HPA Axis Functioning in Borderline Personality Disorder:
A Meta-Analysis
E. M. K. Drews (11013923) Research Internship Report Supervision: Prof. Dr. Arnoud Arntz
Amsterdam, 09.08.2017
Research Master Psychology Specialization: Clinical Psychology
Table of Contents
Abstract ... 1
Introduction ... 3
Stress Reactivity in Borderline Personality Disorder ... 3
The Hypothalamic-Pituitary-Adrenal Axis: A Major Stress Response System ... 3
Abnormal HPA Axis Functioning in BPD – A Matter of Comorbidity? ... 6
Current Evidence for Abnormal HPA Axis Functioning in BPD ... 8
Objectives ... 9
Methods ... 11
Protocol and Registration ... 11
Eligibility Criteria ... 11
Information Sources ... 12
Search ... 13
Study Selection ... 13
Data Collection Process ... 14
Data Items ... 15
Risk of Bias in Individual Studies ... 16
Summary Measures ... 16
Power Calculation ... 17
Synthesis of Results ... 17
Risk of Bias Across Studies ... 18
Additional Analyses ... 18
Study Selection ... 20
Identification ... 21
Eligibility ... 21
Included ... 21
Study Characteristics ... 22
Risk of Bias Within Studies ... 25
Results of Individual Studies ... 25
Synthesis of Results ... 26
BPD subjects versus HC subjects. ... 26
BPD subjects versus MDD subjects. ... 38
BPD subjects versus PD subjects. ... 44
Risk of Bias Across Studies ... 52
Comparison of basal cortisol in BPD subjects and HC subjects ... 52
Comparison of continuous cortisol in BPD subjects and HC subjects ... 52
Comparison of basal cortisol in BPD subjects and MDD subjects. ... 53
Comparison of basal cortisol in BPD subjects and PD subjects. ... 54
Discussion ... 55
Summary of Evidence ... 55
Strengths and Weaknesses of the Meta-Analysis ... 59
Weaknesses ... 59 Strengths ... 60 Implications ... 62 Theoretical implications ... 62 Practical implications. ... 62 Conclusion ... 64 References ... 65
Appendix B: Search strategy as used for PsycINFO ... 83
Appendix C: Items of the adjusted quality assessment ... 84
Appendix D: Quality ratings for the individual studies ... 86
List of Tables
Table 1: Associations between cortisol, childhood adversity, MDD, and PTSD ... 8
Table 2: Characteristics of all comparisons included in the current meta-analysis ... 23
Table 3: Subgroup Analyses: Basal Cortisol Values (BPD vs. HC) ... 36
Table 4: Subgroup Analyses: Basal Cortisol Vaues (BPD vs. MDD) ... 42
List of Figures
Figure 1. Schematic Illustration of the HPA Axis ... 4
Figure 2. PRISMA Flow Diagram ... 21
Figure 3. Forest Plot: Basal Cortisol Values (BPD vs. HC). ... 27
Figure 4. Forest Plot: Pharmacological Challenges (BPD vs. HC) ... 28
Figure 5. Forest Plot: Psychosocial Stress (BPD vs. HC) ... 29
Figure 6. Forest Plot: Recovery from Psychosocial Stress (BPD vs. HC). ... 29
Figure 7. Forest Plot: Continuous Cortisol Output (BPD vs. HC) ... 31
Figure 8. Forest Plot: Comparison of Psychosocial Stress Tests (Stress; BPD vs. HC) ... 35
Figure 9. Forest Plot: Comparison of Psychosocial Stress Test (Recovery; BPD vs. HC) .... 35
Figure 10. Forest Plot: Basal Cortisol Values (BPD vs. MDD) ... 39
Figure 11. Forest Plot: Pharmacological Challenges (BPD vs. MDD) ... 39
Figure 12. Forest Plot: Basal Cortisol Values (BPD vs. PD) ... 45
Figure 13. Forest Plot: Psychosocial Stress (BPD vs. PD) ... 46
Figure 14. Forest Plot: Recovery from Psychosocial Stress (BPD vs. PD) ... 46
Figure 15. Forest Plot: Comparison of PD Control Groups (Basal; BPD vs. PD). ... 49
Figure 16. Funnel Plot: Basal Cortisol Values (BPD vs. HC) ... 52
Figure 17. Funnel Plot: Continuous Cortisol Values (BPD vs. HC) ... 53
Figure 18. Funnel Plot: Basal Cortisol Values (BPD vs. MDD) ... 53
Abbreviations
ACTH Adrenocorticotropic Hormone
ANS Autonomic Nervous System
APA American Psychological Association
AUC Area under the curve
BPD Borderline Personality Disorder
CAR Cortisol Awakening Response
CC Clinical Control subjects
CI Confidence Interval
CPD Cluster C Personality Disorder
CRH Corticotropin-Releasing Hormone
DEX-CRH Dexamethasone-Suppressed Corticotropin-Releasing Hormone Test DSM-IV Diagnostic and Statistical Manual of Mental Disorders, 4th Edition
DSM-V Diagnostic and Statistical Manual of Mental Disorders, 5th Edition
DST Dexamethasone Suppression Test
ED Eating Disorder
HC Healthy Control subjects
HPA Axis Hypothalamic-Pituitary-Adrenal Axis
MDD Major Depressive Disorder
NSSI Non-Suicidal Self-Injury
PD Personality Disorder
PD-NOS Personality Disorders not otherwise specified PTSD Post-Traumatic Stress Disorder
SSRI Selective Serotonin Reuptake Inhibitor
Abstract
Borderline Personality Disorder (BPD) has been associated with abnormal hypothalamic-pituitary-adrenal (HPA) axis functioning inconsistently. Therefore, the current meta-analysis investigated HPA axis functioning based on basal, pharmacological, and psychosocial stress paradigms. Published and unpublished English, German, and Dutch literature was searched using six databases, neuroendocrinology-related journals, and prior reviews using keywords related to BPD and the HPA axis. We included case-control studies that compared adult BPD subjects with healthy controls (HC), subjects suffering from Major Depressive Disorder (MDD), and subjects suffering from other Personality Disorders (PD). Articles were independently screened and extracted by five reviewers using predefined data fields including study quality items. This resulted in 804 publications, of which 37 studies (44 comparisons, N = 1835) were included. Effect sizes were calculated based on Hedges’ g. Analyses were based on the random effects model. When compared to HC subjects, cortisol response in BPD was blunted during psychosocial stress (Hedges’ g = -0.29, 95% CI [-0.52, -0.05], p = .017) and recovery therefrom (Hedges’ g = -0.34, 95% CI [-0.54, -0.14], p = .001). Similarly, BPD subjects were characterized by lower cortisol during psychosocial stress (Hedges’ g = -0.80, 95% CI [-1.61, 0.01], p = .051) and recovery thereof (Hedges’ g = -0.74, 95% CI [-1.30, -0.18],
p = .001) in comparison to PD controls. Continuous cortisol output was elevated in BPD
subjects (Hedges’ g = 0.37, 95% CI [0.04, 0.71], p = .030), however, neither basal cortisol (Hedges’ g = 0.13, 95% CI [-0.05, 0.31], p = .161) nor cortisol after pharmacological challenges (Hedges’ g = -0.19, 95% CI [-1.12, 0.74], p = .690) differed between BPD and HC subjects. Also, BPD subjects did not differ from MDD controls on basal cortisol measures (Hedges’ g = -0.25, 95% CI [-0.65, 0.16], p = .231) and in response to pharmacological challenges (Hedges’
g = -0.23, 95% CI [-0.54, 0.08], p = .141). Heterogeneity was high for the basal cortisol
comparison between BPD and MDD subjects (c2= 7.3, p < .121, I2 = 45%) and for the
comparison between BPD and PD subjects (c2= 4.5, p < .210, I2 = 34%). Lastly, subgroup
analyses indicated that younger BPD subjects displayed elevated basal cortisol, while older age in BPD was associated with lowered basal cortisol values. This meta-analysis was limited by inconsistent reporting in individual studies and small samples for some comparisons. Due to the debilitating nature of stress-related symptoms in BPD, more research on elevated continuous cortisol output and blunted cortisol during psychosocial stress is necessary.
PROSPERO registration number: 42017062312
Keywords: Borderline Personality Disorder; Hypothalamic-Pituitary-Adrenal Axis;
Introduction Stress Reactivity in Borderline Personality Disorder
Borderline personality disorder (BPD) is a severe and heterogeneous disorder, which is characterized by affective instability, impulsivity, and interpersonal instability (e.g., Gunderson & Lyons-Ruth, 2008; Sanislow et al., 2002; Skodol et al., 2005). Heterogeneity emerges as five out of nine, seemingly dissimilar criteria need to be met for a diagnosis (American Psychological Association [APA], 2013). Yet homogeneity prevails when considering the role of stress in both the development and manifestation of the disorder (Kuhlmann, Schmidinger, Thomann, & Herpertz, 2013).
On the one hand, heightened stress exposure early in life has been considered a risk factor for the development of BPD (Ball & Links, 2009). On the other hand, increased vulnerability and maladaptive responses to stress are reflected by various BPD symptoms (Grove, Smith, Crowell, Williams, & Jordan, 2017). Accordingly, stress-related symptoms in BPD have been differentiated based on their chronic or acute nature (Zimmerman & Choi-Kain, 2009). Chronic symptoms are considered stable and temperamental features such as dysphoria, intolerance of aloneness, and abandonment concerns. In contrast, acute symptoms are elicited by stress and tend to remit quickly, such as impulsivity, mood reactivity, and self-injurious behaviour. Taken together, research indicates that both chronic and acute symptoms bear on stress responsivity, however, the pathophysiology of stress in BPD remains unclear (Wingenfeld et al., 2010).
The Hypothalamic-Pituitary-Adrenal Axis: A Major Stress Response System
The primary neurobiological system related to stress regulation is the hypothalamic- pituitary-adrenal (HPA) axis (Lightman, 2008; Mitrovic, 2002). Notably, the HPA axis regulates stress responses through feedforward and feedback mechanisms (Morris, Compas, & Garber, 2012). Specifically, the hypothalamus releases corticotropin-releasing hormone (CRH)
in response to a stressor, which is then carried to the anterior pituitary gland. This stimulates the adrenocorticotropic hormone (ACTH) production in the anterior pituitary gland, which prompts the adrenal cortex to produce the glucocorticoid hormone cortisol (Buitelaar, 2013; Zimmerman & Choi-Kain, 2009). Cortisol has a variety of metabolic functions, such as increasing blood sugar levels and suppressing the immune system (Randall, 2010). However, cortisol also restores homeostasis following exposure to stress through a negative feedback loop. Hence, and as illustrated in Figure 1, cortisol downregulates its secretion by inhibiting CRH and ACTH (Mitrovic, 2002; Zimmerman & Choi-Kain, 2009). In essence, both feed-forward and feedback loops modulate the HPA axis and consequently impact the stress response (Carrasco & Van der Kar, 2003).
Figure 1. Schematic illustration of the HPA axis. The three endocrine glands, i.e., the
hypothalamus, the pituitary gland, and the adrenal cortex, are regulated by neuroendocrine feedback interactions. Thereby, the HPA axis regulates responses to stress and the maintenance of homeostasis. Blue arrows represent the HPA axis feed-forward function; red arrows reflect the HPA axis feedback function.
Due to its central role in regulating stress responses, the HPA axis is one of the most thoroughly investigated physiological systems in clinical psychology and psychiatry (Nicolson, 2008). In fact, several observational and experimental approaches have been developed to study different components of the HPA axis. First, basal and continuous measurements serve as proxies for spontaneous hormone secretion during the day. Typically, cortisol secretion follows a diurnal rhythm (Fries, Dettenborn, & Kirschbaum, 2009), i.e. cortisol increases sharply upon awakening and declines over the course of the day (Girshkin, Matheson, Shepherd, & Green, 2014). Consequentially, altered basal cortisol values have been
Hypothalamus Anterior
Pituitary
Adrenal
Cortex Target tissue
ACTH CORT
commonly interpreted as a sign for dysregulation of the HPA axis resulting from chronic stress or illness (Lupien & Seguin, 2013).
Second, pharmacological manipulations, such as the dexamethasone suppression test (DST), are employed to determine the feedback functioning of the HPA axis (Carroll, 1985). As the DST inhibits ACTH secretion, decreased cortisol output usually follows. Therefore, the inability to suppress cortisol indicates a dysregulated feedback response (Phillips et al., 2006). According to this, previous studies indicated that different doses of dexamethasone are associated with disease-specific responses (Tajima-Pozo et al., 2013). Hence, prior research suggested that the 1-mg DST allows for differentiation of endogenous depression (Carroll, Martin, & Davies, 1968; Young, Watson, & Akil, 1986) and that the 0.5-mg DST enables the differential diagnosis of major depressive disorder (MDD) and post-traumatic stress disorder (PTSD; Yehuda, Halligan, Golier, Grossman, & Bierer, 2004). Carrasco and colleagues (2007) further claimed that cortisol responses to the 0.25-mg DST distinguish BPD and PTSD individuals, while Rinne and colleagues (2002) noticed that the dexamethasone-suppressed corticotropin-releasing hormone test (DEX-CRH) mirrors trauma history (Rinne et al., 2002). Third, psychosocial stress tests enable the examination of reactivity to experimental and acute real-life stressors. For instance, reactivity to interpersonal stressors is commonly investigated with the Trier Social Stress Test (TSST; Kirschbaum, Pirke, & Hellhammer, 1993), which consists of delivering a speech and performing mental arithmetic in front of an audience. Typically, cortisol is measured before stressor onset, up to 25 minutes after stressor onset, and more than 25 minutes after stressor onset, to estimate baseline cortisol values, the stress response, and recovery from stress.
Moreover, HPA axis activity can be measured based on several methods and assays. In brief, blood1 and saliva samples are taken when cortisol levels at specific moments in time are
of interest. In contrast, daily cortisol output is typically examined with urinary cortisol or averaged salivary or blood cortisol samples over a period of 12-24 hours (Lupien & Seguin, 2013). Finally, hair cortisol can be used to scrutinize long-term cortisol values. In sum, the measurement of cortisol varies depending on the method used and period assessed (Morris et al., 2012).
Abnormal HPA Axis Functioning in BPD – A Matter of Comorbidity?
As mentioned earlier, HPA axis functioning has been deemed relevant for a broad range of psychopathologies, such as MDD, PTSD, and childhood adversity (Charmandari, Tsigos, & Chrousos, 2005; Heim et al., 2008)2. Since aberrant HPA axis functioning seems to reflect
abnormal stress responsivity in these psychopathologies and further relates directly to chronic and acute BPD symptoms (Cackowski et al., 2014), striking parallels between pathological stress responsivities in BPD, MDD, PTSD, and childhood adversity will be outlined in the following. An illustration of prevailing cortisol patterns in MDD, PTSD, and childhood adversity can be found in Table 1 (adjusted from Baumeister et al., 2014).
First, childhood adversity induces fundamental changes in stress regulation capacities (Wingenfeld, Spitzer, Rullkötter, & Löwe, 2010) and research indicates that HPA axis functioning mediates this relationship (Egliston, McMahon, & Austin, 2007). In particular, childhood adversity has been associated with HPA axis hyperactivity (Heim & Nemeroff, 2001; McGowan et al., 2009), elevated cortisol responses to pharmacological challenge tests (Tyrka et al., 2009), and diminished responses to psychosocial stressors (Carpenter et al., 2007). As many BPD patients experience trauma - such as emotional, physical, and sexual abuse - early in life (Holm & Severinsson, 2008), a high prevalence of childhood adversity in BPD patients has been observed (40% to 80%; Zanarini, Frankenburg, Hennen, Reich, & Silk,
2 In agreement with the neuroendocrine literature (e.g., Baumeister et al., 2014), childhood adversity refers to
trauma exposure in childhood, whereas PTSD additionally describes exposure to trauma in adulthood. Generally, it should be noted that childhood adversity increases the risk for PTSD, but does not necessarily lead to symptoms of trauma in adulthood (Xie et al., 2010).
2006). Since prior research indicates a close link between childhood adversity and BPD development (Goodwin, 2005), we may expect that BPD patients display a hyperreactive HPA axis, increased cortisol suppression, and lowered cortisol following psychosocial stressors when compared to unaffected counterparts.
Second, BPD is highly comorbid with a range of stress-related disorders, most notably MDD (83%) and PTSD (56%; Zanarini et al., 1998). Both disorders are known for abnormal HPA axis functioning (Leonard & Myint, 2009; Yehuda, 2006). However, endocrine patterns for these disorders seem reversed (Baumeister, Lightman, & Pariante, 2014; Musselman et al., 1998). Consequently, MDD has been characterized by a hyperactive HPA axis, non-suppression of cortisol following pharmacological challenges, and increased cortisol responses to psychosocial stressors. In contrast, PTSD has been associated with a hypoactive HPA axis, enhanced cortisol suppression following pharmacological challenges, as well as elevated cortisol after exposure to psychological stressors (Kendall-Tackett, 2000; Kellner & Yehuda, 1999).
Importantly, earlier findings related to PTSD and MDD appear controversial. In particular, one meta-analysis could neither establish differences in basal cortisol nor in cortisol suppression following pharmacological challenges between PTSD patients and healthy controls (Klaassens, Giltay, Cuijpers, van Veen, & Zitman, 2012), while another meta-analysis demonstrated both lower basal cortisol values and increased cortisol suppression following pharmacological challenge tests in PTSD patients (Morris et al., 2012). Interestingly, the authors reported similar differences for individuals suffering from both PTSD and MDD, while a third meta-analysis found no basal cortisol differences between MDD patients and healthy controls (Burke, Davis, Otte, & Mohr, 2005). Another meta-analysis on pediatric depression demonstrated higher basal cortisol values and non-suppression following pharmacological challenges (Lopez-Duran, Kovacs, & George, 2009). Although the last meta-analysis is limited
by its ambiguous comparison group, these findings seem to indicate that comorbid disorders moderate HPA axis activity in BPD patients (Wingenfeld et al., 2009). However, evidence for this conjecture is sparse and has not been evaluated quantitatively yet.
Table 1
Associations between cortisol, childhood adversity, MDD and PTSD (adjusted from Baumeister et al.,2014)
Cortisol measure Childhood adversity MDD PTSD
Awakening ? ↑ ↓
Afternoon - ↑ ?
Daily output ↑ ↑ (↓)
Post-DST ↑ ↑ ↓
Post-TSST ↓ ↑ ↑
Note. Stress responsivity in MDD, PTSD, and childhood adversity is reflected by diverging,
but abnormal HPA axis responses. Evidence on associations between childhood adversity and cortisol awakening response is inconsistent and sparse for afternoon cortisol levels. In PTSD, inconsistent findings for afternoon cortisol output have been reported. Daily cortisol production tends to be lower in PTSD as compared to healthy individuals.
Current Evidence for Abnormal HPA Axis Functioning in BPD
Looking at BPD in particular, prior research demonstrated “increased stress vulnerability, disturbed HPA axis functioning and alterations in the size and activation of structures involved in central stress regulation” (Kuhlmann et al., 2012, p. 130). Put differently, HPA axis functioning seems to reflect hippocampal and amygdala abnormalities as the limbic system strongly influences the endocrine system (Packan & Sapolsky, 1990). In particular, the hippocampus inhibits HPA axis activity (Sapolsky, Krey, & McEwen, 1984), while the amygdala increases HPA axis activity (Herman et al., 2003). Consequently, one might expect that reduced hippocampal and amygdalar volumes and increased amygdalar activation in BPD point to a hyperactive HPA axis (Herman, Ostrander, Müller, & Figueiredo, 2005; Wingenfeld
et al., 2009). And indeed, initial studies indicated enhanced basal and stimulated cortisol release in BPD (Wingenfeld et al., 2009). However, results from individual studies are ambiguous and often contradictory (Scott, Levy, & Granger, 2013). And although earlier reviews summarized studies on HPA axis activity in BPD qualitatively (Wingenfeld et al., 2009; Zimmerman & Choi-Kain, 2009), a quantitative analysis of the existing studies is currently lacking. Since understanding the stress response system has critical implications for chronic and acute symptoms among those with BPD (Scott et al., 2013), such an analysis appears highly relevant to overcome limitations inherent to individual studies and to understand stress regulation capacities in BPD thoroughly.
Objectives
Scientific evidence suggests abnormal basal and stimulated cortisol concentrations in BPD patients (Wingenfeld et al., 2009; Zimmerman & Choi-Kain, 2009). Notwithstanding that HPA axis abnormalities are indicated by a variety of acute and chronic BPD symptoms, a systematic evaluation of HPA axis activity in BPD patients is currently lacking. Therefore, the current meta-analysis examined case-control studies comparing cortisol activity among BPD subjects, healthy control subjects (HC), and clinical control subjects (CC).
Separate analyses were carried out to scrutinize HPA axis functioning in BPD based on three prevailing paradigms, i.e. basal assessments, pharmacological challenges, and psychosocial stress tests. Exploratively, we also examined continuous cortisol output. Based on previous reviews (Wingenfeld et al., 2009; Zimmerman & Choi-Kain, 2009), we expected BPD subjects to display enhanced basal cortisol, non-suppression of cortisol following pharmacological challenges, and increased cortisol responses during stress and recovery phases of psychosocial stress tests. Further, we anticipated that BPD patients suffering from comorbid MDD or PTSD would display comorbidity-specific cortisol responses, hence hypercortisolism
and non-suppression of cortisol in the case of MDD, and hypocortisolism as well as enhanced cortisol suppression in the case of PTSD.
As consideration of factors that influence HPA axis activity facilitates the interpretation of results (Nicolson, 2008), we further examined a range of potential moderators, i.e. gender, age, medication use (antidepressant use and contraceptive use), smoking status, type of challenge, assessment method (material used and timing of the assessment) as well as study quality, in an explorative manner. Those factors were chosen as prior reviews (Wingenfeld et al., 2009; Zimmerman & Choi-Kain, 2009) suggested that these factors might either impact HPA axis functioning in BPD patients in particular or the comparison between BPD and control subjects more generally. First, we included age as a potential moderator since cortisol typically increases with age but as age might also impact acute stress reactivity (e.g., Otte et al., 2005). Second, gender was included as males and females often display differential responses to experimental stressors, for instance, due to female reproductive hormones and contraceptive medication (Nicolson, 2008). Third, we compared different constellations of medication use as it is known that certain classes of drugs have long-term effects on the HPA axis. Fourth, we compared differences based on smoking status as prior research demonstrated that smoking influences cortisol responses to acute stressors and possibly also basal cortisol levels (Nicolson, 2008). Fifth, we scrutinized differential effects of pharmacological and psychosocial challenges as past research indicated that they impact clinical and healthy populations differentially (e.g., Dickerson & Kemeny, 2004; Ribeiro, Tandon, Grunhaus, & Greden, 1993). Sixth, we evaluated various assessment methods to compare endocrine differences in diurnal rhythms between BPD subjects and control subjects and to compare endocrine differences due to assessment of different body fluids. Lastly, we scrutinized studies based on their quality scores to investigate if systematic differences can be ascribed to the methodological rigor of the individual studies.
Methods Protocol and Registration
Review title and timescale, team details, methods, and general information were documented in a web-based protocol on the ‘International Prospective Register of Systematic Reviews’ by the Centre for Reviews and Dissemination (PROSPERO; registration number 42017062312). The guidelines of the PRISMA Statement (Liberati et al., 2009; Moher et al., 2015) were used as a framework. The protocol was published on April 16, 2017, and review stages were updated promptly. Adaptations were made as related to the title, keywords, hand-search procedures, and planned analyses. These adjustments were deemed necessary as the initial selection of journals did not yield satisfactory findings and as the analyses had to be adjusted to the characteristics of the included studies.
Eligibility Criteria
Studies were included if (1) formally diagnosed3 BPD subjects were compared to HC or CC subjects above 18 years of age, (2) HPA axis-related dependent variables such as CRH, ACTH, or cortisol were studied, (3) baseline indices of HPA axis activity, HPA axis responses to pharmacological challenges (i.e. DST and DEX-CRH), or psychosocial stress tests (e.g., TSST) were measured, and if (3) sufficient statistical information, such as group means and standard deviations, was disclosed. We excluded studies that (4) were only published in abstract form, (5) did not contain primary data, e.g., reviews or editorials, (6) only reported on cortisol assessments in personality disorders in general rather than in BPD specifically, (7) included individuals with endocrine disorders4, and when (8) data for BPD participants were not reported separately from other participants. Published and unpublished reports were
3 We accepted diagnostic instruments as for instance the Structured Clinical Interview for DSM-IV Axis II
Disorders (SCID-II; First, Gibbon, Spitzer, Benjamin, & Williams, 1997) and the Revised Diagnostic Interview for Borderlines (DIB-R; Zanarini, Gunderson, Frankenburg, & Chauncey, 1989).
4 In particular, we excluded studies focusing on individuals suffering from Cushing’s disease, Addison’s disease,
considered eligible if written in English, Dutch, or German and made available between 19805 to April 2017.
Information Sources
The search was developed and conducted by the first author of this report. Published studies were identified by searching the electronic databases PsycINFO, MEDLINE, Embase, Scopus, and Web of Science. All databases were searched for published articles between 1980 and 2017. The following additional restrictions were imposed: PsycINFO was searched for studies reporting on adult human participants; MEDLINE was searched for English, Dutch, and German articles reporting on human participants; Embase was searched for studies using human adults; Scopus was restricted to English articles as two foreign-language articles in Spanish and Italian were retrieved previously; and Web of Science was searched for articles written in English, German, and other languages not specified. Unpublished studies written in English, Dutch, and German were identified through the ProQuest Dissertations & Theses database (1980-2017). To avoid repeated extractions, we applied deduplicate tools before collection of the articles. The following journals were hand-searched for matching publications: ‘Psychoneuroendocrinology’ (1980-2017), ‘Journal of Personality Disorders’ (1987-2017), ‘Journal of Neuroendocrinology’ (1989-2017), ‘Hormones and Behavior’ (1980-2017). Further, reference lists of prior reviews on HPA axis functioning in BPD (Wingenfeld et al., 2009; Zimmerman & Choi-Kain, 2009) were checked. Also, the Clinical Trials database (http://clinicaltrials.gov/) was searched for ongoing trials, and experts in the field6 were contacted to determine if they had unpublished data available to share. A table listing authors who provided additional details, authors who confirmed to provide details but did not do so
5 BPD was first introduced as a disorder in the DSM-III in 1980 (APA, 1980).
6 We contacted Prof. Anthony Ruocco, Prof. Eric Fertuck, Prof. Katja Wingenfeld, and Prof. Lois W. Choi-Kain
yet, authors who could not provide details, and authors who could not be contacted can be found in Appendix A.
Search
Prior to formulating the protocol, a pilot search was conducted to ensure that no systematic review or meta-analysis pertaining to the research questions had been previously published or registered. The Cochrane Database for Systematic Reviews (CDSR; http://www.cochranelibrary.com/cochrane-database-of-systematic-reviews/), PROSPERO, PubMed, and PsycINFO were used as resources. Next, we formulated a search string for searching PsycINFO. The first component consisted of borderline personality disorder, emotionally unstable personality disorder, or borderline patient*. The second component included HPA axis, cortisol, hormone*, and synonyms. The full search strategy including hits per search term for PsycINFO can be found in Appendix B. Similar search terms were used for the other databases.
Study Selection
The eligibility assessment was performed by five reviewers (AEA, ED, GT, LT, PD) in a blinded and standardized manner so that two reviewers rated each article. Immediate agreement was reached in 91% of the cases (k = .67). Remaining disagreements were resolved by consensus. When no agreement could be reached, a sixth reviewer decided (EF). Prior to full-text evaluation of the articles, titles and abstracts were screened based on the in- and exclusion criteria mentioned above. Full texts were considered if all criteria were met or likely met. Articles were included in the quantitative synthesis if all inclusion criteria were reported in the article or if corresponding authors provided crucial details so that all criteria could be considered satisfied.
Data Collection Process
A digital data extraction sheet was developed (ED, AA) based on the recommendations by Tacconelli (2010). We pilot-tested the extraction sheet on seven randomly-selected included studies (AEA, GT, LT, PD), and refined accordingly (ED). Five reviewers (AEA, ED, GT, LT, PD) independently extracted data from all included studies so that extractions were completed in duplicate. Disagreements were resolved by discussion between reviewers. The extraction sheet can be found at https://drive.google.com/open?id=0B1qLzhvKAbnzdnhJZm9JZD NRY1U.
Further, a web-based digitizer (http://arohatgi.info/WebPlotDigitizer/app/) was used to extract means and standard deviations for studies that only reported relevant statistics based on graphs. When labels were missing from graphs, we assumed that means and standard errors were reported. Standard errors were converted based on the following formula S. E. M. to S. D. = S. E. M. x ( ,) as suggested by Zakzanis (2001). Similar to Dickerson and Kemeny (2004), a conservative effect size of Hedges’ g = 0.000 was chosen if ‘no significant changes’ were reported without additional information.
Duplicate reports were removed when identical samples were described (De la Fuente & Mendlewicz, 1996; De la Fuente, Bobes, Vizuete, & Mendlewicz, 2002). When multiple samples were compared, the largest respective samples were chosen7. If multiple measurements were taken during the day, the measurements closest to 0800h and 1600h were taken as morning and afternoon measurements. In line with previous research (Dickerson & Kemeny, 2004), we defined psychological stressors as non-metabolically demanding tasks, which excluded physical stressors, physical-psychological stressor combinations, and studies involving biological challenges. Accordingly, we only included acute psychological laboratory
7 E.g., for the study by Scott and colleagues (2013), the comparison between BPD participants and the
stressors8 such as cognitive tasks, public speaking tasks, and emotion induction procedures. For psychosocial stress tests, the measurement before stressor onset was recorded as baseline measure, the measurement up to 25 minutes after stressor onset was taken as stress measure, and the measurement more than 25 minutes after stressor onset was recorded as recovery measure, as recommended by Burke and colleagues (2005).
Data Items
Information was extracted from each included trial on (1) general information and
identifying features of the study, i.e. the full reference, year and type of publication, publication
status, country of origin, and source of funding, (2) study characteristics, i.e. aim of the study, study design, recruitment procedures and criteria for in- and exclusion, matching of participants, and any requirements prior to the HPA axis assessment, (3) participant
characteristics, i.e. sample sizes, patient status, diagnostic procedures, gender ratio, average
age, symptom severity, comorbid diagnoses, average BMI, smoking status, contraceptive medication and antidepressant use, other psychoactive medication use, assessment of childhood adversity, non-suicidal self-injury (NSSI) and suicidality, (4) study design, i.e. time of sampling and HPA axis measurement procedure, bodily material used, amount of HPA axis measurement points, and details on basal, pharmacological, and psychosocial stress testing, (5)
results, i.e. exclusion of participants, HPA axis-related findings and corresponding statistics,
as well as (6) the quality assessment (see ‘Risk of Bias in Individual Studies’ on page 16 for a detailed description). Due to the high amount of missing data, several interval variables were retrospectively recoded to categorical variables (see ‘Additional Analyses’ on page 18 for a detailed description).
8 Acute laboratory stressor tasks were defined as “tasks that lasted 60 minutes or less and did not serve a function
outside the laboratory setting; this excluded extended stressor challenges, chronic stressor studies (e.g., caregiving), and naturalistic stressors” (e.g., class examinations; Dickerson & Kemeny, 2004, p. 360). As described on page 33 of this report, we included studies using the TSST, emotion induction tasks, conflict discussions, and the Cyberball paradigm in the comparison on psychosocial stress tests.
Risk of Bias in Individual Studies
To gauge the risk of bias in the included studies, five reviewers (AEA, ED, GT, LT, PD) independently assessed the quality of the studies based on an adjusted version of the quality tool developed by Tak and colleagues (2011). The original quality assessment included nine items related to three key domains in clinical neuroendocrine research, i.e. selection of participants, measurement of HPA axis, and assessment of confounders (Tak et al., 2009). We adjusted two items and reworded one item to evaluate the quality of the present comparisons between BPD and control participants9. Moreover, we added two items to control for quality as related to experimental designs. In particular, one item was used to examine manipulation checks. The other item was used to investigate if subjective stress levels were additionally studied. The respective items of the adjusted quality assessment can be found in Appendix C. Further, we calculated separate scores for observational (relative scores) and experimental designs (absolute scores) so that a maximum score of 18 and 22 points was awarded, respectively. Variability in study results due to methodological differences was explored based on the median-split of the relative quality ratings for all included studies.
Summary Measures
Because of the small samples sizes of many included studies, Hedges’ g (Hedges, 1982) was chosen as a measure of cortisol differences between BPD and control subjects as recommended by Klaassens and colleagues (2012). Hedges’ g, which is an unbiased estimate of the standardized mean difference, was calculated for individual and combined effect sizes. All analyses were based on the random effects model as we anticipated differences between the comparisons relating to participant selection and designs used in the individual studies.
9 More precisely, the first item was adjusted to the current target population of BPD subjects. The second item
was modified so that the recruitment of control participants could be examined. The fourth item was changed and assessed the absence of endocrine disorders instead of disease characteristics. These adjustments were deemed necessary as the initial quality tool has been developed to examine HPA axis functioning in somatic disorders (Tak et al., 2011), which are investigated differently in neuroendocrine research than mental disorders such as BPD.
Power Calculation
To investigate if there was sufficient statistical power in the current meta-analysis to detect a small to medium effect size, a power calculation according to the procedures described by Borenstein, Hedges, Higgins, & Rothstein (2009) was carried out. Assuming a medium level of between study variance, statistical power (1 - b) set at .80, and α = .05, we calculated that 22 studies with a mean sample size of 30 subjects per condition were required to detect a small effect size of Hedges’ g = .20. Alternatively, six studies with 20 subjects were needed to detect a medium effect size of g = .50. As we included 31 studies for the comparison of basal cortisol values (BPD group: M = 23, range: 4 – 92; HC group: M = 20, range: 6 – 229), this meta-analysis was sufficiently powered. In contrast, most other comparisons were likely insufficiently powered to detect small to medium effects. For instance, the comparison of basal cortisol values between BPD (M = 17, range: 16 – 24) and MDD subjects (M = 18, range: 10 – 33) was based on five studies and therefore insufficient to detect a small or medium effect size.
Synthesis of Results
Individual and combined effect sizes were calculated using Comprehensive Meta-Analysis (Version 3, Biostat, Englewood, NJ, USA). Three meta-analyses were conducted to compare BPD subjects to HC subjects, BPD subjects to MDD subjects, and BPD subjects to PD subjects, respectively. Heterogeneity between studies was measured with the chi-square Q-statistic (Cochran, 1954), which tests the null hypothesis that all the variation in effects is due to sampling error. Heterogeneity was further examined with the I2 index, which indicates the proportion of true variance to observed variance. Generally, I2 values of 25%, 50%, and 75% are interpreted as representing small, moderate, and high levels of heterogeneity (Higgins, Thompson, Deeks, & Altman, 2003).
Risk of Bias Across Studies
Publication bias was examined by inspecting the funnel plots for the basal cortisol comparisons between BPD and HC as well as CC subjects and for the comparison of continuous cortisol output between BPD and HC subjects. These comparisons were chosen as they covered all studies included in the current meta-analytic investigation. Further, publication bias was formally tested based on Egger’s test, where the standard normal deviate is regressed on precision, which is defined as the inverse of the standard error (Borenstein, 2005). Biased effect size estimates were further investigated with Duval and Tweedie’s trim and fill procedure (Duval & Tweedie, 2000), which calculates the effect of potential data censoring including publication bias on the outcome of the meta-analysis (O’Mara, 2008). All calculations were based on the random effects model.
Additional Analyses
Subgroup analyses were carried out to examine if the combined effect sizes vary in relation to the a priori determined set of variables, i.e. comorbidity status (MDD and PTSD), gender, age, medication use (antidepressant use and contraceptive use), smoking status, type of challenge, assessment method (material used and timing of assessment), and study quality. In general, subgroup analyses enable the comparison of the mean effect for different subgroups of studies, as analogous to analysis of variance in individual studies. Put differently, one can examine if studies have a different effect based on subgroup analyses, where the total set of studies is divided into a number of subgroups so that separate effect sizes and heterogeneity estimates for these subgroups can be calculated (Cuijpers, 2016). Due to the high heterogeneity of the principal meta-analysis comparing basal cortisol between BPD and HC subjects, a mixed effects model was chosen for all subgroup analyses. This model makes use of the random effects model to combine studies within each subgroup and the fixed effects model to yield an overall effect. Thereby, the mixed effects model does not assume equal between-studies
variance for all subgroups, resulting in separate variance computations for the different subgroups.
Due to the high amount of missing data, MDD and PTSD, medication use, and smoking status were coded as present, absent, or unclear. For example, studies coded as ‘MDD present’ included a certain amount of BPD subjects suffering from comorbid MDD, whereas studies coded as ‘MDD absent’ did not include any BPD subjects suffering from comorbid MDD. Gender was coded as mostly female (≥ 75% female subjects), mixed (25%-75% female subjects), or mostly male (≤ 25% female subjects). Age was coded as old (≥ 29 years) or young (< 29 years) based on a median-split calculated from the studies that examined basal cortisol values. Material was coded as blood or saliva10. Timing of assessment was coded as morning (AM) or afternoon (PM). Study quality was coded as high (> 11) or suboptimal (≤ 11) based on a median-split calculated from the studies that examined basal cortisol values. Type of challenge was coded individually. Hence, pharmacological challenges were coded as 1-mg DST, 0.5-mg DST, 0.25-mg DST, or DEX/CRH. Psychosocial stress tests were coded as Trier Social Stress Test, Conflict Discussion, Emotion Induction, or Cyberball.
10 No distinction between serum and plasma could be made as some authors did not reported which material they
Results
Study Selection
Searching PsycINFO, MEDLINE, Embase, Scopus, Web of Science and the ProQuest Dissertations & Theses database yielded 804 citations. After removing duplicates, 502 publications remained. Titles and abstracts of those publications were screened. When screening the respective abstracts, 135 publications did not report on group comparisons between BPD and healthy or clinical subjects above 18 years of age based on case-control designs, 48 publications did not contain HPA-axis related dependent variables such as ACTH, CRH, or cortisol, 38 publications were only published in abstract form, 172 publications did not contain primary data, 22 publications did not report outcomes separately for BPD individuals and other subjects, two publications were neither published in English, German, or Dutch, and one publication reported on HPA axis assessments in various personality disorders rather than in BPD specifically.
Full texts of the remaining 84 publications were examined in greater detail. Of those, 15 publications did not report on group comparisons between BPD and HC or CC subjects above 18 years of age based on case-control designs, 15 publications reported insufficient statistical information, four publications reported on HPA axis assessments in various personality disorders rather than in BPD specifically, two studies were only published as abstracts, one publication did not contain HPA-axis related dependent variables such as ACTH, CRH, or cortisol, and one publication reported on individuals suffering from neuroendocrine disorders. The remaining 37 publications were included in the quantitative synthesis. As seven publications reported on multiple comparisons11
, three separate meta-analyses were based on 44 comparisons. Qualifying studies are marked with an asterisk in the reference section.
11 The following articles included multiple comparisons: Aleknaviciute et al. (2016), Deckers et al. (2015),
Figure 2. PRISMA flow diagram illustrating the number of publications identified and the
number of publications that were in- and excluded during different stages of the review process.
Records identified through database searching (n = 747) Embase (n = 135) MEDLINE (n = 109) ProQuest (n = 57) PsycINFO (n = 81) Scopus (n = 21) Web of Science (n = 344) Sc re eni ng In cl u d ed E li gi b il it y Id en ti fi cat ion
Additional records identified through other sources (n = 57)
Clinical Trials Database (n = 7)
‘Hormones and Behavior’ (n = 3)
‘Journal of Personality Disorders’ (n = 17)
‘Psychoneuroendocrinology’ (n = 13)
Wingenfeld et al. (2009) (n = 4)
Zimmerman & Choi-Kain (2009) (n = 13)
Records after duplicates removed (n = 302) Automatic removal (n = 185) Manual removal (n = 117) Records excluded (n = 418) (1) Inappropriate comparison (n = 135)
(2) No HPA axis outcomes (n = 48)
(4) Abstracts (n = 38) (5) No primary data
(n = 172)
(6) Personality disorders in general (n = 1)
(8) Separate outcomes not reported (n = 22)
(9) Other languages (n = 2)
Full-text articles assessed for eligibility
(n = 84)
Full-text articles excluded, with reasons (n = 42)
(1) Inappropriate comparison (n = 15)
(2) No HPA axis outcomes (n = 1)
(3) Insufficient statistics (n = 15)
(4) Abstracts (n = 2)
(6) Personality disorders in general (n = 4)
(7) Neuroendocrine disorders (n = 1)
(8) Separate outcomes not reported (n = 4)
Studies included in quantitative synthesis (meta-analysis)
(n = 37)
Comparisons included in quantitative synthesis (meta-analysis)
(n = 44) Records screened
Study Characteristics
All 37 studies selected for the meta-analytic comparison were case-control studies published in English. As shown in Table 2, 19 publications exclusively reported on basal cortisol comparisons, seven publications reported on cortisol comparisons following pharmacological challenges, and 11 publications reported on cortisol comparisons during and after psychosocial stress tests12
. In total, 802 BPD subjects were compared to 1094 control subjects. In particular, 33 studies compared BPD subjects (n = 740) to HC subjects (n = 897), seven studies compared BPD subjects (n = 114) to MDD subjects (n = 116), and four studies compared BPD subjects (n = 74) to PD subjects (n = 81). Overall, 159 effect sizes were computed with each study contributing an average of four effect sizes. Most participants were female (83%) and the average age was 30 years (range: 19 – 69). The average body mass index (BMI) of all participating individuals was 24 (range: 21 – 27). However, 19 studies did not report the average BMI of participating individuals. Regarding medication use, 18 studies involved washout periods, 10 studies included participants taking both endocrine and non-endocrine medications, one study only included non-non-endocrine medications, and two studies did not include participants taking medications. The impact of endocrine medications was not specified in five studies and one study did not report on medication use. Further, most cortisol measures were based on saliva (k = 14), plasma (k = 13), and serum (k = 8). For one study, it was not mentioned whether serum or plasma cortisol was used and one study examined urine cortisol. Lastly, most studies were carried out in the morning (k = 21), while the remaining studies were either carried out in the afternoon (k = 15) or during night-time (k = 1).
12It should be noted that the studies reporting on pharmacological challenges and psychosocial stress tests usually
included basal assessments, leading to 31 comparisons of basal cortisol between BPD and HC subjects, five comparisons of basal cortisol between BPD and HC subjects, and four comparisons of basal cortisol between BPD and PD subjects.
Table 2
Characteristics of all comparisons, grouped by publication, as included in the current meta-analysis
ID Study BPD Sample Control Sample Assessment
Cf. n Age Sex BMI Med n Age Sex BMI Type Time Paradigm Quality 1.1 Aleknaviciute et al., 2016 HC 26 29.2 100 24.5 1 35 28.6 100 22.4 Saliva PM PSY 15 1.2 Aleknaviciute et al., 2016 PD 26 29.2 100 24.5 1 20 26.1 100 23.5 Saliva PM PSY 15
2.1 Beeber et al., 1984 MDD 13 NR NR NR 4 10 NR NR NR Serum PM PHA 6
3.1 Bromundt et al., 2013 HC 14 30.1 100 26.7 3 10 25.7 100 21.4 Saliva AM BASAL 7
4.1 Carrasco et al., 2003 PD 16 NR NR NR 1 14 NR NR NR Plasma AM BASAL 5
5.1 Carrasco et al., 2007 HC 32 30.6 59 NR 1 18 29.7 61 NR Plasma AM PHA 12
6.1 De la Fuente et al., 2002 MDD 20 32.4 70 NR 1 20 35.8 75 NR Plasma PM PHA 10
7.1 Deckers et al., 2015 HC 22 31.4 100 NR 3 24 28.6 100 NR Saliva PM PSY 13
7.2 Deckers et al., 2015 PD 22 31.4 100 NR 3 23 31.9 100 NR Saliva PM PSY 13
8.1 Feliu-Soler et al., 2013 HC 35 30.2 91 24.5 3 15 30.6 87 22.9 Saliva PM PSY 14
9.1 Fernando et al., 2012 HC 24 26.9 96 24.0 4 41 33.0 68 23.7 Saliva AM PHA 14
9.2 Fernando et al., 2012 MDD 24 26.9 96 24.0 4 33 33.4 58 23.6 Saliva AM PHA 14 10.1 Fernando et al., 2013 HC 32 27.9 100 24.7 3 32 29.5 100 23.3 Saliva PM BASAL 16
11.1 Garbutt et al., 1983 HC 15 28.0 40 NR 1 15 31.0 40 NR Serum AM BASAL 11
12.1 Hollander et al., 1994 HC 12 31.2 67 NR 1 15 32.0 33 NR Plasma AM BASAL 8
13.1 Inoue et al., 2015 HC 39 24.4 0 23.7 1 229 25.5 0 23.2 Saliva PM PSY 14
14.1 Jobst et al., 2016 HC 22 30.0 100 NR 3 21 29.7 100 NR Plasma AM PSY 11
15.1 Jogems-Kosterman et al., 2006 HC 22 33.2 100 24.0 3 22 35.7 100 24.7 Saliva AM BASAL 8 16.1 Kahl et al., 2005a HC 16 25.9 100 24.2 4 20 26.1 100 23.1 Serum AM BASAL 7 16.2 Kahl et al., 2005a MDD 16 25.9 100 24.2 4 10 24.2 100 25.1 Serum AM BASAL 7 17.1 Kahl et al., 2005b HC 12 26.8 100 23.6 3 20 26.1 100 23.2 Serum AM BASAL 9 17.2 Kahl et al., 2005b MDD 12 26.8 100 23.6 3 18 31.9 100 24.4 Serum AM BASAL 9 18.1 Kahl et al., 2006a HC 16 25.9 100 24.2 4 20 26.1 100 23.1 Serum AM BASAL 9
ID Study BPD Sample Control Sample Assessment
Cf. n Age Sex BMI Med n Age Sex BMI Type Time Paradigm Quality 18.2 Kahl et al., 2006a MDD 16 25.9 100 24.2 4 12 30.0 100 25.7 Serum AM BASAL 9 19.1 Kahl et al., 2006b HC 12 26.3 100 25.9 1 12 25.6 100 21.8 Serum PM BASAL 5 20.1 Kontaxakis et al., 1987 MDD 13 26.4 0 NR 1 13 43.4 100 21.8 Serum PM DST 11
21.1 Lee et al., 2012 HC 4 33.3 75 NR 1 9 28.3 0 NR Plasma PM PHA 14
22.1 Lieb et al., 2004 HC 23 28.8 100 63.3* 1 24 28.2 38 NR Plasma AM PHA 12
23.1 Lyons-Ruth et al., 2011 HC 16 21.1 100 NR 3 19 22.5 100 65.8* Saliva PM PSY 11
24.1 Martial et al., 1997 HC 5 NR 100 NR 1 6 NR 100 NR Saliva AM BASAL 11
25.1 Nater et al., 2010 HC 15 32.6 100 24.9 1 17 27.2 100 NR Serum PM PSY 15
26.1 Paris et al., 2004 HC 30 27.7 100 NR 1 22 29.0 100 21.4 Saliva AM BASAL 9
27.1 Rausch et al., 2015 HC 35 26.5 100 24.9 0 26 26.3 100 NR BloodÑ AM BASAL 14
28.1 Rinne et al., 2000 HC 12 32.5 100 NR 5 9 27.1 100 22.8 Saliva AM BASAL 9
29.1 Roepke et al., 2010 HC 31 29.0 100 26.1 3 30 28.0 100 NR Plasma AM BASAL 10
30.1 Scott et al., 2013 HC 30 30.4 100 NR 3 33 23.7 100 21.1 Serum PM PSY 13
31.1 Simeon et al., 2007 HC 8 43.4 25 NR 0 11 27.1 100 NR Saliva AM PSY 10
32.1 Simeon et al., 2011 HC 14 35.1 43 NR 1 13 69.0 45 NR Plasma AM PSY 13
33.1 Sinai et al., 2015 HC 92 29.5 100 NR 4 57 39.4 35 NR Plasma PM BASAL 7
34.1 Steiger et al., 2001 HC 34 24.4 100 22.0 1 25 24.6 100 NR Plasma AM BASAL 12 35.1 Steinberg et al., 1997 HC 10 33.6 50 NR 1 11 30.1 100 22.3 Plasma AM BASAL 12
35.2 Steinberg et al., 1997 PD 10 33.6 50 NR 1 24 39.3 45 NR Plasma AM BASAL 12
36.1 Walter et al., 2008à HC 9 18.7 76 NR 2 12 18.7 38 NR Plasma PM PSY 5
37.1 Wingenfeld et al., 2007• HC 21 28.1 100 24.4 1 24 27.7 76 NR Saliva CONT CONT 9
Note. * Only weight in kg indicated; Ñ Not stated if serum or plasma has been assessed; à Demographic details only reported for total sample; • Overnight cortisol examined as a continuous HPA axis measure. The gender ratio is indicated as percentage of participating females. Medication use is defined as: 0 = no medication, 1 = medication washout, 2 = only non-endocrine medications, 3 = mixture of endocrine and non-endocrine medications, 4 = medication use unclear, 5 = medication use not reported. Multiple comparisons within one study were analysed separately.
Risk of Bias Within Studies
Quality was assessed for all studies as related to the selection of participants, quantification of HPA axis function, control for confounding, and experimental HPA axis designs. On average, studies assessing basal cortisol received a quality score of 11 of 18 (range: 5 – 16). Further, studies using experimental designs were awarded an average score of 11 out of 22 (range: 5 – 19) Average scores for the different sections were approximately similar13. It should be noted, however, that just a fraction of publications (k = 4) reported on blinding of HPA axis assessors. All quality ratings for the individual studies, grouped by paradigm, can be found in Appendix D.
Results of Individual Studies
Outcomes of the three separate meta-analyses comparing BPD subjects to HC subjects, MDD subjects, and PD subjects can be found in the ‘Synthesis of Results’ section. Forest plots were included for the basal cortisol comparisons (BPD vs. HC, BPD vs. MDD, BPD vs. PD), pharmacological challenges (BPD vs. HC, BPD vs. MDD), psychosocial stress tests (BPD vs. HC, BPD vs. PD), continuous cortisol assessments (BPD vs. HC) and to visualize the comparisons of different psychosocial stress tests as well as different control groups for the basal comparison between BPD and PD subjects. Subgroup analyses were reported for the basal cortisol comparisons between BPD and HC, MDD, as well as PD subjects. Further, we evaluated publication bias based on funnel plots for the basal cortisol comparisons between BPD and HC, MDD, as well as PD subjects and for the continuous cortisol comparison between BPD and HC subjects. Additional visualizations, such as funnel plots for the comparison of pharmacological challenges, are available on request.
13 I.e. average scores for the different sections were 5/8 for participant selection, 4/6 for quantification of HPA
Synthesis of Results
BPD subjects versus HC subjects. First, meta-analysis revealed that basal cortisol
levels were generally higher in the BPD compared to HC subjects, but this association did not reach statistical significance (Hedges’ g = 0.37, 95% Confidence Interval (95% CI) [-0.06, 0.80], p = .090). There was significant heterogeneity (c2 = 450.9, p < .001; I2 = 93%).
Removing one study (Inoue et al., 2015) considerably decreased heterogeneity (c2 = 74.9, p <
.001; I2 = 61%). Taking into account the deviant effect size of this study (Hedges’ g = 6.80,
95% CI [6,13, 7.47], p < .001) and the high relative weight of the study due to its large control sample (n = 229), the study was considered an outlier and removed from all subsequent analyses, as recommended by Baker and Jackson (2008). A forest plot including the study by Inoue and colleagues (2015) can be found in Appendix E. As shown in Figure 3, the adjusted overall effect size was non-significant (Hedges’ g = 0.13, 95% CI [-0.05, 0.31], p = .161).
Second, post-DST values did not differ significantly between BPD subjects and healthy controls (Hedges’ g = -0.19, 95% CI [-1.12, 0.74], p = .690). The combined effect size was associated with high heterogeneity (c2 = 24.6, p < .001; I2 = 88%). The corresponding forest
plot can be found in Figure 4.
Third, responses to psychosocial stressor tests were evaluated. Both the response to stress (Hedges’ g = -0.29, 95% CI [-0.52, -0.05], p = .017) and recovery from stress (Hedges’
g = -0.34, 95% CI [-0.54, -0.14], p = .001) were characterized by significantly lower cortisol
values in BPD subjects. Heterogeneity was small for both the response comparison (c2 = 12.0,
p = .212; I2 = 25%) and recovery comparison (c2 = 8.6, p = .475; I2 = 0%). Forest plots for the
comparisons of responses to stress and recovery from stress between BPD and HC subjects can be found in Figures 5 and 6.
Figure 3. Comparison of basal cortisol values between BPD and HC subjects excluding one study.
Study name Statistics for each study Hedges's g and 95% CI
Hedges's Standard Lower Upper Weight Relative
g error Variance limit limit Z-Value p-Value (Random) weight
Aleknaviciute et al., 2016 -0,886 0,268 0,072 -1,411 -0,361 -3,309 0,001 4,48 3,91 Bromundt et al., 2013 -0,312 0,402 0,162 -1,101 0,476 -0,776 0,438 3,19 2,78 Carrasco et al., 2007 -0,589 0,296 0,088 -1,169 -0,009 -1,991 0,047 4,19 3,65 Deckers et al., 2015 0,313 0,292 0,085 -0,260 0,885 1,071 0,284 4,23 3,69 Feliu-Soler et al., 2013 -1,001 0,320 0,102 -1,627 -0,374 -3,129 0,002 3,95 3,44 Fernando et al., 2012 0,814 0,264 0,070 0,297 1,331 3,087 0,002 4,53 3,95 Fernando et al., 2013 0,242 0,248 0,061 -0,244 0,728 0,976 0,329 4,70 4,10 Garbutt et al., 1983 0,040 0,355 0,126 -0,656 0,737 0,114 0,909 3,60 3,14 Hollander et al., 1994 1,305 0,415 0,173 0,490 2,119 3,141 0,002 3,09 2,69 Jobst et al., 2016 0,119 0,300 0,090 -0,469 0,706 0,396 0,692 4,15 3,62 Jogems-Kosterman et al., 2006 -0,130 0,296 0,088 -0,711 0,451 -0,438 0,662 4,18 3,65
Kahl et al., 2005a 0,444 0,332 0,110 -0,206 1,095 1,338 0,181 3,82 3,33
Kahl et al., 2005b 0,304 0,358 0,128 -0,398 1,005 0,849 0,396 3,58 3,12
Kahl et al., 2006a 0,461 0,332 0,111 -0,191 1,112 1,386 0,166 3,82 3,33
Kahl et al., 2006b 1,120 0,426 0,181 0,285 1,955 2,629 0,009 3,01 2,62 Lee et al., 2012 0,538 0,569 0,324 -0,577 1,653 0,946 0,344 2,11 1,84 Lieb et al., 2004 0,196 0,288 0,083 -0,368 0,759 0,680 0,496 4,27 3,73 Lyons-Ruth et al., 2011 0,916 0,390 0,152 0,150 1,681 2,345 0,019 3,29 2,87 Martial et al., 1997 0,163 0,555 0,308 -0,924 1,250 0,294 0,769 2,18 1,90 Nater et al., 2010 -0,594 0,353 0,125 -1,286 0,098 -1,682 0,093 3,62 3,16 Paris et al., 2004 0,000 0,276 0,076 -0,542 0,542 0,000 1,000 4,39 3,83 Rausch et al., 2015 0,443 0,259 0,067 -0,064 0,951 1,714 0,087 4,58 3,99 Rinne et al., 2000 0,895 0,445 0,198 0,022 1,768 2,010 0,044 2,86 2,49 Roepke et al., 2010 0,075 0,253 0,064 -0,421 0,570 0,295 0,768 4,65 4,05 Scott et al., 2013 0,338 0,251 0,063 -0,154 0,830 1,347 0,178 4,67 4,07 Simeon et al., 2011 0,074 0,374 0,140 -0,658 0,806 0,198 0,843 3,44 3,00 Sinai et al., 2015 0,000 0,168 0,028 -0,329 0,329 0,000 1,000 5,58 4,86 Steiger et al., 2001 -0,230 0,261 0,068 -0,741 0,282 -0,880 0,379 4,56 3,97 Steinberg et al., 1997 -0,054 0,420 0,176 -0,877 0,768 -0,129 0,897 3,06 2,66 Walter et al., 2008 -0,183 0,431 0,186 -1,029 0,663 -0,424 0,672 2,96 2,58 0,131 0,093 0,009 -0,052 0,314 1,401 0,161 -2,50 -1,25 0,00 1,25 2,50 HC higher BPD higher
Study name Outcome Statistics for each study Sample size Hedges's g and 95% CI
Hedges's Standard Lower Upper Weight Relative
g error Variance limit limit Z-Value p-Value BPD HC Total (Random) weight
Fernando 2012 post-DST 0,353 0,256 0,065 -0,148 0,855 1,381 0,167 24 41 65 1,19 27,01 Carrasco 2007 post-DST -1,383 0,321 0,103 -2,013 -0,754 -4,305 0,000 32 18 50 1,14 25,85 Lieb 2004 post-DST 0,574 0,293 0,086 0,000 1,149 1,961 0,050 23 24 47 1,16 26,37 Lee 2012 post-DST -0,383 0,564 0,318 -1,489 0,722 -0,680 0,497 4 9 13 0,91 20,77 -0,190 0,476 0,227 -1,124 0,744 -0,399 0,690 83 92 175 -2,50 -1,25 0,00 1,25 2,50 HC higher BPD higher
Figure 5. Comparison of cortisol values during stress between BPD and HC subjects.
Figure 6. Comparison of cortisol values during recovery from stress between BPD and HC subjects.
Study name Outcome Statistics for each study Sample size Hedges's g and 95% CI
Hedges's Standard Lower Upper Weight Relative
g error Variance limit limit Z-Value p-Value BPD HC Total (Random) weight
Aleknaviciute et al., 2016 Stress -0,820 0,266 0,071 -1,341 -0,298 -3,079 0,002 26 35 61 9,41 13,47
Deckers et al., 2015 Stress 0,137 0,290 0,084 -0,432 0,706 0,472 0,637 22 24 46 8,35 11,94
Feliu-Soler et al., 2013 Stress -0,546 0,309 0,095 -1,151 0,059 -1,768 0,077 35 15 50 7,65 10,95
Jobst et al., 2016 Stress 0,009 0,299 0,090 -0,578 0,595 0,029 0,977 22 21 43 8,00 11,44
Lyons-Ruth et al., 2011 Stress 0,000 0,371 0,137 -0,727 0,727 0,000 1,000 16 12 28 5,79 8,28
Nater et al., 2010 Stress -0,881 0,362 0,131 -1,591 -0,171 -2,431 0,015 15 17 32 6,00 8,58
Scott et al., 2013 Stress -0,228 0,250 0,062 -0,718 0,262 -0,911 0,362 30 33 63 10,22 14,61
Simeon et al., 2007 Stress 0,000 0,444 0,197 -0,870 0,870 0,000 1,000 8 11 19 4,30 6,16
Simeon et al., 2011 Stress -0,300 0,376 0,141 -1,036 0,436 -0,798 0,425 14 13 27 5,66 8,10
Walter et al., 2008 Stress 0,000 0,430 0,185 -0,844 0,844 0,000 1,000 9 11 20 4,53 6,48
-0,286 0,120 0,014 -0,520 -0,051 -2,390 0,017 197 192 389
-2,50 -1,25 0,00 1,25 2,50
HC higher BPD higher
Study name Outcome Statistics for each study Sample size Hedges's g and 95% CI
Hedges's Standard Lower Upper Weight Relative
g error Variance limit limit Z-Value p-Value BPD HC Total (Random) weight
Aleknaviciute et al., 2016 Recovery -0,745 0,264 0,070 -1,263 -0,227 -2,819 0,005 26 35 61 14,31 14,90 Deckers et al., 2015 Recovery -0,229 0,291 0,085 -0,800 0,341 -0,788 0,431 22 24 46 11,80 12,29 Feliu-Soler et al., 2013 Recovery -0,403 0,306 0,094 -1,003 0,198 -1,314 0,189 35 15 50 10,65 11,09 Jobst et al., 2016 Recovery -0,009 0,299 0,090 -0,596 0,578 -0,030 0,976 22 21 43 11,15 11,61 Lyons-Ruth et al., 2011 Recovery 0,000 0,371 0,137 -0,727 0,727 0,000 1,000 16 12 28 7,27 7,57
Nater et al., 2010 Recovery -0,653 0,355 0,126 -1,349 0,042 -1,841 0,066 15 17 32 7,94 8,27
Scott et al., 2013 Recovery -0,405 0,252 0,063 -0,899 0,088 -1,611 0,107 30 33 63 15,78 16,43
Simeon et al., 2007 Recovery 0,000 0,444 0,197 -0,870 0,870 0,000 1,000 8 11 19 5,08 5,28
Simeon et al., 2011 Recovery -0,719 0,386 0,149 -1,476 0,037 -1,863 0,062 14 13 27 6,71 6,98
Walter et al., 2008 Recovery 0,206 0,432 0,186 -0,640 1,052 0,477 0,633 9 11 20 5,37 5,59
-0,344 0,102 0,010 -0,544 -0,144 -3,374 0,001 197 192 389
-2,50 -1,25 0,00 1,25 2,50
Exploratory analysis: continuous HPA axis outcomes. As four studies compared
continuous cortisol output between BPD and HC subjects, an exploratory meta-analysis was carried out to evaluate the corresponding combined effect size (see Figure 7). Continuous HPA axis measurements refer to assessments of diurnal or overnight cortisol release and are frequently used to measure cortisol secretion under natural, non-laboratory conditions (Lieb et al., 2004)14. As cortisol secretion follows a diurnal rhythm, continuous HPA axis measures represent a more balanced assessment than singular cortisol assessments. Therefore, we reported continuous HPA axis measures separately from basal cortisol assessments.
For the current comparison, we included two studies measuring salivary cortisol during the day (Bromundt et al., 2013; Lieb et al., 2004), and two studies measuring urinary cortisol either during night-time (Wingenfeld et al., 2007) or over the course of 24 hours (Simeon et al., 2007)15. The overall comparison indicated that BPD was characterized by significantly higher continuous cortisol values when compared to HC subjects (Hedges’ g = 0.37, 95% CI [0.04, 0.71], p = .030). No heterogeneity could be detected (c2 = 1.9, p = .603; I2 = 0%).
14 It should be noted that basal cortisol assessments can theoretically be examined in a naturalistic and ambulatory
manner, as well. However, and in contrast to the continuous HPA axis assessments, all studies included for the comparison of basal cortisol assessed cortisol under controlled conditions in a laboratory environment.
15 For the study by Lieb and colleagues (2004), we included seven determinations in 2 h intervals over the course
of the day (called ‘total daily cortisol’ in the primary study). For the study by Bromundt and colleagues (2013), we included seven cortisol determinations in undefined intervals over the course of the day. Both cortisol assessments were calculated based on cortisol areas under the curve (AUCs). Wingenfeld and colleagues (2007) utilized urinary cortisol assessments collected over three consecutive nights (7 PM to 7 AM) and reported an averaged value. The study by Simeon and colleagues (2007) took place from 10 AM at the first day until 10 AM at the second day and reported on the total cortisol output over 24 hours.
Figure 7. Comparison of continuous cortisol output between BPD and HC subjects.
Study name Statistics for each study Sample size Hedges's g and 95% CI
Hedges's Standard Lower Upper Weight Relative
g error Variance limit limit Z-Value p-Value BPD HC Total (Random) weight
Bromundt et al., 2013 0,193 0,401 0,161 -0,592 0,979 0,482 0,630 14 10 24 6,23 18,19 Lieb et al., 2004 0,625 0,294 0,086 0,048 1,201 2,124 0,034 23 24 47 11,57 33,78 Simeon et al., 2007 -0,051 0,444 0,197 -0,921 0,819 -0,115 0,908 8 11 19 5,07 14,82 Wingenfeld et al., 2007 0,398 0,297 0,088 -0,183 0,979 1,343 0,179 21 24 45 11,37 33,21 0,371 0,171 0,029 0,036 0,706 2,170 0,030 66 69 135 -2,00 -1,00 0,00 1,00 2,00 HC higher BPD higher
