Based on a sample of refugees and migrants from Syria and other Arabic-speaking countries, we examined in the current work whether stress resonance would constitute a vulnerability factor for developing PTSD symptoms after war-related trauma. Next to subjective stress resonance, we focused on three different markers of the stress systems: (1) cortisol as a neuroendocrine marker of the HPA axis, as well as (2) heart rate and (3) heart rate variability as physiological markers of sympathetic and parasympathetic nervous system (SNS, PNS) functioning. As a strong paradigm that reliably triggers stress resonance even in strangers, we used the empathic Trier Social Stress Test (TSST [16, 24]). In this paradigm, native Arabic-speaking participants watched native German-speakers undergo a standardized psychosocial stress test. The different stress markers were simultaneously captured in the stressed targets and passive observers. In a previous data set stemming from the same study, we found a moderate positive association between trauma exposure and PTSD symptoms [25]. In the present work, we draw on data from a subsample of this earlier study. Based on the above literature, we expected that stress resonance would moderate the link between trauma exposure and PTSD symptomatology, such that the positive association between trauma and PTSD symptoms would be stronger in individuals exhibiting higher stress resonance.
Our sample consisted of N = 66 targets (55 females) taking part in a stress test, and N = 67 observers of the opposite sex (11 females) watching the targets during the stress test (opposite-sex dyads were chosen as per our usual approach [16]). In the first testing session, two observers simultaneously watched one target. Given implementation of COVID-19 contact restrictions after testing of this first triad, all subsequent testing sessions were executed in a one-to-one observation set-up (i.e., dyads). Targets were German (age M = 26.0 years, SD = 5.22 years), and observers Arabic native speakers (age M = 28.7 years, SD = 4.85 years). Of the observers, 35 had entered Germany as refugees (age M = 30.4 years, SD = 4.59 years) and 32 as migrants (age M = 26.9 years, SD = 4.53 years). The majority of refugees and migrants came from Syria (88.6 and 71.9%, respectively; see Table S1 for countries of origin).
Participants were recruited through social media advertisement and by distributing flyers. Trained students conducted a structured telephone screening to determine participants' eligibility. All participants needed to be aged between 20 and 40 years. Targets were required to speak German as the native language. Observers needed to speak Arabic as the native language and German at an intermediate (B1) level, come from an Arabic-speaking country of origin, and live for at least six months in Germany. Within observers, we further differentiated between refugees and migrants based on definitions of UNHCR [8]: Individuals who had been forced to flee their home countries because of conflict or persecution were considered as refugees. In contrast, individuals who had chosen to leave their home to work, study, or join family in a new country were considered as migrants. Refugees were included if they reported a war-related trauma (i.e., they had fled war, violence, or persecution) in the absence of other major trauma (e.g., child maltreatment, severe accident; queried based on a list of traumatic events compiled by the authors). In contrast, migrants were supposed to be free of any major traumatic experience. Given known effects of weight on autonomic activity [28], an additional exclusion criterion for participants of both target and observer groups was a BMI of less than 18 or more than 30. We further excluded participants who reported chronic illnesses, psychiatric disorders (e.g., anxiety disorders), or the regular use of medication, hormone-based birth control, or other substances (e.g., alcohol, drugs) known to affect HPA axis or autonomic activity [29, 30]. Refugees were not excluded if they had a diagnosis of PTSD or depression during the last two years. To facilitate recruitment, we further decided to include occasional and regular cigarette smokers in the observer group due to high prevalence of smoking in many Middle Eastern countries [31].
The study was approved by the Ethics Board of the medical faculty of Leipzig University, Germany (ethics number: 405/18-ek). Observers came to the lab for a first visit lasting about 2.5 h, in which they provided informed consent, filled out online questionnaires, and completed an empathy task (EmpaToM [32]; not subject to the current manuscript). For a second lab visit lasting about 4.5 h, observers were invited back together with an unfamiliar target person of opposite sex. This visit was scheduled in the early afternoon to control for variation in diurnal cortisol secretion [33].
Upon arrival at the laboratory, targets and observers were placed in separate rooms. Participants first underwent a rapid drug test to screen for recreational drug use. To equalize blood sugar levels, they were offered a snack and a glass of juice. To avoid saliva sample contamination, participants were instructed to not eat or drink anything other than water throughout the remainder of the testing session. Then, participants were equipped with a chest belt recording an electrocardiogram (ECG) to measure SNS and PNS activity. After a resting period of 40 min, participants were guided into either the testing or the adjacent observation room. While targets attended the TSST [24], observers passively watched the procedure (preparatory anticipation and stress phase) through a one-way mirror. Targets were aware of being observed during their performance, although they did not know by whom. After the TSST, target and observer separately left their rooms, and rested alone during the 60 min recovery phase.
The TSST is a standardized laboratory paradigm that reliably elicits social-evaluative stress [34, 35]. It consists of a preparatory anticipation phase (5 min), followed by a video-recorded mock job interview (5 min) and a mental arithmetic task (5 min) in front of an evaluation committee. Given safety rules due to the COVID-19 pandemic, for 60.6% of the testing sessions, the committee members were sitting in a separate room from the participant. Their image was projected via live video feed at large-scale on the wall facing the target. During the TSST instruction period, the committee members shortly entered the TSST testing room to demonstrate their actual presence in the laboratory and thus the realness of the video situation.
Throughout the testing period, ten ratings of subjective stress experience and ten cortisol samples were collected simultaneously from targets and observers. In addition, a continuous ECG was collected from 50 min prior to 50 min post stressor onset (for assessment timeline, see Fig. 1). Participants received financial compensation for study participation. The study was performed in agreement with the Declaration of Helsinki. All participants gave their written informed consent and could withdraw from the study at any time.
In both targets and observers, we assessed participants' subjective stress experience at ten times throughout the test session using a 7-point Likert scale asking for the level of currently experienced stress ("How stressed do you feel at this moment?"). Assessments took place 20 min prior to stressor onset (baseline), after stress anticipation (2 min prior to stressor onset), and at 10, 20, 25, 30, 40, 50, 60 and 70 min after stressor onset (see Fig. 1).
In parallel to the assessments of subjective stress experience, participants collected ten salivary samples for cortisol analysis using Salivettes (Sarstedt, Nümbrecht, Germany; Fig. 1). Participants placed the collection swab in their mouth for 2 min and refrained from chewing. Salivettes were stored at -20 °C until analysis. Cortisol levels were assessed using a time-resolved fluorescence immunoassay with intra- and inter-assay variabilities of less than 10 and 12%, respectively [36].
All participants wore a Zephyr Bioharness 3 chest belt (Zephyr Technology, Annapolis, Maryland, USA), which recorded a continuous ECG at a frequency of 250 Hz for altogether 100 min (from -50-+50 min relative to stressor onset). A trained student manually corrected artifacts in the ECG raw data using python-based in-house software. Two students re-checked all corrections made. For further analysis, each ECG recording was split into 5-min timeframes related to specific phases of the testing protocol (Fig. 1). The baseline phase was represented by timeframes 1 and 2 (from -40 to -30 min), the anticipation phase by timeframe 3 (from -7--2 min), the stress phase by timeframes 4 and 5 (from 0-+10 min), and the recovery phase by timeframes 6-12 (from +15-+50 min). For each timeframe, we calculated averages for heart rate (HR) and heart rate variability (HRV; specifically the root mean of the square successive differences [RMSSD]) per participant using the python package "hrv-analysis" [37]. To ensure sufficient data quality, we excluded timeframes in which more than 10% of the recording had to be cut.
Trauma was measured with the official Arabic ("Iraqi") version of the trauma events section of the Harvard Trauma Questionnaire (HTQ [26, 38]). Participants indicated whether they had experienced each of 42 traumatic events (e.g., oppression, imprisonment, combat exposure) before coming to Germany. Trauma scores were calculated by a sum of "yes" responses. In our study, the trauma events section of the HTQ demonstrated excellent reliability (Cronbach's α = 0.92).
We assessed PTSD symptom severity with the first 16 items of the trauma symptom section of the HTQ [38], which correspond to symptoms of PTSD according to the DSM-IV. Participants indicated the degree to which they were distressed in the past week by trauma symptoms such as "feeling detached or withdrawn from people" or "trouble sleeping" on a Likert scale ranging from 1 ("not at all") to 4 ("extremely"). Higher mean scores indicated more severe PTSD symptoms. The HTQ manual recommends a cut-off score of 2.5 to identify clinically significant levels of PTSD [26], which was the case in five observers. Reliability was good (Cronbach's α = 0.89).
As a specific facet of empathy potentially linked to stress resonance, we assessed trait personal distress based on the Interpersonal Reactivity Index (IRI [39]). The IRI was translated into Arabic by a native, bilingual speaker and back-translated by another bilingual individual to ensure linguistic equivalence [40]. The personal distress scale measures self-oriented feelings such as discomfort, anxiety and worry when being confronted with others' negative emotional states or situations. It consists of seven items rated on a Likert scale from 1 ("does not describe me well") to 5 ("describes me very well"), aggregated to a sum score. In the current study, reliability was relatively poor (Cronbach's α = 0.49).
To approach normal distribution, cortisol, HR, and HRV data were log-transformed and subsequently winsorized to three standard deviations [29]. For all participants and stress markers, we calculated change scores from baseline to stress peak. While change scores indicated stress reactivity in targets, they reflected vicarious stress (i.e., stress reactivity independent of target stress) in observers.
For cortisol, the baseline level (sample 1 at -20 min) was subtracted from the individual peak level chosen from samples at +10, +20, +25, +30, or +40 min (samples 3-7), since cortisol typically peaks within this timeframe after stressor onset [41, 42]. A significant physiological stress response was defined as a cortisol increase of at least 1.5 nmol/l above baseline levels [43]. For HR, the mean of timeframes 1 and 2 (baseline) was subtracted from the maximum of timeframes 3 (anticipation), 4 and 5 (stress task). For HRV, the minimum of timeframes 3 - 5 was subtracted from the mean of timeframes 1 and 2 (baseline). This was done because other than HR, HRV decreases with stress experience [44]. Hence, higher change scores indicated stronger increases in cortisol and HR, and stronger decreases in HRV in response to stress. If HR / HRV data was only available for the anticipation phase, but missing for the stress phase, we refrained from computing a change score, as it was impossible to verify whether the anticipation sample represented peak reactivity. For subjective stress, we subtracted the baseline rating (at -20 min) from the peak level chosen from ratings at -2 and +10 min. Since baseline levels can affect the reactivity of physiological parameters (law of initial value [45]), we adjusted cortisol, HR, and HRV change scores for baseline levels by extracting the standardized change score residuals from a regression model. Residualized change scores were chosen rather than percentage change scores as the latter are highly sensitive to baseline levels (i.e., particularly small baseline levels can yield exaggerated percentage changes).
To gain a measure of physiological and subjective stress resonance, we subtracted observer change scores from the corresponding target's change scores, and computed an absolute value of this difference. By doing so, we obtained a measure of how closely the observer's stress reactivity resembled the one of the target. The absolute difference score was multiplied by (-1), such that higher values indicated higher stress resonance.
Data on vicarious stress and stress resonance was complete for subjective stress and cortisol. Due to technical issues, HR and HRV data was missing for two observers and two targets (of whom one observer and one target belonged to the same dyad). In addition, HR and HRV data quality was too low (i.e., >10% needed to be cut) in one observer during both baseline and stress timeframes, and in one target during both stress timeframes, precluding the computation of change scores. Thus, overall, data on HR and HRV vicarious stress was missing for three observers, and data on HR and HRV stress resonance was missing for five observers.
We first examined the number of targets and observers showing a significant cortisol stress response (cortisol increase >1.5 nmol/L from baseline [43]). To detect possible differences in our variables of interest between refugees and migrants, we compared both groups with regard to their vicarious stress responses and stress resonance across all four stress markers, as well as regarding trauma experience and PTSD symptoms.
Regression analyses were carried out to test for the relation between stress resonance and PTSD symptoms. We initially planned to conduct multiple linear regression analyses, but because assumptions were violated in most models (e.g., presence of heteroscedasticity/unequal variances of residuals), we applied multiple logistic regressions. To this end, we created a dichotomous outcome measure of PTSD symptoms above and below the sample mean. All continuous predictors were z-standardized to handle possible issues of multicollinearity. For each stress marker, the base model included trauma and the covariates sex and age as predictors of PTSD symptoms. In a second step, we tested whether adding stress resonance (operationalized as the absolute value of the difference between target and observer change scores) and its interaction with trauma would improve model fit, which was evaluated by comparing both models using analysis of variance. This procedure resulted in one model per stress marker (subjective stress, cortisol, HR, HRV), yielding four final models. Given multiple testing, we applied Bonferroni correction, such that p-values were compared to a significance threshold of α = 0.0125.
Due to significant correlations between observers' stress resonance and vicarious stress for subjective stress, HR, and HRV, we re-computed analyses by introducing vicarious stress instead of stress resonance as a moderator to the regression models. Further, since the drug screening conducted before the TSST was positive for n = 4 targets (6.1%) and n = 3 observers (4.5%), we re-computed analyses by excluding the seven dyads in which either target or observer had a positive test result.