Stress-Induced Cortisol Reactivity as a Predictor of Success in Treatment for Affective Dimensions

Abstract

Background:

Response rates to first-line treatments for depression and anxiety remain unsatisfactory. Identification of predictors or moderators that can optimize treatment matching is of scientific and clinical interest. This study examined the role of prolonged laboratory-induced stress cortisol reactivity as a predictor of outcome for a treatment of affective dimensions (TAD). Patients received 15-sessions of a treatment targeting reductions in negative affect or increases in positive affect (Craske et al., 2019). A second aim was to examine whether treatment type would moderate the association between cortisol reactivity and treatment outcome.

Methods:

Thirty-five participants underwent a 36-minute intermittent stress induction task composed of a mental arithmetic task and a fear-potentiated startle task one week before treatment initiation. Cortisol was collected at five-time points with reactivity being quantified as peak during the task minus basal level of cortisol the evening before the assessment. Using multilevel modeling, we examined the associations between cortisol reactivity and slopes of symptom improvement.

Results:

Cortisol reactivity was related to treatment outcome, with average and higher levels of stressinduced cortisol response predicting greater decreases in symptoms throughout treatment and 6-month follow-up. Treatment condition differences (moderation) were not observed in the effect of cortisol reactivity on symptoms.

Conclusion:

Our findings demonstrate the impact of greater cortisol stress reactivity on treatment outcome. Future studies should investigate how to enhance this therapeutic benefit through capitalizing on endogenous diurnal fluctuations or exogenous cortisol manipulation.

1.1. Introduction

First-line treatments for depression and anxiety disorders remain unsatisfactory due to high non-response and drop-out rates (Carpenter et al. 2018; Cuijpers et al., 2018; Gaynes et al., 2008; Loerinc et al., 2015). The identification of predictors and moderators that may facilitate treatment selection with the aim of more personalized treatments is therefore of great scientific and clinical interest and has been a focus of the NIMH Research Domain Criteria framework (Insel, 2014; Kazdin, 2007). Cortisol has been discussed as a marker of distress and investigated for its pathophysiological relationship to depression, anxiety, stress, and trauma (Dhama et al., 2019; Fries et al., 2009; Heim et al., 2008; Islam et al., 2018; Miller, Chen, & Zhou, 2007; Rauch et al., 2015; Xu et al., 2018; Zorn et al., 2017). It is a product of the hypothalamic-pituitary-adrenal axis (HPA-axis) and helps mobilize the organism in response to challenges by assembling and restocking energy stores and containing the immune response to maintain allostasis (Boucher and Plusquellec, 2019; Haglund et al., 2007). Although cortisol reactivity is a natural part of the endocrine stress response, it can be detrimental to long-term health if it translates into chronically elevated or reduced levels (Kumari, Shipley, Stafford, & Kivimaki, 2011; Lovallo, 2015). Altered cortisol reactivity has been associated with depression (Burke et al., 2005; Fiksdal et al., 2019; Morris et al., 2013, 2017; Trueba et al., 2016; Zorn et al., 2017), anxiety (Wichmann et al., 2017a; Wintermann, Kirschbaum, & Petrowski, 2016), and stress (Bunea, Szentágotai-Tătar, Miu, 2017; Goldman-Mellor, Hamer, & Steptoe, 2012; Mazurka et al., 2016).

Although the direction of the response of cortisol reactivity (hyper- or hypo-reactivity) varies across studies and even within related presentations of psychopathology, assessing its impact on treatment could be beneficial. Specifically, recent studies have shown that higher pre-treatment cortisol reactivity to standardized stressors predict a more favorable treatment outcome. Dieleman et al. (2016) observed that higher pre-treatment cortisol reactivity to an acute stressor (a mental arithmetic task and social competence interview) predicted more significant improvements in depressive symptoms following cognitivebehavioral therapy (CBT) in anxious children. Whereas no significant relationship between cortisol reactivity and anxiety symptoms was found in this study, another study demonstrated that higher cortisol reactivity to a psychosocial stressor (the Trier Social Stress Test [TSST], Kirschbaum et al., 1993) prior to 5 weeks of CBT for individuals with panic disorder predicted superior reductions in agoraphobic avoidance at post-treatment (Wichmann et al., 2017b). There is also evidence that stress-induced cortisol reactivity is beneficial for those experiencing trauma symptoms. For example, Rauch et al. (2015) found that higher pre-treatment cortisol responses to script-driven imagery predicted superior treatment improvement in prolonged exposure for Post-Traumatic Stress Disorder (PTSD). By contrast, higher cortisol reactivity to a pre-treatment presentation of virtual-reality-based combat stimuli was related to significantly worse outcomes at post-treatment following six sessions of virtual reality exposure therapy for PTSD (Norrholm et al., 2016); but no association was observed six months after treatment. Relatedly, cortisol reactivity during in-vivo exposures was unrelated to outcome in those with panic disorder (Meuret et al., 2015).

Other studies have examined the effects of cortisol administration (cortisone) on fear reduction during exposure in social anxiety disorder, specific phobia, and PTSD (de Quervain et al. 2009; Soravia et al., 2006, 2014). Higher cortisol levels during exposure (due to exogenous cortisol administration but also endogenous cortisol levels in and of themselves) were related to significantly greater fear reductions. Similarly, higher levels of endogenous cortisol, assessed during a pre-treatment stress test before treatment for those with trauma symptoms (Zantvoord et al., 2019) and during in-vivo exposures for those with panic disorder and agoraphobia (Meuret et al., 2015; Siegmund et al., 2011), also predicted better treatment outcomes. Mechanisms by which cortisol exerts a positive effect on exposure-based outcomes include: dampening stimulus-elicited fear, increasing approach behaviors to feared stimuli, impairing fear-based memory retrieval when exposed to a salient cue, and enhancing memory consolidation when undergoing new learning (Soravia et al., 2006, 2013; de Quervain et al., 2011; 2019; Lass-Hannemann & Michael, 2014; Singewald et al., 2015).

Notwithstanding, the predictive value of cortisol reactivity during exposure and benefits in psychotherapy for depression and anxiety is less clear. Two recent meta-analysis by Fischer et al. (2017) and Fischer and Cleare (2017) showed none or a negative impact of cortisol on outcome; while others showed a positive effect (Meuret et al., 2005; Alpers et al., 2003). One reason could be anticipatory anxiety prior to exposures lead to higher pre-exposure levels, creating a ceiling effect. Therefore, cortisol reactivity assessed before treatment onset, as opposed to before an exposure session, may be a more promising predictor of treatment success. There has been ample debate whether fear activation, within the context of exposure therapy, is sufficient to activate the HPA-axis acutely and create a cortisol response (Hjortskov et al., 2004). For example, one study in which patients approached feared animals or situations (spider, snake, and claustrophobia), showed no cortisol reactivity during the actual exposure (Mayer et al., 2017). However, when comparing levels to the ones of non-exposure control days, cortisol may indeed be elevated (a form of reactivity due to anticipatory anxiety). This effect was illustrated in exposure therapy for agoraphobia (Schumacher et al., 2014), panic disorder (Meuret et al., 2015), and driving phobia (Alpers et al., 2003) all of which found elevated cortisol during exposure (in addition to before and after in the latter two studies) when compared to non-exposure control days. In line with this, cortisol secretion by the HPAaxis is thought to act as a mechanism to prepare us for meaningful events by maintaining allostasis (Boucher and Plusquellec, 2019).

The aim of the current study was to replicate previous findings of the impact of pre-treatment stressinduced cortisol reactivity on treatment benefits; moreover, to extend beyond prior studies by using a transdiagnostic sample as well as using a prolonged laboratory stressor. A second aim and extension was to evaluate whether pre-treatment cortisol reactivity would moderate the therapeutic benefits of different types of interventions. Participants were randomly assigned to either an intervention aimed at improving positive affect by explicitly targeting the reward sensitivity system (Positive Affect Treatment or PAT), or to an intervention aimed at reducing negative affect by explicitly targeting the threat sensitivity system, Negative Affect Treatment (NAT; Craske et al., 2016, 2019). Based on prior research on exposure treatment and fear extinction, we expected that higher pre-treatment cortisol reactivity would be associated with greater treatment gains, particularly in an intervention designed to target reductions in negative affect (NAT). We hypothesize that NAT may lead to additional cortisol mobilization and memory consolidation of extinction learning (Craske et al., 2014; Singewald et al., 2015) throughout treatment (but specifically during the first half of treatment which features exposure).

1.2. Material and Methods

1.2.1. Participants

Participants were recruited from the Dallas-Fort Worth and Los Angeles metroplexes for a randomized controlled treatment study comparing two forms of treatment, increasing positive affect (Positive Affect Treatment, PAT) or decreasing negative affect (Negative Affect Treatment, NAT [Craske et al., 2016, 2019]). Another goal was to identify behavioral indicators and biologically-based characteristics of threat sensitivity and reward sensitivity that would moderate the effects of NAT and PAT. Eligible participants had to meet elevated scores on any of the three subscales of the Depression Anxiety and Stress Scale-21 (DASS), represented by either a score ≥ 11 on the depression subscale, ≥ 6 on the anxiety subscale, or ≥ 10 on the stress subscale (Brown, Chorpita, Korotitsch, & Barlow, 1997). Participants also had at least a moderate level of disability on any one of the three Sheehan Disability subscales, which included work or school, social life, and family life including home responsibilities (Sheehan, HarnettSheehan & Raj, 1996). Other entry criteria were age 18 to 65, English-speaking, refrained from other psychosocial treatment and stabilized on psychotropic medications (1 month for beta-blockers and benzodiazepines, three months for selective serotonin reuptake inhibitors and heterocyclics), and willing to refrain from starting or modifying pharmacological treatments until the 6-month follow-up. Exclusion criteria were severe medical and psychiatric conditions, such as neurological, muscular-skeletal diseases, uncontrolled hyper- or hypothyroidism, uncontrolled high blood pressure, history of seizures or epilepsy, severe asthma, and chronic obstructive pulmonary disease, active suicidal ideation or self-harm within the past year or a previous suicide attempt, a history of Bipolar I or II disorder, cyclothymic disorder, schizophrenia-spectrum disorder, intellectual disability, organic brain damage, substance use disorder within the last six months, and current pregnancy.

The present analysis included all participants with a viable baseline cortisol sample the night before the pre-treatment psychophysiological assessment and a viable cortisol sample during the pre-treatment assessment (SMU n=22 and UCLA n=13)¹. An additional 23 participants completed the pre-treatment assessment, but were not included due to a) missing the evening-before cortisol sample (n=14), b) lacking a viable cortisol sample during the assessment (n=1) or c) missing both the evening and laboratory session cortisol samples (n=4) or d) missing all DASS scores (n=4). The study was approved by the ethics committees of both universities and informed consent was obtained from all participants. Participants received compensation for posttreatment and follow-up assessments ($50 and $75, respectively).

1.2.2. Procedure

Prior to treatment, eligible participants completed two 3-hr laboratory assessments, approximately one week apart, to examine whether psycho-biobehavioral characteristics moderate treatment outcomes. The laboratory sessions included a battery of tasks to capture emotion processing, fear learning processes, reward sensitivity, and respiratory dysregulation. The 36-minute intermittent stress induction tasks described here took place during the second laboratory session, completed on average one week before the first treatment session.

Before the stress induction period, participants did a baseline and paced breathing respiration assessment, a fear extinction task, and a reward sensitivity task. The first task of the 36-minute stress period was a standard mental arithmetic task that had been used previously in cardiovascular reactivity research and has shown to elicit anxiety (Vögele & Steptoe, 1992). The task was preceded by a 3-min baseline and followed by a 3-min recovery period. Participants were asked to add and subtract single and double-digit numbers on a computer screen while listening to background noises (voices of people having indistinguishable conversations in a large room) through in-ear headphones. At the end of the task, the experimenter asked the participants the final number and solution. To increase social evaluate threat, participants were told that most people made it toward the end of the sequence. The second task was a fear-potentiated startle (FPS) task, which involved the repetition of eight “safe” and “danger” sequences on a computer screen (Craske et al., 2012; Grillon, Ameli, Foot, & Davis, 1993). During each of the sequences, the screen displayed a graphic of a timer bar that counted the 55 seconds. The progressing timer graphic and screen was colored green in the “safe condition” and red in the “danger condition”. Participants were told that delivery of an aversive stimulus (an electric stimulation to their bicep delivered by two electrodes) could occur during the final 15 seconds of the “danger” sequences, but never in the “safe” condition. To emphasize the final 15 seconds of the “danger” condition, a darkening redness appeared to the timer. Additionally, two auditory startle probes were presented in each safe and each danger condition, at 5 and 35 or 15 and 45 seconds, resulting in 32 total startle probes over the eight sequences. The task lasted 25 minutes, during which participants eventually received only one electric stimulation at the end of the 4^th danger sequence.

Unlike other cortisol reactivity laboratory-induced stressor tasks, which typically last approximately 15 minutes (including the Trier Social Stress Test, TSST [Kirschbaum, Pirke, & Hellhammer, 1993]), the current combined intermittent stress tasks lasted for 36 minutes. The long, intermittent stressor in the present study was designed to be more representative in terms of length and intermittent nature of everyday life stressors. We expected that the stress tasks would elicit a strong cortisol response due to having both elements of social-evaluative threat and uncontrollability, as also found in the TSST (Dickerson and Kemeny, 2004), which uses a similar mental arithmetic task.

1.2.4. Clinical outcome

The treatment outcome measure was the DASS (Brown et al., 1997). The DASS is a 21-item measure of depressive, anxiety, and stress symptoms. The DASS depression subscale evaluates constructs such as lack of interest/involvement, anhedonia, and dysphoria. The anxiety subscale assesses situational worries and autonomic arousal. The stress subscale evaluates irritability, difficulty resting, and nervous arousal in stress. The DASS has adequate psychometric properties, including Cronbach’s α=0.97. Participants responded on 4-point scales ranging from 0 (did not apply to me at all) to 3 (applied to me very much, or most of the time) on the degree to which the statements applied to them over the past week. Participants completed this measure before treatment, at the beginning of every treatment session, post-treatment, and at the 6 months follow up after the last session. For our study, we used the total score on the DASS because it better captures a variety of symptoms in a transdiagnostic sample. Further, the total score on DASS is a reliable and valid measure of general distress (Gloster et al., 2008; Page, Hooke & Morrison, 2007).

1.2.5. Salivary cortisol

Cortisol was collected using Salivettes® tubes with cotton swabs (Sarstedt, Germany). Participants placed the swab in their cheek for two minutes while making a chewing motion to capture saliva. Salivary cortisol was collected four times during the stress tasks (before the mental arithmetic task, before and after the FPS task, and approximately 10 minutes after FPS task), in addition to the evening and morning before (awakening and +30 min) the cortisol session. Cortisol reactivity was quantified as peak cortisol during the stress tasks minus the basal level of cortisol the evening prior. The latter sample served as the baseline, “non-stress”, sample because cortisol’s naturally lowest point in the diurnal cycle is in the late evening. Samples were stored in freezers and subsequently spun at 1600xg for 3 minutes to extract the saliva for analysis with liquid chromatography-mass spectrometry. Cortisol was quantified in ng/ml and converted to nmol/L. The lower limit of quantitation for the assay was 0.17 nmol/L.

1.2.6. Intervention

The Treatment for Affective Dimensions (TAD) was a 15-session randomized control trial (Craske et al., 2019). Participants were randomly assigned to either NAT or PAT. Both treatment arms broadly consisted of three modules that addressed patients’ a) behaviors, b) cognitions, and either c) physiological arousal (NAT), or c) compassion (PAT). In NAT, the goal of treatment was to reduce negative affect and threat sensitivity, whereas the goal of PAT was to increase positive affect and reward sensitivity The first half of treatment (sessions one to seven) focused on behavioral activation in PAT, and on exposure to feared situations and outcomes in NAT. Sessions eight to ten targeted cognitive retraining toward positive affect in PAT and restructuring negative cognitions in NAT. Sessions ten to fifteen included savoring positive experiences in PAT and breathing retraining in NAT (see Craske et al., 2019 for details).

1.2.8. Analytic Strategy

Changes in DASS over time were analyzed using multilevel modeling (MLM). MLM is the recommended approach for analyzing clinical trials (Hamer & Simpson, 2009), as it allows the inclusion of all individuals who provide at least one data point. It thereby allows unbiased estimates of the growth curve parameters when data are missing at random (MAR). We examined whether participants with missing data differed from those with complete data on any demographic or baseline level of the study variables. We then performed pattern mixture modeling to investigate whether growth curve parameters differed for those with missing data compared with those with complete data. Since the raw data of cortisol samples were skewed (skewness= 2.62 [SE= 0.14]), we transformed them using a natural logarithm before analysis. Analyses controlled for different variables that are known to affect cortisol, including age, gender, usage of any medications, and time of day. Since cortisol reactivity is thought to improve therapeutic outcomes through memory consolidation, we also separately examined the effects of psychotropic medication use on therapeutic outcomes (Buffett-Jerrorr & Stewart, 2002; Otto, McHugh, & Kantak, 2010). Variables that did not significantly predict outcomes in the analysis were dropped to conserve degrees of freedom and avoid model overfitting. We followed the Aiken and West (1991) “centering” method to estimate the growth curve at specific levels of predictor variables (e.g., cortisol reactivity). For the two main analyses of our aims, maximum likelihood estimation was used to calculate growth curve parameters, and the “residual” degrees of freedom method was used to calculate dfs for the p values.

Our first aim examined whether cortisol reactivity to a prolonged pre-treatment laboratory stressor, predicted better treatment benefits as measured by the DASS. Higher cortisol reactivity was hypothesized to predict greater improvement through the 6-month follow-up. Initial analyses showed that a logarithmic growth curve model with a Toeplitz error covariance matrix provided the best fit for the data, based on AIC and BIC. In addition to the previously listed variables that will be controlled (age, gender, psychotropic medication usage, and time of day), we also controlled for baseline levels of the DASS.

The growth curve model for testing Hypothesis 1 was:

$${\text{Y}}_{\text{ij}}{\text{=b0+b1* Cortisol }}_{\text{i}}{\text{+b2* Time }}_{\text{ij}}{\text{+b3* Cortisol }}_{\text{i}}^{\text{*}}{\text{ Time }}_{\text{ij}}{\text{+b4* Pre-Treatment Y}}_{\text{i}}{\text{+ε}}_{\text{ij}}$$

Y_ij is the outcome measure (DASS-Total) for participant i at assessment j. Cortisol_i is the cortisol reactivity for participant i. Time_ij is the log of time for participant i at assessment j and was scaled so that the regression coefficient (slope) for time reflected the total change from session 1 through follow-up. PreTreatmentY_i is the pre-treatment value of the outcome measure for participant i. Time was centered at the follow-up-assessment. Since pre-treatment levels of outcome were used as a predictor of the growth curve, the growth curve was modeled from session one through follow-up assessment (outcomes were assessed at the beginning of each session).

The second aim examined whether the effect of cortisol reactivity on outcome was moderated by treatment condition. We predicted that higher cortisol reactivity would predict better treatment outcomes for those in NAT but not in PAT through the 6-month follow-up. Thus, treatment condition was added as a moderator to the cortisol terms in the above MLM model. A post-hoc power analysis for the second aim showed that we had greater than .80 power to detect a small effect size (d=0.20).

A missing data analysis using Little’s MCAR Test found that the main study variables (e.g. DASS and Cortisol Reactivity) were not missing completely at random, χ²(69)= 1466.025, p< 0.001.² Pattern mixture modeling was used to assess whether the growth curve was significantly different between those with missing data and those without missing data. None of the growth curve parameters were different for those with, versus those without, missing data (all p’s >.059). These data are consistent with data missing at random, suggesting that the MLM estimates are unbiased estimates of the true parameters.

1.3. Results

1.3.1. Sample Characteristics, Cortisol Levels, and Clinical Change

97.0% (n=32) met the criteria for a current Diagnostic and Statistical Manual of Mental Disorders– 5 diagnosis (American Psychiatric Association, 2013) and 78.8% (n=26) had two or more diagnoses. Eighty-eight percent (n=29) of participants met diagnostic criteria for an anxiety disorder and more than half (51.52%, n=17) for a depressive disorder. Participants were predominantly female (68.6%, n=24) and white (51.4%, n=18). Also, 42.9% (n=15) reported taking psychotropic medication (anti-depressants and benzodiazepines). Of those, six of participants were taking benzodiazepines as needed during treatment, with two who reported taking it before the laboratory session. Pre-treatment DASS scores were in the moderate-to-severe range, M=30.93 (SD=13.30). There were no group differences in demographic variables between the treatment conditions. Moreover, there was no significant difference on the pretreatment DASS scores between NAT (M=30.04, SD=14.43) and PAT (M=32.89, SD= 11.65), t(32)=0.638, p=0.528, d=0.23. Means and SDs of the demographic and baseline characteristics data are presented in Table 1.

Table 1.

Baseline Characteristics of Patients (N=35)

Characteristic	Valuea
Age, meanb, c	35.79 (14.301)
Female sex	24 (68.6)
Race/Ethnicityc
White	18 (51.4)
Asian	4 (11.4)
African American	3 (8.6)
Hispanic/ Latino White	3 (8.6)
Hispanic/ Latino Non-white	5(14.3)
Other	1 (2.9)
Marital Statusc
Married	9 (25.7)
Single, in a relationship	6 (17.1)
Single, not in relationship	13 (37.1)
Other	5 (14.3)
Employed full or part timec	22 (62.86)
Education Levelc
Less than high school	1 (2.9)
High school diploma/GED	4 (11.4)
Some college/2/4 year degree	18 (51.4.5)
Graduate degree	11 (31.4)
Incomec
<10,000– 30,000	8 (23.53)
30,001 – 50,000	4 (11.76)
50,001 – 70,000	7 (17.65)
70,001 – 100,000	7 (20.59)
>100,000	6 (17.65)
Medications
Antidepressants	13 (37.1)
Benzodiazepines	6 (17.1)
Non-psychiatric	14(40.0)
Stimulants	2(5.7)
Unknown	1(28)
Conditionc
PAT	19 (54.3)
NAT	15 (42.9)
DASS Total Score, meanb	30.93(13.30)
Depression, meanb	11.34 (5.97)
Anxiety, me nb	7.88 (5.44)
Stress, meanb	11.70 (4.69)

There were no group differences in demographic variables between the two conditions (NAT vs PAT)

^aData are presented as number (percentage) of patients unless otherwise indicated

^bData are presented as (standard deviation)

^cIndicates missing responses (age=1, race=1, employed=4, income=3, marital status=2,education level=1, condition =1)

Cortisol levels averaged 2.05 nmol/L (SD=2.33) for the evening before the session and 4.28 nmol/L (SD=3.40) for the maximum during the laboratory stressor. A paired-samples t-test using the logtransformed cortisol values showed a significant difference between evening cortisol levels and peak cortisol levels observed during the laboratory stress tasks, t(34)=5.34, p<0.001, d=1.83. There was also a significant difference in a paired-samples t-test between the laboratory “baseline cortisol” level which was collected before the start of the two stress tasks and peak cortisol observed during the laboratory stress tasks t(33)= 2.08, p=0.045, d=0.72. Pre-treatment DASS score was not correlated with cortisol reactivity (using evening before as a reference), r(33)= 0.22, p=0.21.

1.3.2. Effects of Pre-treatment Cortisol Reactivity on Clinical Improvement

In line with our first hypothesis, higher pre-treatment cortisol reactivity was related to greater improvement (more negative slopes of change in DASS), b= −4.20, t(450)= −1.99, p=0.047, d= 0.19 for the cortisol reactivity x Time interaction. ³ To investigate this interaction, we used the “simple slopes” method to estimate predicted slopes of improvement in DASS for participants with various levels of cortisol reactivity. We found that participants with high cortisol reactivity (1SD above the mean) showed significant improvement in DASS over time, b= −4.46, t(450)=−4.93, p<0.001, d=0.46, while those with low cortisol reactivity (1SD below the mean) did not show significant decreases in DASS over time, b= −1.74, t(450)=1.80, p=0.07, d=0.17 (Figure 1). Those with average levels of cortisol reactivity showed significant improvement over time, b= −3.10, t(450)=−4.84, p<0.001, d=0.46, but the slope of improvement for these participants (−3.10) was less than the slope of improvement for those with high cortisol reactivity (−4.46). Although greater pre-treatment cortisol reactivity predicted more negative slopes of change over time, cortisol reactivity was not significantly related to final end-point DASS scores at the 6-month follow-up assessment, b=−6.99, t(450)=−1.11, p= 0.27, d=0.10 for the main effect of cortisol reactivity.

Figure 1.

Higher pre-treatment cortisol reactivity is related to more improvement in DASS scores over time.

We also examined the effects of an alternate characterization of cortisol reactivity. This alternate characterization was calculated as the cortisol level measured right before the stress task (during the laboratory session) minus the cortisol level measured the evening before (as opposed to the peak cortisol during the stress tasks). Using this alternate characterization of cortisol reactivity, we again found a significant cortisol reactivity x Time interaction, b= −2.82, t(433)=−3.17, p=0.002, d=0.30.⁴ This interaction had the same form as the interaction reported above for our primary characterization of cortisol reactivity. Participants with high cortisol reactivity (1SD above the mean using our alternate characterization) showed significant decreases in DASS scores over time, b= −5.15, t(433)=−5.76, p<0.001, d=0.55, while those with low cortisol reactivity (1SD below the mean) did not show significant decreases in DASS over time, b= −1.13 t(433)=−1.25, p=0.212, d=0.12. Again, participants with “average” levels of cortisol reactivity (on our alternate characterization of cortisol reactivity) showed significant improvement in DASS over time, b= 3.14, t(433)=−4.95, p<0.001, d=0.48, but this rate of improvement was slower than that for participants with high cortisol reactivity (using the alternate characterization of cortisol reactivity).

1.3.3. Moderating effects of pre-treatment cortisol reactivity on clinical improvement for NAT and PAT

Counter to our hypothesis, treatment condition did not differentially affect the relationship between cortisol reactivity and slopes of change, b=−6.23, t(445)=−1.37, p=0.17, d=0.13. That is, the effect of cortisol reactivity on slope of improvement in NAT was not different from its effect on slope of improvement in PAT.

1.4. Discussion

We investigated whether cortisol reactivity to a prolonged laboratory stress induction protocol was a predictor of treatment outcome in a transdiagnostic sample of anxious, depressed, and stressed individuals and whether the effects were moderated by treatment condition (with one treatment targeting threat sensitivity and the other targeting reward sensitivity). As hypothesized, participants with higher pre-treatment cortisol reactivity showed faster improvement in DASS scores than did those with lower pre-treatment cortisol reactivity. Indeed, those who were 1SD above the mean in cortisol reactivity improved 2.5 times faster relative to those 1SD below the mean. For average, the high cortisol group improved 1.4 times faster. However, in contrast to our second hypotheses, the relationship between cortisol reactivity and improvement during treatment was not greater in NAT than PAT.

The finding that higher cortisol reactivity predicted better treatment outcome provides further evidence for cortisol reactivity as a biological variable that impacts clinical improvement in psychosocial interventions. Prior research has shown a similar positive association between heightened cortisol reactivity to laboratory stressors and greater reductions in depressive symptoms (Dieleman et al., 2016) and agoraphobic avoidance (Wichmann et al., 2017b) at the end of treatment. The current study replicated and extended those observations to a transdiagnostic sample of individuals with clinical levels of anxiety, depression, or stress. Previous findings also suggested that higher cortisol reactivity predicted lower symptoms at the final point of assessment. By contrast, the current study only found that higher cortisol reactivity predicted quicker reduction in symptoms, but showed no differences at the final assessment (6-month follow-up). This suggests that future research on cortisol reactivity should examine both rates of improvement as well as final levels of the outcome, as we have done here.

In the current study, we also found that the alternate method in calculating cortisol reactivity, computing the difference of cortisol levels the evening before with the cortisol levels prior to the stress task, predicted better improvement during treatment for individuals with average to high levels. Levels before the stress tasks were already greater than the night-before levels, likely due to the influence of the tasks that preceded the stress task, a finding that lends credence to the idea that the general mobilization of the HPA-axis due to stressful situations has predictive value for treatment success.

Possibly, any strong cortisol reactivity could improve treatment outcomes through its influence on memory consolidation or on the process of transferring new information into long term memory (de Quervain et al., 2011; de Quervain et al., 2009). Although cortisol reactivity to stress in a laboratory setting may not reflect cortisol reactivity to more naturalistic exposure settings (Mayer et al., 2017), cortisol is likely elevated as part of the organism’s mobilization to strategize and address a challenge (Boucher and Plusquellec, 2019), which may have led to elevated cortisol levels beginning prior to in-vivo exposures (Alpers et al., 2003; Meuret et al., 2015; Schumacher et al., 2014). Existing literature indicates that higher cortisol levels during treatment exposures (relative to a control day) are associated with better treatment success. For instance, Meuret et al. (2015) observed that such higher cortisol levels during therapeutic exposures for panic disorder and agoraphobia were associated with lower avoidance behavior, lower threat appraisal, and higher perceived control throughout and at the end of treatment. Likewise, Siegmund et al. (2011) observed trend level reductions in panic symptoms and avoidance (as measured by the Panic and Agoraphobia Scale) for those with panic disorder and agoraphobia with higher absolute levels of cortisol during exposures. Additional evidence of the benefits of glucocorticoids has been observed in studies in which glucocorticoids have been administered exogenously and enhance extinction learning through memory consolidation (de Quervain et al., 2019). While the latter studies were limited to examining the association in individuals with a principal diagnosis of panic disorder and agoraphobia undergoing in-vivo exposure therapy, the current study was broader in scope. It suggests that cortisol reactivity to standardized laboratory stressors, assessed before treatment, can predict better treatment outcomes throughout treatment for a heterogeneous sample of patients with clinically relevant levels of anxiety, depression, and stress.

The current study also found that the effect of pre-treatment cortisol reactivity was observed across both treatment interventions and was not limited to an intervention aimed at reducing threat sensitivity, or even more narrowly, exposure therapy. Perhaps increased salivary cortisol reactivity assisted in strengthening memory consolidation by learning new and corrective information relevant to both threat and reward learning. Thus, benefits of cortisol could be due to increased learning through memory consolidation (Soravia et al., 2006, 2013; de Quervain et al., 2011) of the skills provided in both treatments, such as finding a silver lining in PAT and fear reduction through exposure in NAT.

An interesting juxtaposition arises when comparing the findings of cortisol reactivity across different research contexts: on the one hand, predictor studies examining pre-treatment cortisol reactivity to psychosocial laboratory stressors have shown that higher pre-treatment cortisol reactivity predicts superior treatment outcomes (de Quervain et al., 2009; 2019; Dieleman et al., 2016; Wichmann et al., 2017b), as was also the case in this study. On the other hand, some studies have conceptualized cortisol as indicative of an abnormality and in need of curtailment, such as laboratory studies where cortisol reactivity was directly manipulated using instruction sets (Abelson et al., 2014; Salzmann et al., 2018) or in psychotherapeutic interventions trials which have attempted to reduce cortisol reactivity (Gaab et al., 2003; Hammerfald et al., 2006; Lindsay et al., 2018; Storch et al., 2007; Tang et al., 2007). However, conceptualizing cortisol reactivity as detrimental may not be appropriate in all cases because it constitutes functional levels of mobilization that can also have beneficial consequences for therapy. As with many physiological parameters, responding within an intermediate-range can be expected to be adaptive in maintaining allostasis, whereas both over- and under-responding could be dysfunctional.

Several limitations deserve mention. Even though power was sufficient to examine our study aim, larger-scale dismantling studies could refine our understating of which particular components of the treatment in our multi-module protocol best capitalizes on the benefits of endocrine stress reactivity. Therefore, future studies could also attempt to dismantle the different therapeutic skills within the current study to clarify the full benefit of cortisol reactivity. Furthermore, studies on the impact of cortisol reactivity in pharmacological treatments for affective disorders are lacking. An additional limitation of our study is the lack of control for female reproductive influences and other factors like smoking, caffeine, and alcohol on the test day (Kudielka, Hellhammer, & Wüst; 2009). Finally, we encourage future studies to include additional sampling, including awakening cortisol, one-hour before and at the start of the laboratory protocol.

In conclusion, our findings indicate that average and higher cortisol reactivity to a laboratory stressor predicted greater treatment improvement in individuals with clinical levels of anxiety, depression, or stress. Future studies should investigate how to enhance cortisol reactivity’s therapeutic benefits through the use of naturally occurring high levels of cortisol (diurnal effects of cortisol awakening response, sleep, or napping) or exogenous cortisol manipulation in transdiagnostic samples (Kleim et al., 2014). Some preliminary research has found benefits of elevated cortisol during the morning due to the cortisol awakening response (Meuret et al., 2015). Furthermore, such an awakening response has been observed after a 90-minute nap (Devine and Wolf, 2016). The findings of those studies encourage further evaluation of the therapeutic benefits of endogenous cortisol to confirm the possible benefits of sleep in enhancing therapeutic benefits.

Highlights.

• Study evaluates role of cortisol reactivity in predicting psychotherapy outcome.

• The study recruited a transdiagnostic sample randomized into two therapies.

• Participants underwent intermittent stress induction tasks before treatment.

• Average and higher cortisol reactivity predicted faster reductions in symptoms across study.

• However, results were not moderated by treatment condition.