session (4 hours) is to introduce and discuss the aim of the intervention and to train basic communication skills. In the first module (60 minutes), participants discuss what puts them under stress and what helps them to cope with stress. Therapists seek to put the results of the discussion in context of findings on social evaluative threat and the social dimension of stress [11], on social support [13] and the assumption of gender-specific coping [25]. In the second module (90 minutes), participants and therapists play a board game involving personal questions. The goal of the game is to create a relaxed atmosphere where participants and therapists can get to know each other. In the third module (90 minutes), participants are invited to discuss and practice determinants of good communication, such as empathy, authenticity and unconditional appreciation.

The goal of the second session (2 hours) is to improve the perception and communication of one’s own feelings [29] as a basis for supportive communication. For this purpose, participants are introduced to simple exercises to monitor somatic representations of emotional experiences, which are first demonstrated to the whole group under the guidance of a therapist and then exercised in small groups.

The goal of the third session (2 hours) is to focus on one’s own social network and the different forms of social support that are enacted in this network, especially when under stress [30]. Participants discuss different forms of support and draw an illustration of their social network. They discuss how their social ties are affected by stress and how and with whom they can increase beneficial social interactions when under stress.

The goal of the fourth session (2 hours) is to address self-disclosure and being vulnerable in social interactions in the delicate balance between feeling threatened by anticipated and real social rejection and the relieving experience to share problematic experience with others. These topics are discussed on the basis of personal experiences and with references to Leary’s sociometer theory [31].

The goal of the fifth session (2 hours) is to differentiate between different types and consequences of stress and how to utilize social support when under stress. In the first module (60 minutes), participants and therapists discuss differences between acute and chronic as well as functional and health-impacting stress. In the second module (60 minutes), the use of imagined persons as social support in times when no-one is directly available is discussed and experiences with this are shared with the group.

The goal of the sixth session (4 hours) is, beside the presentation of one new aspect of social support, the review and repetition of contents of the intervention. In the first module (60 minutes), the relevance and importance of physical contact as a form of social support is discussed on the basis of personal experience and scientific research [15]. In the second module (30 minutes), participants discuss what was important for them and possible changes they made during the intervention in small groups. In the closing module (90 minutes), the game from the first session is played again, this time with questions focused on experiences made during the intervention. A feedback round followed by a small reception with snacks and drinks concludes the intervention (60 minutes).

Stress test

The Trier Social Stress Test (TSST, [26]) was used to induce psychosocial stress. The TSST consists of a simulated job interview followed by a mental arithmetic task (5 minutes each) in front of an audience of two confederates. The TSST has repeatedly been found to induce profound endocrine and cardiovascular responses in 70–80% of the subjects tested [26]. Neuroendocrine, physiological and psychological data were collected before, during and after the test (for details see below). To account for circadian rhythm in cortisol secretion [32], all TSST took place between 1:55pm and 6:00pm (mean = 3:58pm; no differences between conditions: F(1,35) = 0.02, p = 0.91). The appointment was on average 23.55 days (SD = 6.99) after the last day of the intervention (range: 16–37 days, with 75% of appointments occurring within 26 days) in the intervention condition. After completion of data collection, participants were fully debriefed on the nature of the TSST and compensated with 50 CHF or study credits.

Measures

Demographic data including age and body mass index (BMI) were assessed online together with the inclusion criteria before condition randomization took place. At baseline (week 1), post-intervention (week 4) and at follow-up (week 8) the following questionnaires were administered online:

The German version of the Perceived Stress Scale [PSS; 33], consisting of 14 items rated on a 5-point scale, was used to assess the degree to which situations in the last few days have been appraised as stressful. PSS scores can be in the range between 0 and 56 with higher values indicating more perceived stress. In a probability sample of the United States, Cohen [34] found mean scores for women of 20.2 (SD 7.8).

The German version of the State Trait Anxiety Inventory [STAI; 35] was used to assess trait anxiety. The STAI trait form consists of 20 statements each to be rated on a 4-point scale with a total score of 20–37 interpreted as little or no anxiety, 38–44 as moderate anxiety and a score of 45–80 as extreme anxiety.

A German change-sensitive symptom list [ASS-SYM; 36] was employed to assess experience of relaxation, well-being, discomfort and preoccupation. The ASS-SYM has 6 subscales (physical and psychological exhaustion, nervousness and inner tension, psychophysiological deregulation, performance and behavioral problems, burden of pain, problems with self-determination and -control) with 8 items each, rated on a 4-point scale. Additionally a sum scale, which can have scores between 0 and 192, is calculated to reflect the general level of symptoms and problems. Norms for a heterogenous German sample showed values between 26 and 55 for the 34^th to 66^th percentile.

The availability of social resources was assessed with the short version of the German Social Support Questionnaire [FSozU K-22; 37]. The FSozU K-22 includes 22 statements about the perceived and anticipated availability of social resources to be rated on a 5-point likert scale. It covers the dimensions emotional and practical support and social integration. A score is calculated as mean of the sum of all items and can have values between 1 and 5. Norms for a representative German sample showed a mean of 4 (SD 0.66) and values from 3.69 to 4.41 for the 30^th to 68^th percentile.

At post-intervention assessment, two questionnaires were administered to the intervention condition to evaluate perceived group climate and therapeutic alliance:

The German version of the Group Climate Questionnaire—Short Form [GCQ-S; 38] was employed to assess perceived engagement (i.e. group cohesion, cognitive understanding, self-disclosure and empathy) and conflict (i.e. anger, detachment, confrontation and mistrust) with 4 items each. Both scales are calculated as mean values of the four corresponding items and can have values between 0 and 6.

The German version of the Working Alliance Inventory—short revised [WAI-SR; 39] was employed to asses the therapeutic alliance, i.e. agreement on tasks, agreement on goals and development of an affective bond. The 12 items of the WAI-SR were rated on a 5-point scale from 1 to 5. The score represents the mean of the 12 items and can have a value between 1 and 5. The wording of the questionnaire was adapted to fit our intervention [e.g. 40].

Furthermore, to assess physiological and emotional responses during and to the TSST, the following parameters were assessed:

To assess state anxiety before and after the TSST, the German version of the State Trait anxiety inventory [STAI; 35] was administered before the TSST (-45 minutes), after the introduction to the TSST (-20 minutes), immediately after the TSST (0 minutes) and in the recovery phase (50 minutes).

To assess salivary free cortisol levels, nine saliva samples were collected using Salivette collection devices (Sarstedt, Sevelen, Switzerland) at -45, -35, -25, -10, 0, 10, 20, 35 and 50 minutes. Sampling time lasted approximately 1 minute during which subjects chewed on the cotton swabs as regularly as possible. Salivettes were stored at −20°C until biochemical analysis took place. After thawing, biochemical analyses were conducted in the bio-chemical laboratory of the Clinical Psychology and Psychotherapy department at the University of Zurich, Switzerland, by means of a highly sensitive liquid chromatography–tandem mass spectrometry (LC–MS/MS) method [41]. Since the use of oral contraceptives and the menstrual cycle phase has been shown to influence the activity of the HPA-axis and therefore the endocrine response to stress [42], participants reported the use of oral contraceptives, menstrual cycle length and the first day of their last menses. Based on these informations, menstrual cycle phase, i.e. follicular or luteal phase, on the day of the TSST was estimated. The follicular phase was defined as the period between the first day and 14 days before the end of the menstrual cycle, while the luteal phase was defined as the last 14 days of the cycle [43].

Electrocardiography (ECG) was recorded continuously at 1000 Hz during 1 hour and 40 minutes using the wireless physiological recording system BioNomadix® (Biopac Systems, Inc., Goleta, CA). Recorded ECG data were filtered using the software AcqKnowledge (Biopac Systems, Inc., Goleta, CA) with a FIR bandpass filter from 0.5 Hz to 35 Hz with 8000 coefficients. The resulting heart period series were visually examined for artifacts and corrected when necessary using the VivoSense® software (Vivonoetics, Inc., San Diego, CA). To assess changes in heart rate over the course of the experiment 7 mean heart rate slots of 5 minutes each were calculated. Slots represented the mean heart rate at baseline (-30 to -25 minutes), in the preparation phase of the TSST, during the free speech in the TSST, during the mental arithmetic task in the TSST and in the recovery phase with three slots starting 5, 15 and 25 minutes after the TSST ended. Furthermore, heart rate variability was calculated as RMSSD, the square root of the mean of the sum of the squares of differences between adjacent normal-to-normal intervals. This parameter is one of the most commonly used measure derived from interval differences [44]. Shaffer and Ginsberg [45] presented norms for short-term measurements (around 5 minutes) of RMSSD with a range of 19 to 75 and a mean of 42 milliseconds (SD 15).

Deviating from the study protocol we did not perform assessments of alpha-amylase and electromyography. Data of electrodermal activity, skin temperature, and respiratory activity were not analyzed. Analysis of two questionnaires (PASA and BFW/E) were not reported.

Statistical analysis

All calculations were carried out using the statistic software IBM^® SPSS^® Statistics Version 27. Two sample t-tests were calculated to compare demographic characteristics and baseline values. Associations of age with psychometric or physiological measures were calculated as Pearson correlation coefficients.

For the comparision of psychological and physiological variables between the two conditions over the course of the study and the TSST (condition by time interaction), separate covariance pattern models were employed using the mixed procedure of SPSS. Covariance pattern models allow to deal with missing data and the correlation between repeated measures and were therefore used instead of the previous planned MANCOVA. Normal distribution and homoskedasticity of the residuals were visually inspected and tested using the Kolmogorov-Smirnov and the Levene’s test. When normality and homoskedasticity were not met, calculations were repeated with Box-cox transformed data. For better readability, the original values were used for tables and figures.

Covariance pattern models for each outcome were built in the following order: 1) A basic model (model 1) was built including time, condition (i.e. intervention and control), and the interaction of condition by time as fixed effects and time as a repeated effect on the level of the participant. 2) As participants were randomly assigned to one of the two conditions, identical baseline values were assumed for the psychological variables at pre-intervention (t0). Therefore, a model 2 was built for psychological variables including post(t1), follow-up time(t2), the interaction of condition and t1, and the interaction of condition and t2 as fixed effects (based on [46], p. 128 ff). As in model 1, time was defined as repeated effect. 3) Effects of the possible covariates age or BMI were tested by adding them to the model 1 or 2 as fixed effects and comparing model fit of the models including age or BMI to model fit without them. 4) To test the primary interest, i.e. differences of the two conditions over the course of the measurements, a simplified model including the same predictors as model 1 or 2 except the interaction of condition by time was built and model fit was compared to model 1 or 2. When there was a significant model fit improvement in the model with the interaction effect, estimations of unstandardized regression coefficients were reported. The effect size Cohen’s d was calculated by dividing the unstandardized regression coefficient with pooled within-group standard deviation of the outcome [47].

All models employed maximum likelihood estimation. Model fit differences were tested with the likelihood ratio test. To achieve optimal fit whilst reducing the parameters, the covariance structure of the repeated effect was set to compound symmetry in models for psychological variables measured pre-, post and follow-up and to first-order ante dependence for psychophysiological variables measured before, during and after the TSST. Due to bad ECG data quality ten subjects had artifact corrections between 3 and 10%. Heart rate and RMSSD analyses were repeated with these 10 subjects excluded, but results did not differ (data are not shown).

For cortisol and STAI state responses areas under the response curve were calculated with respect to increase (AUCi) using the trapezoidal method as an indicator for the integrated response in the TSST [48]. As we were interested in the reaction of cortisol to the TSST, the first three cortisol measurements which took place before the introduction of the TSST were simplified to one baseline measure by calculating their mean. There were two values missing in assessments of the STAI state due to data transfer interruptions which were replaced by multiple imputations (20 iterations). Pooled parameters for the testing of condition differences in AUCi STAI state were reported.

Distribution between conditions in use of hormonal contraceptives, menstrual cycle phase and frequency of non-responders were tested with Chi-square tests of independence. Descriptive statistics are presented for the GCQ and WAI measures.

For all analyses, significance level was set at α = 5%.

Psychophysiological stress responses

Cortisol responses over time did not differ significantly between conditions, shown by no significant improvement of model fit when the interaction condition by time was added to the model (model included BMI as covariate, χ2 (6) = 9.80, p = .13, Box-Cox transformed values χ2 (6) = 8.25, p = .22; see Fig 2) and no differences between conditions in the integrated cortisol response AuCi: intervention condition M = 35.50, SD = 199.81; control condition M = 40.39, SD = 169.81, F(1, 34) = .004, p = .95). Conditions did not differ in the prevalence of cortisol non-responders (cortisol increase to baseline < 1 nmol/l: control condition N = 7/41%; intervention condition N = 9/45%, χ2 (1) = .06, p = .82). Also, there was no significant condition by time interaction effect in the model for heart rate (model included BMI, χ2 (6) = 3.98, p = .68, Box-Cox transformed values χ2 (6) = 2.30, p = .89) or RMSSD responses to the TSST (model included age, χ2 (6) = 8.22, p = .22, Box-Cox transformed values χ2 (6) = 7.28, p = .30; see Fig 3). Regarding STAI state anxiety responses to the TSST, there was no significant improvement in model fit when the interaction condition by time was added (model included age, χ2 (3) = 4.03, p = .26; Box-Cox transformed values χ2 (3) = 3.89, p = .27;); see Fig 2). However, participants in the intervention condition showed a significant lower integrated response in state anxiety in comparison to controls (pooled results from multiple imputations for two missing values: AuCi intervention condition M = 191.71, SEM = 120.13; control condition M = 619.91, SEM = 217.81, F(1, 34) = 4.38, p = .045, partial η² = .11; original data: AuCi intervention condition M = 153.21, SEM = 118.21; control condition M = 592.63, SEM = 227.55, F(1, 32) = 5.51, p = .025, partial η² = .15). All variables measured over the course of the TSST showed significant changes over time (fixed effect of time: cortisol F(6, 78.56) = 4.41, p = .001; heart rate F(6, 62.11) = 24.32, p < .001; RMSSD F(6, 60.89) = 19.98, p < .001; STAI state F(3, 51.41) = 24.92, p < .001).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Salivary free cortisol and state anxiety responses in the Trier Social Stress Test between conditions.

Values are means and standard error means.

https://doi.org/10.1371/journal.pone.0252568.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Heart rate and RMSSD responses in the Trier Social Stress Test between conditions.

Values are means and standard error means.

https://doi.org/10.1371/journal.pone.0252568.g003

Effects on perceived stress

Comparing model 2 with and without interaction of condition by time showed a significant improvement of model fit for PSS when the interaction was included (model included age, χ2 (1) = 7.91, p = .005). The effect of condition by time was significant at post-intervention with the intervention condition showing a decline in PSS values (b = -4.70, 95%CI [-8.43; -.97], F(1, 95.97) = 6.26, p = .014, d = -.60) but not at follow-up (b = .90, 95%CI [-2.96; 4.76], F(1, 95.34) = .22, p = .64). The analysis was repeated with no restrictions regarding minimal attendance of the intervention and did show a less pronounced but still significant effect for model fit improvement when the interaction effect was added (N = 47, χ2 (1) = 4.66, p = .031, condition by time at post-intervention: b = -4.08, 95%CI [-7.94; .21], F(1, 100.03) = 4.38, p = .039; condition by time at follow-up: b = .46, 95%CI [-3.52; 4.45], F(1, 99.33) = .05, p = .82).

There was no indication for differences in conditions over the course of the study in any other psychological variables, as the inclusion of the condition by time interaction did not lead to a significant improvement in model fit (STAI trait: χ2 (1) = .66, p = .42; ASS-SYM: χ2 (1) = 1.32, p = .25; FSozU K-22, model with age: χ2 (1) = 2.94, p = .09; see Table 1).

Evaluation of the SSSM

Results of the GCQ-S indiated high levels of engagement (M = 4.93, SD = 0.62) and very little interpersonal conflict (M = 0.49, SD = 0.63). Furthermore, the overall alliance (WAI-SR) with therapists (M = 3.40, SD = 0.58) and participants in the SSSM condition (M = 3.41, SD = 0.54) was good and comparable. On the three subscales, development of an affective bond was rated high (M = 4.05, SD = 0.58 for therapists, M = 3.98, SD = 0.34 for participants), ratings for agreement on tasks and agreement on goals were one point lower on the likert scale (task M = 3.08, SD = 0.87 for therapists, M = 3.18, SD = 0.84 for participants, goal M = 3.09, SD = 0.70 for therapists, M = 3.08, SD = 0.81 for participants).