A meta-analysis of blood and salivary cortisol levels in first-episode psychosis and high-risk individuals

Dysregulated cortisol responses and glucose metabolism have been reported in psychosis. We performed a random-effects meta-analysis of cortisol responses in first-episode psychosis (FEP) and psychosis risk states, taking into consideration glucose metabolism. A total of 47 studies were included. Unstimulated blood cortisol levels were significantly higher (g = 0.48, 95%CI: 0.25 - 0.70, p < 0.001) in FEP, but not in psychosis risk states (g = 0.39, 95%CI: -0.42 - 1.21, p = 0.342), compared to controls. Cortisol awakening response (CAR) was attenuated in FEP (g = -0.40, 95%CI: -0.68 - -0.12, p = 0.006), but not in psychosis risk states (p = 0.433). Glucose and insulin levels were positively correlated with unstimulated blood cortisol levels in FEP. Our meta-analysis supports previous findings of elevated blood cortisol levels and attenuated CAR in FEP. Future research should focus on identifying the common denominators for alterations in stress hormones and glucose metabolism.


Introduction
The hypothalamic-pituitaryadrenal (HPA) axis is one of the major biological systems responsible for the stress response. Activation of the HPA axis begins with the release of corticotropin-releasing hormone (CRH) by the hypothalamus, which in turn stimulates the production of the adrenocorticotropic hormone (ACTH) by the anterior pituitary gland. Subsequently, ACTH stimulates the adrenal cortex to release glucocorticoids that include cortisol. The HPA axis acts through a negative feedback loop, whereby cortisol decreases the production of CRH and ACTH. Glucocorticoids regulate a number of metabolic and immune processes (Oppong and Cato, 2015;Vegiopoulos and Herzig, 2007). Chronic exposure to glucocorticoids leads to the development of obesity and insulin resistance through several mechanisms (Geer et al., 2014). Specifically, glucocorticoids have been shown to increase food intake by altering expression of appetite-regulating molecules, such as neuropeptide Y or leptin (Solano and Jacobson, 1999). Moreover, glucocorticoids decrease post-receptor insulin signaling by altering expression of downstream signaling molecules, such as IRS-1, p39MAPK and PTP1B (Almon et al., 2005(Almon et al., , 2007. Finally, glucocorticoids promote hepatic gluconeogenesis (Goldstein et al., 2002), lipolysis (Harvey et al., 2018) and fat accumulation in the liver (Marino et al., 2016).
Accumulating evidence suggests that aberrant activity of the HPA axis might be involved in the pathophysiology of psychosis (Pruessner et al., 2017). Indeed, it has been shown that cortisol may enhance dopaminergic signaling in various brain regions, especially in the mesolimbic system (Walker et al., 2008). For instance, Pruessner et al. (2004) found that exposure to psychosocial stress triggers release of dopamine in the ventral striatum as measured by a decrease in [11C] raclopride binding in healthy volunteers. In this study, a positive correlation between dopamine release and cortisol response was observed. Apart from interactions with the mesolimbic system, glucocorticoids interact with their receptors located in other brain regions, including the hippocampus and the prefrontal cortex that are responsible for learning and memory processes. Chronically elevated levels of cortisol can lead to reduced neurogenesis and synaptic plasticity in the hippocampus, contributing to cognitive impairment (Kim et al., 2015). A recent metaanalysis of post mortem studies revealed decreased volume and neuron numbers in several subfields of the left hippocampus in patients with schizophrenia (Roeske et al., 2020). However, there are certain mechanisms that protect the hippocampus against neurotoxic effects of cortisol. For instance, dehydroepiandrosterone (DHEA) and its sulfate (DHEA-S) pass through the blood-brain barrier, serve as the local source of neurosteroids and protect the hippocampus from the effects of corticosterone (Kimonides et al., 1999;Melcangi et al., 2011).
Previous meta-analyses of the HPA axis activity have demonstrated the following alterations: 1) blunted cortisol awakening response (CAR) in first-episode psychosis (FEP) and schizophrenia) (Berger et al., 2016); 2) attenuated cortisol response to social stress in schizophrenia (Ciufolini et al., 2014); 3) increased morning cortisol levels in blood and saliva of individuals with schizophrenia and FEP (Girshkin et al., 2014); 4) increased baseline cortisol levels in blood of individuals with FEP (Hubbard and Miller, 2019) and 5) increased baseline cortisol levels in saliva of individuals at clinical high risk (CHR) of psychosis (Chaumette et al., 2016). However, most meta-analyses in this field have addressed single parameters of the HPA axis activity and have not considered cortisol levels in blood and saliva separately as well as their association with other hormonal and metabolic alterations reported in this population, including impaired glucose homeostasis (Pillinger et al., 2017) and altered levels of DHEA or DHEA-S (Misiak et al., 2018). Importantly, assessment of blood cortisol levels provides the information on the total fraction of cortisolfree and bound to proteins. In turn, collection of saliva samples is non-invasive and provides the opportunity to determine the free fraction of cortisol that elicits biological activity (Kirschbaum and Hellhammer, 1989;Kudielka et al., 2012). Indeed, the acinar cells that are located in the salivary glands prevent the transition of proteins and protein-bound molecules to saliva (Kirschbaum and Hellhammer, 1994). Another shortcoming of previous research is the lack of a systematic comparison of baseline cortisol levels and cortisol levels in response to stimulation. Finally, there is some evidence that unaffected relatives of individuals with psychosis show increased emotional reactivity to daily life stress, increased ACTH levels in response to stress, increased pituitary volume and reduced hippocampal volume (Aiello et al., 2012). These observations provide the rationale for investigating the association between genetic liability for psychosis and the HPA axis responses. Therefore, in this study we aimed to perform a meta-analysis investigating baseline cortisol levels and cortisol levels in response to stimulation in blood and saliva in subjects with first-episode psychosis (FEP), CHR individuals, as well as unaffected relatives of individuals with schizophrenia, taking into consideration a number of potential moderators.

Search strategy
Five databases (MEDLINE, ERIC, CINAHL Complete, International Pharmaceutical Abstracts as well as the Academic Search Ultimate and the Health Source: Nursing/Academic Edition) were searched from their inception until 1st Feb 2021 by two independent reviewers (BS and MW). The following keywords were used: "chr" OR "uhr" OR "arms" OR "risk" OR "siblings" OR "relatives" OR "offspring" OR "first-episode" OR "antipsychotic-naïve" OR "drug-naïve" OR "unmedicated" OR "never medicated" AND "psychosis" OR "schizophr*" AND "cortisol" OR "HPA axis" OR "glucocorticoid". The third reviewer (BM) was involved in resolving all disagreements. Online searches were conducted in agreement with the PRISMA guidelines (Moher et al., 2009). The protocol of this meta-analysis was registered in the PROSPERO database (registration number: CRD42021185601). The PRISMA checklist was presented in Supplementary Table 1.

Eligibility criteria
Publications were considered eligible if they met the following criteria: 1) reported the blood and/or salivary levels of unstimulated (baseline) cortisol and/or CAR as measured by the area under the curve with respect to increase (AUCi) or ground (AUCg) (Pruessner et al., 2003); 2) necessary data (mean and SD for baseline blood and/or salivary levels of cortisol for the CAR as measured by the AUCi as well as sample size) were available in the article or upon request from the corresponding author; 3) cross-sectional studies comparing blood and/ or salivary levels of cortisol and/or CAR as measured by the AUCi or AUCg between persons with FEP or CHR individuals or unaffected individuals at familial risk (FHR) of psychosis (first-and second-degree relative of individuals with schizophrenia) and healthy controls or between high-risk converters and non-converters and 4) English language full-text articles. Three subgroups of individuals were included: 1) patients with FEP; 2) CHR individuals diagnosed using the Comprehensive Assessment of At-Risk Mental State (CAARMS) (Yung et al., 2005) or the Structured Interview for Prodromal Syndromes (SIPS) (Miller et al., 2003), including those with schizotypal personality disorder and 3) FHR individuals (first-and second-degree relatives of patients with schizophrenia). The following types of publication records were excluded: 1) animal model studies; 2) non-original studies (e.g., reviews, commentaries and editorials); 3) publications without necessary data; 4) studies that did not include healthy controls or did not record transition to overt psychosis. In case of overlapping authorship of potentially eligible publications, corresponding authors were contacted to provide necessary data from the largest sample of non-overlapping participants. In case of a lack of response from the corresponding author, only the largest study was included. Similarly, corresponding authors of eligible publications were contacted in case of unclear findings or a lack of necessary data to perform meta-analysis.

Data extraction
The following data were retrieved: 1) age; 2) sex; 3) body-mass index (BMI); 4) cigarette smoking status; 5) unstimulated (baseline) salivary levels of cortisol (i.e., single cortisol levels in the absence of stimulation or before stimulation or the values from the first measurement during the day); 6) unstimulated blood levels of cortisol; 7) CAR as measured by the AUCi or AUCg; 8) type of assay used to measure the level of cortisol; 9) type of biological material used to measure the level of cortisol (saliva, serum or plasma); 10) number of biological material samples used to calculate the level of cortisol; 11) conditions of biological material sampling (fasting or non-fasting, time of sampling); 12) medication status; 13) illness duration; 14) scores of the Positive and Negative Syndrome Scale (PANSS) (Kay et al., 1987) and 15) the levels of other biochemical parameters determined by at least six studies. The terms "baseline", and "unstimulated" with respect to cortisol levels are used interchangeably. Variables were retrieved as mean and SD or a number of cases.
Conversion methods were used to calculate mean and SD in case of data provided as median and interquartile range (IQR) or 95% confidence interval (95 %CI) or standard error (SE). The median was included as an approximation of the mean (Higgins and Green, 2011). In turn, SD was calculated using the following formulas: 1) SD = IQR/1.35; 2) SD = SE × ̅̅̅ ̅ N √ or 3) SD = ̅̅̅̅ N √ × (upper limit of 95 %CI -lower of 95 %CI)/ 3.92 (Higgins and Green, 2011;Hozo et al., 2005). Effect size estimates for biochemical parameters that were tested as potential moderators were calculated as described elsewhere (Lipsey and Wilson, 2001).

Quality assessment
The Newcastle-Ottawa Scale (NOS) (Wells et al., 2000) was used to assess quality of eligible studies. The NOS is a "star system" developed to evaluate three categories of quality: 1) the selection of study groups (the maximum score is 4 stars); 2) the comparability of study groups (the maximum score is 2 stars) and 3) the ascertainment of exposure or outcome of interest (the maximum score is 3 stars). The total NOS score ranges from 0 to 9 stars, where higher scores indicated better quality. Matching patients and healthy controls for age and sex was considered to allocate stars for the comparability of studied groups.

Data analysis
Between-group differences in baseline blood and salivary levels as well as CAR were tested as main outcome measures. The Cochran Q and I 2 statistics were used as the measures of heterogeneity. Random-effects models were used due to between-study differences in sample size, recruitment procedures, diagnoses or clinical manifestation captured by FEP or high-risk states and methods used to measure cortisol responses. Indeed, random-effects models are recommended if eligible studies are heterogeneous from a clinical and methodological point of view, and thus it is unreasonable to assume that they share a common effect (Tufanaru et al., 2015). The Hedges' g was calculated as the effect size estimate. The number of effect size estimates included in each analysis was reported as the "k" measure. The leave-one-out sensitivity analysis was performed to investigate if any single study accounted for heterogeneity. All analyses were carried out separately for studies of individuals with FEP, CHR and FHR. Meta-regression analyses were performed in case of continuous moderators that were assessed by at least six studies and categorical moderators if each category was represented by at least four studies (Fu et al., 2011). Publication bias was evaluated using the Egger's test if specific outcome measures were determined by at least 10 studies (Sterne et al., 2008). The level of significance was set at p < 0.05. The STATISTICA software, version 12.5, was used to perform data analyses.
Blood cortisol levels were significantly higher in patients with FEP (k = 22, g = 0.48, 95 %CI = 0.25-0.70, p < 0.001) but not individuals at high risk of psychosis (k = 7, g = 0.39, 95 %CI = − 0.42-1.21, p = 0.342) compared to healthy controls (Table 3, Fig. 2). Heterogeneity was significant in these analyses [FEP vs. controls: I 2 = 83.3%, Q = 125.5, p(Q) < 0.001; high-risk individuals vs. controls: I 2 = 94.6%, Q = 110.8, p(Q) < 0.001] The leave-one-out sensitivity analysis revealed that no single study accounted for heterogeneity in studies investigating unstimulated blood cortisol levels in subjects with FEP and those at high risk of psychosis (Supplementary Table 4). Standardized mean differences for the levels of glucose (k = 9, β = 0.85, 95 %CI = 0.28 -1.43, p = 0.004) and insulin (k = 7, β = 2.11, 95 %CI = 0.60 -3.62, p = 0.006) were associated with significantly higher standardized mean differences for blood cortisol levels in studies of individuals with FEP (Supplementary Table 5, Supplementary Fig. 1, Supplementary Fig. 2). These moderators were not tested separately for high-risk individuals due to low number of studies. Other variables were not significantly associated with effect size estimates for blood levels of cortisol (mean difference in age, difference in percentage of males, type of assay, fasting, difference in percentage of smokers, mean difference in BMI, percentage of antipsychotic-naïve patients or high-risk individuals, the NOS score, as well as effect size estimates for blood lipids, DHEA and DHEA-S). Results of the Egger's test were not significant for studies of individuals with FEP, indicating
The leave-one-out sensitivity analysis revealed that after removing single studies (Chaumette et al., 2016;Day et al., 2014;Pruessner et al., 2013a; salivary cortisol levels were significantly higher in CHR individuals compared to healthy controls with no evidence of heterogeneity (Supplementary Table 4). However, no significant betweengroup differences in salivary cortisol levels were found after removing single studies from the subgroup analysis of patients with FEP. Metaregression analysis (Supplementary Table 5, Supplementary Fig. 3) demonstrated that the NOS score was significantly associated with effect size estimates for salivary cortisol levels in the subgroup analysis of studies that included CHR individuals (k = 10, β = − 0.22, 95 %CI = -0.39 -− 0.06, p = 0.009). Other potential moderators were not found to be significantly associated with effect size estimates for salivary cortisol in studies of CHR individuals (mean difference in age, difference in the percentage of males, number of saliva samples, single measure at awakening vs. other sampling schedules, type of assay and the percentage of antipsychotic-naïve CHR individuals). Due to the low number of studies that included patients with FEP, only the mean difference in age and the NOS score were tested in meta-regression analyses. However, none of them was found to be significantly associated with the effect size estimates of salivary cortisol in studies of patients with FEP. Importantly, none of eligible studies investigating unstimulated salivary cortisol levels measured the parameters of glucose homeostasis. Results of the Egger's test were not significant for studies of CHR individuals (k = 10, regression intercept = 1.21, 95 %CI = -0.73 -3.15, p = 0.189).
controls. The leave-one-out sensitivity analysis (Supplementary Table 4) revealed that the CAR remained attenuated after removing any single study, except from the findings of the studies by Pruessner et al. (2013b; (k = 3, g = -0.42, 95 %CI = -0.83-0.01, p = 0.050). After removing these studies heterogeneity was not significant [I 2 = 61%, Q = 5.1, p(Q) = 0.077]. Heterogeneity remained not significant in sensitivity analysis. The difference in CAR between high-risk individuals and controls was not significant after removing any single study in sensitivity analysis. Due to low number of studies, meta-regression analyses and the Egger's test were not performed. Moreover, none of eligible studies investigating CAR measured the parameters of glucose homeostasis.

Unstimulated blood cortisol levels
Our systematic review and meta-analysis demonstrated significantly higher unstimulated levels of blood cortisol in patients with FEP but not in high-risk individuals with medium effect size estimates and significant heterogeneity. The difference in unstimulated levels of blood cortisol was not associated with between-group differences in age, sex, BMI and cigarette smoking status. Moreover, quality of studies, type of assay, fasting and the use of antipsychotics were not significantly correlated with effect size estimates. These findings are consistent with the results of previous meta-analysis of blood cortisol in patients with FEP (Hubbard and Miller, 2019). However, for the first time, we investigated a number of other biological alterations as potential moderators (the levels of glucose, insulin, DHEA and DHEA-S as well as lipid profile). We found that the effect size estimates for blood cortisol levels are significantly associated with those for the levels of glucose and insulin. It should be noted that abnormal glucose homeostasis parameters in unmedicated patients with FEP have previously been reported by several meta-analyses (Greenhalgh et al., 2017;Perry et al., 2016;Pillinger et al., 2017). Although a cross-sectional design of studies included in this meta-analysis does not allow to provide insights into causality, two scenarios need to be considered. First, it is likely that the enhanced activity of the HPA axis contributes to impaired glucose homeostasis. Indeed, glucocorticoids may liberate energy substrates through enhancing hepatic gluconeogenesis and reducing glucose utilization in fight-or-flight circumstances (Geer et al., 2014). Second, it cannot be ruled out that impaired glucose homeostasis in FEP accounts for elevated blood cortisol levels. For instance, it has been shown that glucose load amplifies cortisol responses to psychosocial stress in healthy volunteers (Gonzalez-Bono et al., 2002;von Dawans et al., 2021).
Of note, our findings do not explain the mechanisms underlying elevated blood cortisol levels in patients with FEP and concomitant   glucose homeostasis alterations. However, certain processes should be taken into consideration. Convincing evidence indicates that traumatic life experiences, especially those acting in childhood, increase a risk of psychosis (Beards et al., 2013;Misiak et al., 2017;Varese et al., 2012). Their impact seems to be long-lasting as patients with psychosis and a history of childhood trauma show poor clinical and functional outcomes (Alameda et al., 2017;Misiak and Frydecka, 2016;Mondelli et al., 2015;Pruessner et al., 2019Pruessner et al., , 2021. Neuroimaging studies have also identified some neural substrates of childhood trauma in psychosis. Indeed, it has been reported that a history of early-life stress is associated with reduced volumes of the hippocampus and the amygdala in adult patients with psychosis (Aas et al., 2012;Hoy et al., 2012). In addition, more proximal stressors play an important role in triggering relapse of psychosis (Martland et al., 2020). However, it is important to note that a recent meta-analysis demonstrated poor concordance between naturally occurring psychosocial stressors and cortisol responses in psychosis (Cullen et al., 2020b). Nevertheless, the authors of this meta-analysis noted that this observation should be interpreted with caution due to limited statistical power at the level of meta-analysis and individual studies, heterogeneity of study measures and populations as well as differences in timing of cortisol measurement with respect to the onset of stressor. Moreover, experiencing psychosocial stress is neither necessary nor sufficient to trigger the onset of psychosis, and may contribute to the development of psychopathology that falls beyond the psychosis spectrum. This observation suggests that certain factors might moderate the association between psychosocial stress and a risk of psychosis or cortisol responses. These moderators might include individual psychological characteristics (e.g., personality traits, coping strategies or resilience) (Mizuno et al., 2016;Oswald et al., 2006;Piotrowski et al., 2020), social environments (Gayer-Anderson et al., 2015) or even genetic backgrounds (Stramecki et al., 2020). Interestingly, it has also been observed that a history of childhood trauma might contribute to insulin resistance in patients with schizophrenia and FEP Perry et al., 2021;Tosato et al., 2020). Apart from the associations with psychosocial stress, it cannot be excluded that elevated blood cortisol levels in FEP simply reflect a broad context of biological dysregulations. For instance, it has been shown that individuals with FEP show an elevated allostatic load index that captures a number of cardiovascular, metabolic, neuroendocrine and immune-inflammatory alterations, including elevated cortisol levels (Piotrowski et al., 2019). Finally, it is also important to note that we did not find significant correlations of effect size estimates for blood levels of cortisol with those for DHEA and DHEA-S. Elevated levels of DHEA and DHEA-S, two hormones that exert anti-glucocorticoid activities (Kimonides et al., 1999), have been welldocumented in patients with FEP (Misiak et al., 2018). However, based on our meta-analysis, it might be implied that the HPA axis and DHEA/DHEA-S responses in FEP are not linearly associated.

Unstimulated salivary cortisol levels
Another important observation is that we found no convincing evidence that unstimulated salivary cortisol levels are altered in subjects at high risk of psychosis or those with FEP compared to healthy controls. Notably, heterogeneity was not significant in these analyses. After removing two single studies (Chaumette et al., 2016;Day et al., 2014) and the joint data from studies by Pruessner et al. from the subgroup analysis of CHR individuals, baseline salivary cortisol levels were significantly higher in CHR individuals compared to healthy controls. It should be noted that Chaumette et al. (2016) recruited help-seeking healthy controls as opposed to other eligible studies, which might explain why between-group differences in salivary cortisol were not significant. In case of other studies, it is difficult to indicate factors that might account for the results of sensitivity analysis. Importantly, the quality of studies was strongly and negatively associated with the effect size estimates. Nevertheless, a lack of significant difference with respect to unstimulated salivary cortisol levels between individuals with FEP and healthy controls might be surprising in light of our findings on blood cortisol levels. Although assessment of blood levels provides the information on total cortisol fraction, while salivary cortisol levels capture the unbound and biologically active fraction of cortisol, both measures are highly correlated Hellhammer, 1989, 1994). Additional studies of salivary cortisol in psychosis are warranted, and should include multiple measurements, controlling for potential confounding factors. Moreover, methods like experience sampling, which allows to explore cortisol responses to daily hassles in real life environments, might provide more insights into the dynamics of salivary cortisol in subjects at the early phase of psychosis development (Myin-Germeys et al., 2018).

CAR
The present meta-analysis showed significantly attenuated CAR in subjects with FEP. Heterogeneity was not significant in this analysis. No significant differences with respect to the CAR were found between high-risk individuals and healthy controls. Only the study by Cullen et al. (2014) revealed a blunted CAR in children, who were relatives of individuals with schizophrenia, suggesting that the CAR might be associated with familial liability to psychosis. Notably, this was the only study that included unaffected FHR individuals. Altogether, these findings are in agreement with those of a previous meta-analysis (Berger et al., 2016), and would suggest that the CAR becomes attenuated after the onset of psychosis.
The CAR captures the marked increase in cortisol levels that appears during the first 30-45 min. after morning awakening (Stalder et al., 2016). It is independent of baseline or daytime cortisol levels and cortisol responses to psychosocial stress (Fries et al., 2009;Kidd et al., 2014). Studies included in the present meta-analysis also demonstrated interesting correlates of the CAR in subjects with FEP. Indeed, an attenuated CAR was associated with unfavorable paternal parenting under the age of 16 years (Pruessner et al., 2013b), greater cortical volume reductions (Pruessner et al., 2015;2017a) and poor treatment response (Mondelli et al., 2015). However, it should be noted that we were not able to perform meta-regression analyses for the CAR.

Sex differences in cortisol response
It is important to note that the vast majority of eligible studies did not investigate sex differences in cortisol responses. Indeed, sex differences were investigated only in one dataset (Pruessner et al., 2013b;. The authors found that the CAR was significantly lower in males with FEP compared to females with FEP and male controls. Also, a significant positive correlation between CAR and hippocampal volumes was found in males with FEP (Pruessner et al., 2015). Sex differences were also reported in a study by Carol et al. (2016) that was excluded from the present meta-analysis due to overlap with findings from a larger sample (Carol et al., 2017). More specifically, the authors found a significantly blunted CAR in CHR males but not in CHR females.
Sex differences with respect to cortisol responses might be of particular relevance to the pathophysiology of psychosis. For instance, females with schizophrenia tend to show later age of psychosis onset, more affective and less negative symptoms as well as less severe course of illness (Abel et al., 2010). These observations might be attributed to differences in hormonal homeostasis between females and males that are important with respect to cortisol responses. For instance, it has been -1,4 -1,0 -0,6 -0,2 0,2 0,6 1,0 1,4 reported that testosterone and cortisol levels are positively correlated, and this association does not change with age (Harden et al., 2016). Sex differences in testosterone levels in patients with schizophrenia have also been explored in our previous meta-analysis (Misiak et al., 2018). Although we were not able to address these aspects in FEP due to a low number of studies, we found that total testosterone levels among individuals with multiple-episode schizophrenia are elevated in females and reduced in males, likely due to sex-dependent impact of antipsychotics. Moreover, greater hippocampal volume reductions have been reported in males with schizophrenia (Adriano et al., 2012).

Limitations
Our meta-analysis is characterized by certain limitations. Although it was based on a large number of studies, in the majority of them sample size was relatively low and representativeness was limited. In case of certain subgroup analyses, the number of eligible studies was also low. Moreover, it should be noted that we were unable to perform separate subgroup analyses of CHR and FHR individuals. In case of blood cortisol levels, subgroup analysis of high-risk individuals included only one study of CHR subjects. Moreover, due to a lack of data, we were unable to address the effects of clinical heterogeneity in the CHR construct. Another limitation is that our meta-analysis was based on a limited number of studies that compared cortisol measures in CHR converters and non-converters. Findings on the salivary cortisol levels should also be interpreted with caution due to the effects of quality scores. Importantly, the present meta-analysis does not provide insights into causal mechanisms underlying reported cortisol alterations. In addition, it is important to note that heterogeneity was not fully explained and was high in certain analyses, e.g., those of blood cortisol levels. Finally, results of this meta-analysis should be interpreted with caution due to the high number of statistical tests performed overall.

Conclusions
In summary, most robust findings of the present meta-analysis confirm elevated blood cortisol levels and attenuated CAR in patients with FEP. However, the present findings also allow to build new observations upon existing evidence (Berger et al., 2016;Chaumette et al., 2016;Hubbard and Miller, 2019;Pruessner et al., 2017). First, we showed that the HPA axis dysfunction is associated with impairments of glucose homeostasis widely demonstrated in FEP. This association needs to be explored further to clearly indicate direction of causality, and might open new possibilities to develop early interventions aimed to improve somatic health in FEP. Second, it is of importance to initiate studies that combine various measures of the HPA axis activity as discrepant observations together with the effects of methodological factors are reported in the present meta-analysis. These studies should strictly follow existing recommendations on the measurement of the HPA axis activity and take into consideration potential sex differences (Miller et al., 2013;Ryan et al., 2003a;2003b;;Stalder et al., 2016). Third, available evidence does not support the utility of the HPA axis measures in predicting transition to psychosis in CHR individuals. However, these observations are based on limited number of studies, and thus additional research in this field is needed. A clear definition of HPA axis dysfunction is also warranted due to the fact that the efficacy of anti-glucocorticoid treatments in psychosis has not been sufficiently documented (Garner et al., 2016).