Hypothalamic-Pituitary-Adrenal Axis Dysregulation in Depressed Children and Adolescents: A Meta-Analysis

Summary

Research findings on the hypothalamic-pituitary-adrenal (HPA) axis and pediatric depression reflect a variety of methodological approaches that tap different facets of HPA-axis functions. Partly owing to the methodological heterogeneity of studies, descriptive reviews of this area have produced inconsistent conclusions. Therefore, we conducted formal meta-analyses of pertinent studies in order to advance our understanding of HPA-axis dysregulation in pediatric depression. We examined: a) 17 published studies of HPA-axis response to the dexamethasone suppression test in depressed youth (DST; N=926) and b) 17 studies of basal HPA-axis functioning (N=1,332). We also examined descriptively studies that used corticotropin releasing hormone (CRH) infusion, and those that used psychological probes of the HPA-axis. The global standardized mean effect size difference in HPA-axis response to the DST between depressed and non-depressed youth was .57, z = 4.18 p< .01. The global standardized mean difference effect size in basal HPA-axis functioning was .20, z = 4.53, p < .01. Age, sex, timing of sampling, dexamethasone dosage, or type of control group was not a significant source of variability for the DST or basal studies. In addition, when compared to non-depressed peers, depressed youth have a normative response to CRH infusion but an overactive response to psychological stressors. In conclusion, the HPA-axis system tends to be dysregulated in depressed youth, as evidenced by atypical responses to the DST, higher baseline cortisol values, and an overactive response to psychological stressors. This pattern of dysregulation suggests anomalies within the axis's negative feedback system and CRH production, but intact pituitary and adrenal sensitivity.

In the quest for biological markers of depression, researchers have explored the link between hypothalamic-pituitary-adrenal (HPA) axis functioning and adult depression for at least forty years, extending the inquiry to pediatric depression in more recent decades. Much of this work has involved the examination of cortisol response to biological and psychological probes or the assessment of basal cortisol levels. However, recent descriptive reviews of the pediatric literature have yielded inconsistent findings, with some reviewers concluding that the association between dysregulated HPA-axis and child depression is inconclusive at best (Birmaher et al., 1996; Birmaher & Heydl, 2001), while others suggesting that HPA-dysregulation in pediatric depression is consistent with that found in adults (Kaufman et al., 2001).

While earlier studies typically compared responses to the dexamethasone suppression test (DST) of depressed and non-depressed pediatric samples, there has been an increase in research using other methodologies to examine the HPA-axis, such as CRH infusion tests, psychological stress exposure, and measurements of basal levels at key times during the day. These different methodologies tap different components of the HPA-axis. Therefore, a careful analysis of group differences in cortisol functioning, as assessed by various methods, is needed to better understand of the specific components of the HPA-axis that may be associated with major depression in pediatric populations.

For example, upon administration of dexamethasone, the HPA-axis suppresses the production of corticotropin-releasing hormone (CRH) by the hypothalamus and adrenocorticotropic releasing hormone (ACTH) by the pituitary, eventually leading to a decrease of cortisol secretion (normative response). The non-suppression of cortisol after DST challenge indicates a dysregulation within the HPA-axis negative feedback mechanism (Burke et al., 2005). Such dysregulation appears to reflect tonic HPA-axis functioning and is usually considered an indication of hypercortisolaemia. In contrast, exogenous infusion of CRH results in the rapid release of ACTH by the pituitary gland followed by an increase in cortisol production by the adrenal gland. Relative suppression (limited increase) of ACTH and cortisol release after CRH infusion suggests pituitary under-sensitivity to CRH. Likewise, non-stress induced cortisol levels (basal) obtained at key hours of the day may reflect distinct components of the HPA-axis. For example, while high levels of total cortisol production during a 24-hour cycle indicate possible tonic hypercortisolaemia, samples obtained upon awakening (during the cortisol awakening response) are not necessarily reflective of tonic HPA-axis functioning and instead reflect stress reactivity (Edwards et al., 2001; Schmidt-Reinwald et al., 1999). Finally, while DST and CRH tests, and to a lesser extent basal cortisol levels, can mirror key areas of the HPA-axis system, responses to psychological probes may reflect differences within various HPA-axis components, including dysregulation of the negative feedback system, oversensitivity to CRH or ACTH, and/or differences in cognitive factors (e.g., attention, depressive rumination) that could lead to acute activation of the HPA-axis.

In addition to differences in the methods used to assess HPA-axis functioning, studies have also differed along a number of dimensions that may have contributed to the inconsistencies of past findings, such as age (e.g., Dahl et al., 1992), time of data collection (e.g., Forbes et al., 2006; Mannie et al., 2007), or whether the comparison group consisted of youths with psychiatric diagnoses other than depression (e.g., Casat et al., 1994) or normal controls (e.g., Feder et al., 2004; Forbes et al., 2006). While previous descriptive reviews have examined these factors as possible moderators of the MDD-cortisol relationship, there is no quantitative review of the effects of these variables on the various indices of HPA-axis functioning in depression.

We conducted a meta-analytic review of the HPA-axis literature on pediatric depression with the overall goal of addressing several of the above noted issues. Specifically, we focused on comparing the different research methodologies used. To this end, we grouped studies according to whether they relied on a) biological probes such as the dexamethasone suppression test or corticotropin releasing hormone infusion, b) basal cortisol levels throughout the day, and c) psychological stress probes. In addition to assessing the main effects of each experimental approach, we examined the putative effects of age, sex, and age and sex distribution imbalances across MDD and control samples, timing of sampling during the day, and type of control group.

Methods

Study Selection and Search Strategy

We restricted our search to peer-reviewed journals and sought published articles, which reported on HPA-axis functioning of depressed youth. We used PubMed, PsycINFO, and Google Scholar™ to search for studies from the earliest available date up to February 15th, 2009. We searched abstracts and titles using iterations of the following search terms: depression, depressed, WITH child, children, childhood, adolescents, adolescence, WITH cortisol, HPA, CRH, and hormones. We also searched the reference lists of recent reviews of this area, as well as the reference lists of primary sources. Finally, we contacted key researchers in the field to ensure we did not miss any eligible studies.

After duplicate findings were eliminated, our search strategy yielded 157 unique articles. The abstracts and methods of these articles were reviewed and those that met full inclusion criteria were chosen for the meta-analysis. Articles were included if all of the following conditions were met: a) the design included a depressed group consisting of children or adolescents with diagnosed major depressive disorder or dysthymia, b) the depressed group did not include bipolar cases, c) the design included a non-depressed comparison group, d) the methodology entailed either a biological challenge relevant to the HPA axis, experimental or naturally occurring psychological stress, or characterization of diurnal variation of plasma, salivary, or urinary cortisol levels, and e) the outcome was reported as mean cortisol differences between MDD and non-depressed groups, or differences in rates of non-suppressors in the case of DST studies.

We then examined the selected studies for potential sample overlap. When two articles had sample overlap but presented non-redundant data (e.g., obtained at different times or during different protocols) the two articles were kept in the analysis and clustered as described in the statistical section below. If the two articles presented sample overlap and reported redundant data, the study with the largest sample size was selected.

Based on our inclusion criteria and procedures, we identified 17 DST articles, 17 articles presenting basal cortisol levels, 4 articles using corticotropin releasing hormone (CRH) infusion, and 3 articles using psychological stressors. Therefore, we were able to conduct separate meta-analysis only on studies using the DST and those reporting basal cortisol levels. The limited number of CRH infusion and psychological stressor studies precluded a formal meta-analysis and thus for these methodologies we present only a descriptive comparison of effect sizes.

Effect Size Calculation

Effect sizes were calculated for two types of outcome data, namely: 1) mean cortisol differences between MDD and control groups and 2) difference between MDD and control group in rate (χ²) of suppression vs. non-suppression to the DST. Hedge's G (Hedges, 1982) weighted effect size was used as metric for all mean comparison. Hedge's G adjusts for differences in sample sizes and yields a more conservative metric than Cohen's D (Grissom & Kim, 2005). Studies presenting χ² statistics were also transformed to a Hedge's G scale and adjusted with Hedge's sample size weight adjustment. Studies that reported comparisons between an MDD group and more than one control sample were treated as independent samples nested within studies, yielding 18 DST and 27 basal samples nested within 17 and 17 studies, respectively.

Meta Analytic Approach

We used a hierarchical multilevel mixed-model approach (Kalaian & Raudenbush, 1996) similar to that employed by Dickerson and Kemeny (Dickerson & Kemeny, 2004). This approach allows the use of multiple effect sizes per study, as it controls for the non-independence of observations when a single study provides multiple comparisons varying across potential moderator variables (e.g., time of the day). Therefore, this model accounts for the embedded nature of the data in which some effect sizes are embedded within studies. We included multiple effect sizes per study as long as they represented comparisons that differed in at least one potential moderator. However, when the effect sizes were simple repeated observations of the same condition (e.g., morning basal samples obtained on 3 consecutive days), the data were pooled and presented as a single effect size. For all computations we used SAS PROC MIXED syntax designed for multi-level models (Singer, 1998) with special application for meta-analysis (van Houwelingen et al., 2002; Normand, 1999).

First, we calculated the overall effect size using maximum likelihood in a random effects model. Our model employed a two level structure. At the first level (within-study level), each observed effect size within a study (d_ij) is a function of the true effect size for the study (β_0j) and within-study measurement error (R_ij):

$${\text{d}}_{\text{ij}}={\beta }_{\text{0j}}+R_{\text{ij}}.$$

At the second level (between-study level), each true study effect size is a function of the overall effect size across all studies (γ₀₀) and random residual error:

$${\beta }_{\text{0j}}={\gamma }_{00}+U_{\text{0j}}.$$

This structure was used to calculate 1) the overall effect size (γ₀₀), 2) whether the overall effect size was significantly different from zero, and 3) the heterogeneity of effect sizes across studies. The heterogeneity, as observed in the variance of the random residual error (U_0j), indicates whether there is variance that was not accounted for by the random effects model. However, we proceeded to examine possible factors that explain differences between studies even if heterogeneity was not observed, as estimates of heterogeneity have some limitations (Ioannidis, 2008), and significant moderating factors may be present even when the variability between studies is not large (see Hall & Rosenthal, 1995).

Then, we expanded the previous model to examine the effects of two types of within- and between-study factors. First, we entered time of the day for each cortisol assessment as a within-study level 1 factor. Second, between-study level 2 factors included type of control group (non-psychiatric vs. psychiatric), mean age of subjects (years), and sex (% male). For the DST studies only we also included DST dosage (.5mg vs. 1mg) as a between-study factor.

Publication Bias

Assessment and control for publication bias was conducted via a modified version of the Trim and Fill method as proposed by Duval and Tweedie (2000). First, a funnel plot was created by plotting the effect sizes against standard error (specifically 1/SE). The possible number of missing studies was estimated via a modified rank-base multiple iteration method leading to the L₀ statistic (Duval & Tweedie, 2000). The L₀ statistic was originally defined on a “naive” random effect size which assumed that all measurements were independent (i.e. sampled in clusters of size 1). To ensure that our analysis was not unduly weighted by studies with replicate measures as expected, given our multilevel hierarchical modeling, we estimated the mean effect size and pooled standard error within each study sample, and used these as the primary points in our trim-and-fill analysis. We computed a “Clustered L₀”, by comparing the mean effect sizes to a “grand mean” intercept from a random effects model that assumed that measurements were sampled in clusters of varying size. When one or more missing studies were identified (Cluster L₀ > 0.50), the missing data were estimated based on the symmetric data (in our case, cluster mean effect sizes and pooled variances) to fill the most extreme lower left cluster from our funnel plot. Using this estimated data, a new overall fixed effects model was fitted and reported.

Results

Dexamethasone Suppression Test: Meta-Analysis

Main effects

We identified 17 studies comparing MDD and non-MDD controls on post-DST cortisol levels (total 18 samples with N = 926; see Table 1). The studies contributed a total of 49 effect sizes. Using a mixed, random effects model, the pooled effect size (k = 49) for group differences in DST response was .57 (z = 4.18, p < .01; 95% CI .28 – .86) indicating greater cortisol production (or less suppression) after the DST among depressed children and adolescents than among non-MDD controls. The estimate of heterogeneity was significant (.26), z = 2.33, p < .01, suggesting large between-study variability. Figure 1 presents the effect sizes and 95% confidence intervals for the studies included in this analysis.

Table 1.

Study characteristics and mean effect size differences in cortisol response to the DST among depressed youth and comparison groups.

ID	Author	Year	% Inpatient	Control Type	Sex MDD (%males)	Sex-Cont (%males)	Age-MDD	Age-Cont	Time	Dosage	N-MDD	N-Cont	G
1.01	Birmaher et al.	1992	0	Norm	50	66	14.9	15.3	23	1	44	38	0.03
1.02	Birmaher et al.	1992	0	Norm	50	66	14.9	15.3	16	1	44	38	0.30
1.03	Birmaher et al.	1992	0	Norm	50	66	14.9	15.3	8	1	44	38	0.39
2.01	Casat et al.	1994	100	Psych	55	78	10.1	10.8	16	0.5	11	9	-1.08
2.02	Casat et al.	1994	100	Psych	55	78	10.1	10.8	8	0.5	11	9	-0.20
3.01	Dahl et al.	1992	48	Norm	41	41	15.3	14.8	8	1	27	34	0.33
3.02	Dahl et al.	1992	48	Norm	41	41	15.3	14.8	9	1	27	34	0.28
3.03	Dahl et al.	1992	48	Norm	41	41	15.3	14.8	10	1	27	34	0.12
3.04	Dahl et al.	1992	48	Norm	41	41	15.3	14.8	11	1	27	34	-0.16
3.05	Dahl et al.	1992	48	Norm	41	41	15.3	14.8	12	1	27	34	-0.14
3.06	Dahl et al.	1992	48	Norm	41	41	15.3	14.8	13	1	27	34	0.20
3.07	Dahl et al.	1992	48	Norm	41	41	15.3	14.8	14	1	27	34	0.05
3.08	Dahl et al.	1992	48	Norm	41	41	15.3	14.8	15	1	27	34	-0.08
3.09	Dahl et al.	1992	48	Norm	41	41	15.3	14.8	16	1	27	34	-0.07
3.10	Dahl et al.	1992	48	Norm	41	41	15.3	14.8	17	1	27	34	-0.24
3.11	Dahl et al.	1992	48	Norm	41	41	15.3	14.8	18	1	27	34	-0.05
3.12	Dahl et al.	1992	48	Norm	41	41	15.3	14.8	19	1	27	34	-0.12
3.13	Dahl et al.	1992	48	Norm	41	41	15.3	14.8	20	1	27	34	0.00
3.14	Dahl et al.	1992	48	Norm	41	41	15.3	14.8	21	1	27	34	-0.22
3.15	Dahl et al.	1992	48	Norm	41	41	15.3	14.8	22	1	27	34	0.16
3.16	Dahl et al.	1992	48	Norm	41	41	15.3	14.8	23	1	27	34	0.09
4.01	Doherty et al.	1986	100	Psych	57	57	11.5	11.5	8,4.11	1	26	64	0.67
5.01	Extein et al.	1982	100	Psych	40	50	15	16	24	1	15	12	0.97
5.02	Extein et al.	1982	100	Psych	40	50	15	16	12	1	15	12	2.39
5.03	Extein et al.	1982	100	Psych	40	50	15	16	8	1	15	12	2.50
5.04	Extein et al.	1982	100	Psych	40	50	15	16	16	1	15	12	3.27
6.01	Fristad et al.	1988	100	Norm	70	29	9.7	9.3	16	0.5	63	21	0.72
6.02	Fristad et al.	1988	100	Norm	70	29	9.7	9.3	8	0.5	63	21	0.78
6.03	Fristad et al.	1988	100	Psych	70	71	9.7	9.2	8	0.5	63	14	0.81
6.04	Fristad et al.	1988	100	Psych	70	71	9.7	9.2	16	0.5	63	14	0.84
7.01	Ha et al.	1984	100	Psych	75	75	15.38	14.75	23	1	8	4	-0.04
7.02	Ha et al.	1984	100	Psych	75	75	15.38	14.75	16	1	8	4	0.23
8.01	Hsu et al.	1983	100	Psych	51	51	15	15	16	1	14	66	0.77
9.01	Livingston et al.	1984	100	Psych	50	73	9	10.18	16	0.5	4	11	0.23
10.01	Naylor et al.	1990	100	Psych	68	73	11.12	9.68	4	1	25	48	0.44
11.01	Petty et al.	1984	100	Psych	77	77	9.35	10.58	16	0.5	13	6	0.62
11.02	Petty et al.	1984	100	Psych	77	77	9.35	10.58	23	0.5	13	6	0.67
12.01	Pfeffer et al.	1989	100	Psych	71	71	10.5	10.5	8	1	18	27	0.08
12.02	Pfeffer et al.	1989	100	Psych	71	71	10.5	10.5	23	1	12	26	0.27
12.03	Pfeffer et al.	1989	100	Psych	71	71	10.5	10.5	16	1	19	26	0.28
12.04	Pfeffer et al.	1989	100	Psych	71	71	10.5	10.5	16	0.5	16	31	0.30
12.05	Pfeffer et al.	1989	100	Psych	71	71	10.5	10.5	8	0.5	17	31	0.52
12.06	Pfeffer et al.	1989	100	Psych	71	71	10.5	10.5	23	0.5	14	25	0.71
13.01	Poznanski et al.	1982	0	Psych	55	55	10.2	9	16	0.5	9	9	1.00
14.01	Robbins et al.	1983	100	Psych	50	50	15.5	15.5		0.5	23	9	0.51
15.01	Steingard et al.	1990	0	Psych	94	94	11.2	11.2	16	0.1	56	32	0.35
16.01	Targun & Capodano	1983	100	Psych	51	51	15.05	15.05		1	17	103	0.58
17.01	Young et al.	2006	0	Norm	46	47	10	10.3	8	1	11	32	1.07
17.02	Young et al.	2006	0	Norm	46	47	10	10.3	16	0.5	11	32	1.19

Figure 1.

Forrest plot of standardized effect size and 95% confident intervals for comparative studies examining response to DST between depressed and non-depressed children and adolescents.

Effect Size Predictors

We used mixed-modeling meta-regression techniques to examine possible predictors of effect sizes. Table 2 presents the regression coefficients of all predictors tested. None of the factors examined significantly predicted effect sizes in DST response.

Table 2.

Summary of regression estimates for predictors of mean effect size differences in cortisol response to the DST among depressed youth and comparison groups.

	Estimate	SEM	Z
Intercept	0.5731	0.1372	4.18**
Level 1 Variable
Time (Vs. Afternoon)
Daily Average	0.0403	0.3725	0.11
Morning	0.0859	0.1076	0.80
Evening	-0.0726	0.1186	-0.61
Level 2 Variables
Non Psych Control	-0.1026	0.3094	-0.33
MDD Group % Inpatient	0.0005	0.0032	0.17
Mean Age (Years)	0.0181	0.0559	0.32
Mean % Male	-0.0135	0.0097	-1.39
Dosage DST (0.5 mg)	0.1058	0.1835	0.58

Publication Bias

No missing studies in the direction of the null hypothesis were detected via the trim and fill method (Duval & Tweedie, 2000; Cluster Lo = -0.40 after 1 iteration). Figure 2 presents the funnel plot of cluster effect sizes by the inverse mean study sample standard error.

Figure 2.

Inverse standard error vs. effect size in DST cortisol studies. Connected dots represent comparisons in the same sample/cluster. Circles represent average effect size and pooled precision of each cluster.

Basal HPA Axis Functioning Studies: Meta Analysis

Main Effects

We identified 17 studies comparing MDD and non-MDD controls on basal (not stress induced) cortisol levels at different times during the day (27 total samples with N = 1,332; see Table 3). The studies contributed a total of 28 effect sizes. Using a random effects model, the pooled effect size (k = 36) for group differences in basal cortisol levels was .20 (z = 4.53, p < .01; 95% CI .11 – .29) indicating a tendency for depressed children and adolescents to have higher basal cortisol levels than non-MDD controls. The estimate of heterogeneity was not significant (.0000), z = .00, p > .20, suggesting limited between-study variability. Figure 3 presents the effect sizes and 95% confidence intervals for the studies included in this analysis.

Table 3.

Study characteristics and mean effect size differences in basal cortisol among depressed youth and comparison groups.

ID	Author	Year	% Inpatient	Control Type	Sex MDD (%males)	Sex-Cont (%males)	Age -MDD	Age-Cont	Time* (hour)	N-MDD	N-Cont	G
1.01	Birmaher et al.	1996	39	Norm	71	50	10.4	10.1	17	28	14	0.47
2.01	Casat et al.	1994	100	Psych	55	78	10.1	10.8	8	11	9	-0.90
3.01	Coplan et al.	2002	0	Norm	53	50	13.43	13.8	UK	22	36	0.47
3.02	Coplan et al.	2002	0	Psych	53	53	13.43	13.4	UK	22	40	0.07
4.01	Dahl et al.	1989	8	Norm	46	65	14.8	15.7	M24h	48	40	0.32
5.01	Dahl et al.	1991	48	Norm	37	41	15.2	14.9	M24h	32	27	0.39
6.01	Doherty et al.	1986	100	Psych	57	57	11.5	11.5	8	26	64	1.38
6.02	Doherty et al.	1986	100	Psych	57	57	11.5	11.5	23	26	64	-0.35
7.01	Extein et al.	1982	100	Psych	40	50	15	16	8	15	12	0.01
7.02	Extein et al.	1982	100	Psych	40	50	15	16	16	15	12	0.12
7.03	Extein et al.	1982	100	Psych	40	50	15	16	24	15	12	-0.27
8.01	Feder et al.	2004	15	Norm	59	59	9.3	9.2	M24h	72	15	0.05
8.02	Feder et al.	2004	15	Psych	59	59	9.3	8.7	M24h	72	31	0.23
9.01	Forbes et al.	2006	0	Norm	37	56	13.98	13.4	17	40	32	0.25
9.02	Forbes et al.	2006	0	Psych	37	64	13.98	13.4	17	40	14	0.00
9.03	Forbes et al.	2006	0	Norm	68	61	10.49	10.4	17	76	44	0.24
9.04	Forbes et al.	2006	0	Psych	68	67	10.49	10.5	17	76	18	0.06
10.01	Goodyer et al.	1996	0	Norm	42	43	12.98	13.08	8	82	40	-0.24
10.02	Goodyer et al.	1996	0	Norm	42	43	12.98	13.08	12	82	40	0.23
10.03	Goodyer et al.	1996	0	Norm	42	43	12.98	13.08	20	82	40	0.38
10.04	Goodyer et al.	1996	0	Psych	42	27	12.98	12.09	8	82	11	.24
10.05	Goodyer et al.	1996	0	Psych	42	27	12.98	12.09	12	82	11	.36
10.06	Goodyer et al.	1996	0	Psych	42	27	12.98	12.09	20	82	11	.49
11.01	Kaufman et al.	1997	50	Norm	46	54	9.75	9.5	17	26	13	-0.08
12.01	Luby et al.	2003	0	Norm	45	41	4.76	4.7	11	40	44	0.02
12.02	Luby et al.	2003	0	Norm	45	41	4.76	4.7	20	40	44	0.24
12.03	Luby et al.	2003	0	Psych	45	54	4.76	4.5	11	40	33	0.02
12.04	Luby et al.	2003	0	Psych	45	54	4.76	4.5	20	40	33	-0.03
13.01	Naylor et al.	1990	100	Psych	68	73	11.12	9.7	11	25	48	0.35
14.01	Pfeffer et al.	1989	100	Psych	71	71	10.5	10.5	8	17	27	0.06
14.02	Pfeffer et al.	1989	100	Psych	71	71	10.5	10.5	16	17	29	0.21
14.03	Pfeffer et al.	1989	100	Psych	71	71	10.5	10.5	23	17	26	0.22
15.02	Puig-Antich et al.	1989	0	Norm	64	38	9.5	9.1	M24h	45	8	0.10
15.01	Puig-Antich et al.	1989	0	Psych	64	60	9.5	8.7	M24h	45	20	0.15
16.01	Rao et al.	2008	0	Norm	48	43	15	15.1	20	30	25	0.49
17.01	Rao & Poland.	2008	0	Norm	44	38	15.8	15.6	M12p	16	16	0.60

Figure 3.

Forrest plot of standardized effect size and 95% confident intervals for comparative studies examining basal cortisol levels between depressed and non-depressed children and adolescents.

Effect Size Predictors

We used mixed-modeling meta-regression techniques to examine possible predictors of effect sizes. Table 4 presents the regression coefficients for these predictors. None of the factors examined predicted average effect size differences between the MDD and control groups.

Table 4.

Summary of regression estimates for predictors of mean effect size differences in basal cortisol among depressed youth and comparison groups.

	Estimate	SEM	Z
Intercept	0.1990	0.0439	4.53**
Level 1 Variable
Time (Vs. Afternoon)
Daily Average	0.0986	0.1971	0.50
Morning	0.0502	0.1945	0.26
Evening	-0.0353	0.1998	-0.18
Level 2 Variables
Non Psych Control	0.0499	0.0879	0.57
MDD Group % Inpatient	0.0004	0.0011	0.35
Mean Age (Years)	0.0190	0.0134	1.41
Mean % Male	0.0003	0.0042	0.08

^**p < .01

Publication Bias

No missing studies in the direction of the null hypothesis were detected via the trim and fill method (Duval & Tweedie, 2000; Cluster L₀ = -1.0 after 1 iteration). Figure 4 presents the funnel plot of cluster effect sizes by the inverse mean study sample standard error.

Figure 4.

Inverse standard error vs. effect size in basal cortisol studies. Connected dots represent comparisons in the same sample/cluster. Circles represent average effect size and pooled precision of each cluster.

Additional Descriptive Analysis of HPA-Axis Probes

CRH Infusion Studies

Four studies compared cortisol levels post CRH infusion, but none found a significant difference in cortisol or ACTH secretion between MDD and non-MDD groups (Birmaher et al., 1996; Dorn et al., 1996; Kaufman et al., 1997; Ronsaville et al., 2006). Effect sizes could be calculated from only one study (Kaufman et al., 1997), which provided data on differences in total cortisol production, peak cortisol level, and pre-to-post test change after administration of CRH in MDD and non-MDD controls. The effect sizes were -.13, -.14, and -.11, respectively.

Psychological (or naturally occurring) Challenge Studies

Surprisingly, we found only three relevant studies of HPA-axis reactivity to psychological stressors in children or adolescents with depression. Effect sizes could be calculated only for Luby et al. (2003) and Rao et al. (2008). For Luby et al. (2003), the mean effect size differences between the MDD group and psychiatric and healthy controls at 30 minutes post-stress were .30 and .21, suggesting higher post stress cortisol in the depressed group. However, Luby et al. (2003) reported blunted reactivity (percentage change) in the depressed group, likely due to already elevated baseline values. Luby et al. (2004) also compared cortisol responses to a laboratory stressor in young children and found reduced percentage change in the MDD group when compared to healthy controls. Furthermore, Rao et al (2008) compared the cortisol response of depressed adolescents and a non-psychiatric control group after a social stress task and found mean effect size differences of .66 in peak cortisol levels, and of .91 to .96 during recovery.

Discussion

Results of our meta-analyses support an association between HPA-axis dysregulation and pediatric depression. Specifically, as compared to control peers, depressed youth tend to have a dysregulated response to the dexamethasone suppression test and moderately higher cortisol levels throughout the day. However, we found no evidence of a dysregulated response to corticotropin releasing hormone in depressed pediatric samples. Furthermore, data on HPA-axis response to psychological stressors in this clinical population were noticeably scant. The three available studies suggest that, compared to non-depressed peers, depressed preschoolers show blunted reactivity (percentage change) but higher peak cortisol levels, while depressed adolescents show higher reactivity and delayed recovery.

Depressed youth's dysregulated HPA-axis response to the DST suggests possible alterations in the mechanism in charge of regulating tonic cortisol levels. Hypercortisolemia has been consistently found in depressed adults (Christensen & Kessing, 2001; Gillespie & Nemeroff, 2005), and has been examined as a potential predictor of various outcomes (e.g., treatment response; Ribeiro et al., 1993) as well as one cause of the brain structural anomalies observed in depressed patients (Videbech & Ravnkilde, 2004). Our results indicate that such HPA-axis dysregulation can also be found in children and adolescents, a conclusion that is in line with, and extends previous reviews by Casat et al. (1989) and Kaufman et al. (2001). However, contrary to previous descriptive reviews, which proposed that age influences HPA axis response in depressed youngsters (Birmaher & Heydl, 2001), our analysis found no significant age differences. Likewise, we found no significant effects of time of sampling and type of control group (psychiatric vs. healthy peers).

Consistent with the DST results, our meta-analysis suggests that depressed youth have higher cortisol levels throughout the day than do non-depressed youth, also reflecting hypercortisolaemia. However, group differences in baseline cortisol did not appear to be a function of sex, age, type of control group, or time of the sample collection. The lack of timing effects are particularly remarkable, as this indicates that the dysregulation is not specific to any time point, as would be expected if the dysregulation was driven by anomalies within circadian systems. Group differences in baseline cortisol also appear to be markedly less robust than differences in DST responses, suggesting that dysregulated responses to the DST do not necessarily translate to tonic HPA-axis dysregulation. Other investigators have also observed this phenomenon. For example, in their meta-analysis Ribeiro et al. (1993) found that while responses to the DST reliably predicted specific outcomes, pre-DST baseline measures were not associated with such outcomes. Therefore, the DST appears to provide a more sensitive index of hypercortisolaemia in pediatric depression than baseline cortisol measures, extending the extant adult literature.

We also examined studies using probes to activate the HPA-axis in depressed youngsters. The 4 published studies that used infusion of corticotropin releasing hormone failed to find a difference in cortisol response between the depressed and control groups (Birmaher et al., 1996; Dorn et al., 1996; Kaufman et al., 1997; Ronsaville et al., 2006). Effect sizes were extremely weak and in the opposite direction of what was expected. This normative cortisol response to CRH infusion among depressed youngsters suggests intact hypothalamic sensitivity to CRH and/or intact adrenal sensitivity to ACTH.

Finally, the three studies of HPA-axis reactivity to psychological stressors (Luby et al., 2003, 2004; Rao et al., 2008) reported equivocal results. Depressed preschool children had blunted activation (reduced base to peak change) compared to non-depressed peers, but mean effect size differences in peak cortisol suggested higher post-stress levels in the depressed groups. In contrast, depressed adolescents showed higher post-stress and recovery cortisol levels when compared to non-depressed peers. These contradictory findings are possibly due to group differences in baseline cortisol levels prior to the stress tasks, with the depressed preschool groups showing higher cortisol levels across the entire protocol (whether because of elevated tonic levels or chronic stress response to the entire laboratory experience), which could constrain the stress response to specific tasks. Additional research on the effects of psychological stressors in pediatric depression using clinical samples of various ages and more methodological controls is warranted.

All in all, our results point towards possible anomalies within HPA-axis tonic system as reflected by dysregulated response to the DST and elevated basal cortisol levels. Evidence for a dysregulated response to psychological stressors is less clear however, as extant research is extremely limited. Yet, the apparent normative response to CRH challenges suggests that if there is a HPA-axis dysregulation to psychological stressors (as found in depressed adults; see Burke et al., 2005), such dysregulation is likely due to factors that affect the production of CRH by the hypothalamus, such as cognitive factors (e.g., rumination, attention) that could trigger hypothalamic activation, or endogenous differences in hypothalamic sensitivity to stress signals.

Although our findings are consistent with descriptive reviews of the adult literature (see Holsboer, 1995), the lack of relevant published meta-analyses of studies of adults precludes direct comparisons of the strength of the HPA-axis-depression association in these two age groups. While some meta-analyses of the DST in adult depression exist (see for example, Mann et al., 2006; Nelson & Davis, 1997; Ribeiro et al., 1993), none of them focused on the DST responses or hypercortisolaemia of depressed vs. non-depressed adults. The only relevant empirical review was conducted by Burke et al. (2005), who examined cortisol responses to psychological stressors in depressed vs. non-depressed adults. In their meta-analysis, the mean cortisol difference between the depressed and non-depressed groups was .27 in stress response and 1.39 during recovery. This is consistent with our findings on pediatric samples using psychological stressors that show a modest group difference in reactivity but much stronger difference during recovery.

The results of our meta-analysis should be considered in the context of various limitations. First, although we examined the effects of the type of comparison group (psychiatric vs. normative), we could not control for diagnostic differences within the psychiatric samples. Unfortunately, most of the psychiatric samples included participants with various diagnoses resulting in diagnostic heterogeneity. Second, although we only included comparisons between participants with documented MDD against controls, we could not control for comorbidity of other disorders within the MDD groups. It is imperative that future studies further examine the effects of comorbidity on the association between HPA-axis dysregulation and pediatric depression. Likewise, due to limited variance between the studies, we were unable to examine factors that had been proposed as possible moderators of the HPA-depression relation, such as depression severity, endogenous subtype, and atypicality (eg., Dahl et al., 1992; Nelson & Davis, 1997), and exposure to traumatic events (e.g., Kaufman et al., 1997; Rao et al., 2008). Finally, due to our strict selection criteria, a number of informative studies were omitted from the meta-analysis (see for example Goodyer et al., 1998, 2003, 2000a, 2000b, 2001, 1991; Halligan et al., 2007; Herbert et al., 1996; Rao et al., 1999, 1996, 1997). Yet, the general findings of the omitted studies are consistent with our results, suggesting HPA-axis dysregulation in pediatric depression.

Our results suggest at least two directions for future work. First, there is a great need for longitudinal research with pediatric populations at familial-risk for depression, who have not yet developed the disorder. Studying such groups will help establish whether HPA-axis dysregulation is a risk factor for eventual depression (potential cause) or a reflection of depressive states (potential symptom). For example, while Rao et al. (1997) reported significant reductions in basal cortisol after remission from an MDD episode, indicating that HPA-Axis dysregulation may be accentuated during depressive states, there is also evidence that HPA dysregulation is already present prior to the onset of depressive symptoms (Halligan et al., 2007) or MDD (Goodyer et al., 2000b). However, most longitudinal studies were conducted with children who were “at-risk” due to the presence of subclinical signs of depression, raising the possibility that the observed cortisol differences prior to MDD onset may have reflected the disease process already underway (see for example Goodyer et al., 2003, 2000a, 2000b). Future studies with asymptomatic children, at familial-risk for depression, may help to clarify the potential impact of HPA-dysregulation in the development of this condition.

Finally, more studies are needed of HPA-axis reactivity to psychological stress in pediatric depression. Such studies should examine the mechanisms by which psychological stressors elicit a stronger cortisol response and/or a more delay regulation phase in depressed individuals when compared to non-depressed peers. This is of great importance given that HPA-axis dysregulation is not unique to depression and is observed in many other disorders, such as post-traumatic stress disorder (de Kloet et al., 2006) and social anxiety (Beaton et al., 2006). Thus, mechanistic research with diverse clinical populations may help us understand how HPA-axis dysregulation may differentially affect the development or course of different psychiatric conditions. For example, is dysregulated cortisol response to stressors observed in depression the result of oversensitivity of the HPA-axis system to stress signals, or to neurocognitive factors prior to the activation of the system? The impact of rumination, attention and other aspects of information processing on HPA-axis functioning in this population should be a specific focus of study.