Morning cortisol as a prospective predictor of MDD

Among a sample of 116 adult women (42% of whom had experienced prior episodes of depression), Harris et al. (Reference Harris, Borsanyi, Messari, Stanford, Brown and Cleary2000) examined the impact of psychosocial risk factors as well as cortisol levels (measured at 8 a.m. and 8 p.m.) on the onset of depressive episodes. Higher cortisol levels at 8 a.m. predicted more onsets of MDD over the next 13 months.Footnote ² This effect was significant independent of the impact of provoking life events and psychosocial vulnerability factors (negative self esteem, chronic negative social interactions, or chronic subclinical depression or anxiety) measured at study entry. It was also not attributable to the residual or scar effects of prior episodes on morning cortisol levels because effects were similarly strong among individuals with and without prior episodes of disorders. Although significant, the predictive power of morning cortisol was relatively weak: an odds ratio [OR] of 1.3, compared to 7.8 for the occurrence of intervening provoking life events or 3.8 for baseline psychosocial vulnerabilities when all variables were entered simultaneously.

A similar finding of high morning cortisol predicting onset of depression was found in a study of 180 adolescents at risk for depression because of either psychosocial factors (such as experiencing two or more recent life events or a history of two or more exit events/separations), high emotionality, or the presence of parental psychopathology (Goodyer, Herbert, Tamplin, & Altham, Reference Goodyer, Herbert and Tamplin2000). Cortisol was measured at 8 a.m. and 8 p.m. over a period of 4 days at study entry, along with a range of psychosocial risk factors, and the occurrence of life events and depressive episodes over the next 12 months was assessed. Adolescents who had one or more elevated morning cortisol peaks at baseline (termed “peak positive,” having one or more morning cortisol samples higher than the 80th percentile of the morning values for that day, defined separately by gender), were more likely to have a depressive episode over the next 12 months. Unlike in the study by Harris et al. (Reference Harris, Borsanyi, Messari, Stanford, Brown and Cleary2000), average level of morning cortisol was not a significant predictor, but rather only the occurrence of peak cortisol levels, suggesting that the presence of extreme morning cortisol values is the factor that confers risk.Footnote ³ Goodyer, Herbert, and Tamplin (Reference Goodyer, Herbert and Tamplin2000) argue that these peak levels may have more potent impact on later functioning because they are more likely to occupy Type I (glucocorticoid) receptors, which are only occupied in the presence of very high cortisol levels. Cortisol peaks, recent life events (in particular, personal disappointments and losses in the past month), and baseline self-reported depressive symptoms each independently predicted the occurrence of a depressive episode in the 12 month follow-up period (with odds ratios of 7.1 for peak morning cortisol, 1.1 for depressive symptoms, 58.9 for disappointments, and 8.8 for losses). Because baseline morning cortisol peaks were not associated with baseline depressive symptoms, baseline emotionality, or the occurrence of recent life events, Goodyer et al. conclude that high morning peaks in cortisol are not of recent psychosocial origin, but rather either of genetic, or early psychosocial origin.Footnote ⁴

Evidence of the latter comes from a recent study by Halligan, Herbert, Goodyer, and Murry (Reference Halligan, Herbert, Goodyer and Murray2007), in which cortisol was measured in the morning (8 a.m.) and evening (8 p.m.) every day for 10 days in a sample of 13-year-olds who either had (N = 48) or had not (N = 39) been exposed to maternal postnatal depression (PND) measured at 6 weeks postpartum (a variety of other intervening risk factors were also assessed). Morning cortisol at age 13 was found to predict high levels of depressive symptomsFootnote ⁵ at age 16, controlling for symptom levels at 13 years and a variety of other possible confounds. In this study, three different measures of morning cortisol (average, maximum level, and variability in 8 a.m. cortisol across the 10 days) were each significant positive predictors of later depressive symptoms, but the maximum level of the 10 measurement days was the best predictor when prior depressive symptoms and covariates were included (OR = 1.37). Of even more interest, exposure to maternal postnatal depression predicted the morning cortisol measures (Halligan et al., Reference Halligan, Herbert, Goodyer and Murray2004), as well as age 16 depressive symptoms. Maximum morning cortisol was a significant mediator of the association between maternal PND and age 16 depressive symptoms (OR = 1.41 for morning cortisol controlling for all covariates and PND). Although we are unable to rule out the possibility of genetic contributions to both maternal PND and adolescent cortisol, this study suggests that maternal postnatal depression is an early psychosocial influence that may have important implications for morning cortisol levels and levels of adolescent depression.

What works in preventive interventions for adolescent depression?

To discuss the efficacy of various types of preventive interventions, some basic “Prevention 101” vocabulary is required. The standard definitions currently used in prevention research are derived from the Institute of Medicine's (IOM) 1994 report (Mrazek & Haggerty, Reference Mrazek and Haggerty1994), which is the output of the Committee on Prevention of Mental Disorders. Prevention refers to those interventions implemented prior to the first onset of a disorder (Muñoz, Mrazek, & Haggerty, Reference Muñoz, Mrazek and Haggerty1996). This explicit focus on “disorder” emphasizes the use of diagnostic outcomes. Notably, however, most prevention studies of depression in youth populations have examined symptom level outcomes, although there is ample justification for examining symptoms in addition to categorical outcomes. That is, depressive symptoms have demonstrated significant concurrent association with impairment (Lewinsohn, Solomon, Seeley, & Zeiss, Reference Lewinsohn, Solomon, Seeley and Zeiss2000) and prospective associations with later full-blown depressive disorder (Pickles et al., Reference Pickles, Rowe, Simonoff, Foley, Rutter and Silberg2001). Another implication of the IOM's definition of prevention is that interventions that occur after the person crosses the threshold for disorder are considered either treatment or maintenance interventions. This distinction between initial onsets and recurrences is important when examining depression outcomes because different psychosocial (and psychobiological) processes may be implicated in first versus later episodes (Lewinsohn, Allen, & Seeley, Reference Lewinsohn, Allen, Seeley and Gottlib1999; Monroe & Harkness, Reference Monroe and Harkness2005).

Interventions are also categorized according to the type of population they target (Mrazek & Haggerty, Reference Mrazek and Haggerty1994). Universal interventions are administered to the general population without restriction. Selective interventions are targeted to individuals who are at heightened risk for a mental disorder. The selected risk factor may be based in biological, psychological, or social domains. Indicated interventions focus on subpopulations who are already manifesting some signs, symptoms, or biological markers of disorder, but whose psychopathology is below the diagnostic threshold.

HPA axis functioning has not, to our knowledge, been previously examined as a risk factor in depression prevention studies in adolescents or adults. It is not entirely clear whether focusing an intervention on participants with abnormalities on particular cortisol measures would be better classified as indicated or selective. Resolution of this debate requires knowing whether the HPA axis changes are part of the etiology of the disorder (indicated), or whether they simply allow us to better identify those individuals at risk for disorder (selective). For the present paper, this question is solved by using the language “targeted,” which refers to selective or indicated interventions, and underscores that the intervention is only given to a subset of the population.

Status of preventive interventions for depression in adolescents

Two recent meta-analyses have comprehensively examined the efficacy of depression prevention in youth populations (Horowitz & Garber, Reference Horowitz and Garber2006; Merry, McDowell, Hetrick, Bir, & Muller, Reference Merry, McDowell, Hetrick, Bir and Muller2004). One of these (Horowitz & Garber, Reference Horowitz and Garber2006) examined 30 studies that randomly assigned participants (under age 21) to intervention or control conditions. Focusing on symptom-level outcomes, the authors reported significantly larger effect sizes at postintervention for selective (d = .30) than for universal interventions (d = .12), with indicated interventions demonstrating intermediate effects (d = 23). At follow-ups ranging from 3 to 8 months, selective and indicated interventions (d = .34 and .31, respectively) were also significantly more effective than universal interventions (d = .02). Thus, across targeted interventions, effects were in the small to moderate range for symptom reduction, with targeted interventions (both selective and indicated) performing better than universal ones.

Far fewer studies have been conducted using diagnostic outcomes, five of which were reviewed by Merry et al. (Reference Merry, McDowell, Hetrick, Bir and Muller2004). Two of these studies were universal, showing significant effects posttreatment but no effects at 1-year follow-up. Three studies were targeted interventions with significant prevention effects on depressive disorders noted both at posttreatment (risk difference = −.13) and at 12-month follow-up (risk difference = −.12). Taking the reciprocal of those risk differences demonstrates that the cited interventions would need to be administered to approximately eight to nine adolescents to reduce the incidence of depression by one case (Merry et al., Reference Merry, McDowell, Hetrick, Bir and Muller2004). Studies of targeted interventions of youth (age range = 10–16 years) conducted since the Merry et al. meta-analysis have yielded a range of findings: (a) substantial reductions in onsets of major depression/dysthymia (Arnarson & Craighead, Reference Arnarson and Craighead2006); (b) significant reductions in anxiety, depressive, and adjustment disorders in participants with elevated depressive symptoms at baselineFootnote ⁷ (Gillham, Hamilton, Freres, Patton, & Gallop, Reference Gillham, Hamilton, Freres, and Gallop2006); and (c) the absence of significant differences in rates of depression between the intervention and comparison groups (Seligman, Schulman, & Tryon, Reference Seligman, Schulman and Tryon2007; Sheffield et al., Reference Sheffield, Spence, Rapee, Kowalenko, Wignall and Davis2006).

In sum, although there has been heterogeneity across studies, there are several examples of interventions (particularly targeted ones) that do decrease risk for subsequent depressive disorder. What are the characteristics of these successful interventions? Do they operate on processes that are known to influence HPA axis activity?

As shown in Figure 2, changes in exposure to objectively determined stressful life events, changes in available social support systems, and changes in individuals' interpretations of stressors or supports are all candidate mechanisms that could be influenced by such interventions and have implications for HPA axis functioning and risk for depression. In addition, changes in sleep, diet, exercise, medication use, or other health behaviors with biological consequences for the HPA axis, could also play a role (see Figure 2, left column). To what extent do effective preventive interventions target these HPA axis relevant processes?

Figure 2.

Access points (A–F) by which interventions may influence the HPA axis, and the possible roles (1–3) of individual differences in HPA axis activity in depression. [A color version of this figure can be viewed online at journals.cambridge.org/dpp]

Description of effective interventions for adolescent depression

It is useful to review the nature of the interventions with documented efficacy to examine their relevance for HPA axis functioning in youth populations. Although, as noted above, we believe that changes in subclinical symptoms are also important, in line with the IOM report, we focus here on studies examining clinical depressive disorder as an outcome. Some of the strongest prevention effects have been found in the work of Clarke and colleagues (Clarke et al., Reference Clarke, Hawkins, Murphy, Sheeber, Lewinsohn and Seeley1995, Reference Clarke, Hornbrook, Lynch, Polen, Gale and Beardslee2001). Across both studies, the Coping with Stress Course resulted in decreased incidence of depressive disorders through 1-year follow-up after completion of the intervention. One caveat to this important work is that many participants had a prior history of depression (see Sutton, Reference Sutton2007). This intervention focuses especially on cognitive restructuring techniques: identifying and challenging irrational or negative beliefs (Clarke et al., Reference Clarke, Hornbrook, Lynch, Polen, Gale and Beardslee2001). For example, the intervention paid particular attention to combating negative cognitions related to the adolescents' having a depressed parent, such as thinking that they were inevitably going to get depressed themselves. Although the intervention appeared to target dysfunctional cognitions, no mediational analyses have yet been reported, so the mechanisms of action are not entirely clear. Hypothetically (as HPA axis activity was not examined), potential influences on HPA axis activity for this intervention would occur at the level of appraisal of events as stressful or not (Figure 2, Point B).

Arnarson and Craighead (Reference Arnarson and Craighead2006) used a broader cognitive behavioral approach that included mood monitoring, pleasant events training, problem solving, and relaxation techniques, in addition to identifying and troubleshooting dysfunctional cognitions. Whereas the results by Clarke and colleagues contained a mix of first onset and recurrent depressive disorder, Arnarson and Craighead examined only first onsets of depression. The investigators found that their intervention resulted in fewer cases of major depression and dysthymia compared to the control condition through 1-year follow-up (Arnarson & Craighead, Reference Arnarson and Craighead2006). Neither mediators of treatment outcome nor component analyses, which focus on the efficacy of different aspects of the treatment protocol, were reported, so it is unclear which of the elements of this intervention were efficacious. This intervention addressed multiple factors relevant to HPA axis activity simultaneously, targeting levels of exposure to stressful events (Figure 2, Point A, via pleasant events training), availability and use of coping resources (Figure 2, Point B, via the problem solving component), appraisal or cognitive interpretation of potential stressors or supports (Figure 2, Point C, via the component addressing dysfunctional cognitions), and health behaviors (Figure 2, Point D, via the relaxation training component).

Another set of studies includes the Penn Resiliency Program (Gillham et al., Reference Gillham, Hamilton, Freres, and Gallop2006) and the APEX/Penn Resiliency Training for College Students program (Seligman et al., Reference Seligman, Schulman and Tryon2007; Seligman, Schulman, DeRubeis, & Hollon, Reference Seligman, Schulman, DeRubeis and Hollon1999). These have substantial overlap, although the programs differ in several ways including intervention structure, targeted population (youth vs. college age) and complexity of the cognitive skills taught (J. E. Gillham, personal communication, June 9, 2007). Both programs implement a cognitive–behavioral framework that includes the following skills: cognitive skills include identifying negative cognitions, examining the evidence for those thoughts, challenging negative thoughts, and developing alternative thoughts; behavioral skills, include training in assertiveness, social skills, problem solving training, and relaxation techniques. Gillham et al. (Reference Gillham, Hamilton, Freres, and Gallop2006) did not examine potential mediators for the effect of the intervention in preventing cases of depressive, anxiety, and adjustment disorder. Seligman et al. (Reference Seligman, Schulman, DeRubeis and Hollon1999) reported that changes in attributional style, hopelessness, and dysfunctional attitudes from preintervention to postintervention mediated the effect of the intervention on depressive symptoms averaged across postintervention and follow-up (see also Seligman et al., Reference Seligman, Schulman and Tryon2007). However, mediational effects for the reduction in depressive episodes over 3-year follow-up were not reported.

Neither of these research programs has conducted component analyses for their broad-based interventions (Sutton, Reference Sutton2007). However, an earlier investigation of the Penn Prevention Program found no substantial differences at postintervention for groups receiving cognitive tools (e.g., challenging negative thoughts) alone, social problem-solving skills alone, or a combined treatment involving both (Jaycox, Reivich, Gillham, & Seligman, Reference Jaycox, Reivich, Gillham and Seligman1994). Thus, this intervention operates on multiple HPA axis-relevant processes including altering appraisals of events and resources (Figure 2, Point C), altering the availability and efficacy of social support by way of improved social problem solving (Figure 2, Point B), and altering stress levels by the use of relaxation techniques (an aspect of Figure 2, Point D).

Finally, a preliminary investigation by Young Mufson, and Davies (Reference Gillham, Hamilton, Freres, and Gallop2006) suggests that interpersonal psychotherapy adolescent skills training (IPT-AST) may also serve as an effective preventive program for adolescents. IPT-AST is a modified version of group IPT for adolescents (Mufson, Gallagher, Dorta, & Young, Reference Mufson, Gallagher, Dorta and Young2004). This intervention focuses on three potential problem areas: role disputes, role transitions, and interpersonal deficits. It also includes psychoeducation about depression, exploration of the relationship between feelings and interpersonal interactions, and communication and interpersonal skill building. The interpersonal skill building component teaches several techniques such as: analyzing difficult interpersonal situations, examining how those situations could have gone differently, expressing emotions constructively, choosing an appropriate time to engage in significant discussions, and negotiating a compromise (Mufson et al., Reference Mufson, Gallagher, Dorta and Young2004). Young et al. (Reference Young, Mufson and Davies2006) reported that the differences in rates of depressive disorders (major depression, dysthymia, depressive disorder not otherwise specified) approached statistical significance with lower incidence in adolescents who received IPT-AST versus referral to the school counselor. This study had a relatively small sample size (N = 41), and no mediators were explored. However, these results tentatively suggest that a focus on interpersonal relationships may also be viable in prevention programs for adolescent depression, and could plausibly influence the HPA axis by way of reducing stressful interpersonal events (Figure 2, Point A), and improving the quantity and quality of social support and coping resources (Figure 2, Point B).

This section has attempted to illustrate several key points. First, multiple interventions, largely focusing on cognitive–behavioral skills, have been effective in preventing depressive disorders in adolescents (see Sutton, Reference Sutton2007, for a more detailed review). Second, there is some indication that cognitive changes may mediate the effects of the cognitive–behavioral interventions (Seligman et al., Reference Seligman, Schulman and Tryon2007), although none of the studies has directly examined that possibility using diagnostic outcomes. Third, initial results suggest that targeting interpersonal functioning may also have preventive effects. Fourth, although several of the interventions inculcate a range of tools, the relative absence of component studies prohibits elucidation of which aspects of the interventions might be particularly effective in preventing depression. Fifth and finally, in contrast to those with positive effects, it must be noted that some studies (Seligman et al., Reference Seligman, Schulman and Tryon2007; Sheffield et al., Reference Sheffield, Spence, Rapee, Kowalenko, Wignall and Davis2006) focusing on cognitions and social problem solving have not found significant preventive effects on depressive disorder, so conclusions about what works must be appropriately nuanced (Sutton, Reference Sutton2007).

These interventions touched upon all the elements (stress exposure, availability of social support resources, interpretation of events, and stress-reducing health behaviors) that we had a priori suggested might be important pathways by which HPA axis activity could be modified in an intervention (although among the possible health behaviors relevant to depression, only relaxation was addressed).Footnote ⁸ Thus, it seems reasonable that changes in the HPA axis could be a candidate pathway by which these successful interventions exerted at least some of their effects. Rather than simply speculating on what might work to influence HPA axis activity, however, we next consider what evidence exists that HPA axis functioning can indeed be altered in the context of a random assignment psychosocial (rather than pharmacological) intervention.

Can HPA axis parameters change as a result of psyschosocial intervention?

To get a better handle on whether “risky” cortisol parameters can be altered by systematic attempts to change psychosocial or cognitive environments among those not already depressed, we turn now to the literature on randomized case-control experiments designed to alter HPA axis activity. There are relatively few of these efforts available, and studies are particularly lacking during the adolescent age period. As a result, we include data from interventions with young children (briefly) and then focus on adult interventions that more closely resemble the types of preventions described above that are typically implemented in attempts to reduce depression in adolescents (currently without the benefit of HPA axis information).

Given accumulating evidence of prenatal and early postnatal influences on the HPA axis, and the contribution of these early HPA axis alterations to increased vulnerability to disorder in adolescence (Halligan et al., Reference Halligan, Herbert, Goodyer and Murray2007), one possibility would be to focus preventive interventions on altering maternal stress experiences during the prenatal period, or maternal depression and behavior in the early postnatal period. Such attempts are a valuable approach with potentially great pay off in the long term, but are not helpful to clinicians who hope to prevent disorder among those who have already reached late childhood and early adolescence. For such individuals the combination of early experience and genetic predisposition may have already created either a risky or robust HPA. Rather, we focus our thoughts on how to supplement existing interventions designed to prevent adolescent depression, with the intention of improving our ability to intervene more immediately to prevent depression in at-risk children and youth.

No studies that we are aware of have tracked changes in glucorticoids over time during the course of a random assignment preventive intervention with adolescents. Two studies, however, have demonstrated that diurnal cortisol profiles in infants, toddlers, and preschoolers are sensitive to a psychosocial intervention involving modifications of caregiver behavior. In each case, the population was children in foster care, and random assignment was made to either a regular foster care setting or an experimental intervention foster care setting (and a regular nonfoster care control group was also available but not part of the random assignment process).

In Dozier and colleagues' (Reference Dozier, Peloso, Lindhiem, Gordon, Manni and Sepulveda2006) study, infants and toddlers who received the Attachment and Biobehavioral Catchup experimental intervention (which involved training caregivers in sensitive responding to infant and toddler signals in 10 weekly sessions) had wake-up to bedtime diurnal cortisol profiles 1 month later that were similar to the nonfoster care control group of children. Those in the regular foster care group also received an educational control intervention (focused on cognitive and language skills) conducted with the same frequency and intensity as the experimental intervention; this group had cortisol profiles that were significantly elevated (at both wakeup and bedtime) over the other two groups.

In the study by Fisher and colleagues, changes in cortisol levels were tracked over time among preschoolers involved in either (a) a foster care experimental intervention group in which consistent and supportive caregiving was taught, (b) a regular foster care group, and (c) a nonfoster care community control group (Fisher, Gunnar, Dozier, Bruce, & Pears, Reference Fisher, Gunnar, Dozier, Bruce and Pears2006; Fisher, Stoolmiller, Gunnar, & Burraston, in press). In this study, the diurnal cortisol patterns of the children in the foster care intervention group remained similar to those of the control children, whereas those in the regular foster group had drops in their morning cortisol levels with increasing time in foster care, resulting in a flattening of their diurnal rhythms. In addition, children in the foster care intervention group who began with low morning cortisol and flattened rhythms had increases in morning cortisol over time, resulting in stronger rhythms.

Both of these studies with infants and toddlers demonstrate that diurnal cortisol profiles appear to be responsive to psychosocial intervention, but have yet to demonstrate whether or not the changes in diurnal cortisol are long lasting. They also have not yet demonstrated whether or not these alterations result in changes in immediate behavioral symptoms or in long-term risk for the development of emotional disorders.Footnote ⁹ In addition, the fact that diurnal cortisol profiles are responsive to psychosocial interventions in infants and young children does not imply that adolescent cortisol profiles are equally malleable. Obviously, the types of interventions implemented in adolescent populations will be different from those employed with younger children, because adolescents have greater responsibility for their own biobehavioral regulation, and because of cognitive development, individual differences in appraisal may become important targets for intervention.

Several studies have focused on the effects of immediate manipulations of mood or social support on salivary cortisol reactivity examined within the context of the same testing session, showing that improvements in mood or the presence of a supportive figure reduced acute cortisol reactions to laboratory-based stressors (Heinrichs, Baumgartner, Kirschbaum, & Ehlert, Reference Heinrichs, Baumgartner, Kirschbaum and Ehlert2003; Kirschbaum, Klauer, Filipp, & Hellhammer, Reference Kirschbaum, Klauer, Filipp and Hellhammer1995). These are of less interest for our purposes than studies in which the interventions were implemented well prior to the stress exposure, such that we know that the manipulation has efficacy beyond the intervention setting itself. Of even more interest are studies in which interventions are conducted and changes in HPA axis activity observed in naturalistic settings. We focus here on the latter two types of studies.

Past theory and research has suggested that HPA axis responding is dependent not only the nature of the stressor encountered, but also on appraisal processes (Herman & Cullinan, Reference Herman and Cullinan1997; Lazarus & Folkman, Reference Lazarus and Folkman1984). Relevant appraisal processes include the individual's interpretation of the stressor, as well as their assessment of their available coping resources (Figure 2, Point C). As a result, researchers have suggested that alteration of cognitive processes related to threat and stressor appraisal, such as those often employed in cognitive–behavioral therapy, may be an effective way to lessen the frequency or severity of HPA axis reactions to everyday events. Others have suggested that interventions aimed at more directly reducing the early physiological manifestations of stress, such as progressive muscle relaxation, might be helpful in reducing HPA axis responses. Several studies have employed randomized case-control designs to examine whether these approaches alter HPA axis activity. Most have focused on responses to a standardized laboratory stressor (the Trier Social Stress Test [TSST]); a few have attempted to influence cortisol responses to real-life stressors or to alter aspects of basal cortisol profiles.

A series of studies has examined the effects of cognitive–behavioral stress management intervention (CBSM) on stress reactivity, comparing individuals randomly assigned to this protocol with control participants (who received CBSM after the posttest; Gaab et al., Reference Gaab, Blattler, Menzi, Pabst, Stoyer and Ehlert2003; Gaab, Sonderegger, Scherrer, & Ehlert, Reference Gaab, Sonderegger, Scherrer and Ehlert2006; Hammerfald et al., Reference Hammerfald, Eberle, Grau, Kinsperger, Zimmermann and Ehlert2006). This group-based intervention involved a combination of cognitive–behavioral techniques, including stress management (cognitive restructuring, problem solving, and self-instruction) and relaxation training (progressive muscle relaxation). In the Gaab et al. (Reference Gaab, Blattler, Menzi, Pabst, Stoyer and Ehlert2003) study, the group intervention was delivered in two sessions (from 10:00 a.m. to 5:00 p.m.) on a weekend, and the effects on the TSST were observed 2 weeks later. Hammerfald et al. (Reference Hammerfald, Eberle, Grau, Kinsperger, Zimmermann and Ehlert2006) used a similar scheduling and intensity of the CBSM training (a total of 10 hr delivered over two weekends), but the TSST was administered 4 months after CBSM treatment.

In both studies, participants in the CBSM training had lower cortisol responses to the TSST, with slightly stronger results when the TSST was examined closer in time to the intervention. These two studies assessed self-reported stress levels, and appraisal processes, including primary appraisal (including items assessing challenge and threat) and secondary appraisal (including self-concept of own competence and control expectancy). They found that it was changes in primary appraisals of stressors that served to partially mediate the intervention effect on TSST responses. This suggests that an intervention aimed at reducing threat and stress appraisals may be most important, at least for reducing acute physiological stress reactions. Although direct attempts to modify threat appraisals are likely to be important, other strategies, including improving perceived social support resources, may indirectly affect threat appraisals and thus also play an important role.

In negotiating stressors encountered in daily life, children and even adolescents are still highly dependent on external coping resources such as parents and friends, and thus interventions aimed more directly at supporting positive social relationships and social support resources may prove to be especially relevant in these younger populations. Future research should involve a component analysis to help identify which aspects of multifaceted interventions have the largest impact on HPA axis activity. It should also examine impacts of the intervention on basal cortisol functioning in addition to stress reactivity, and when examining stress reactivity, should examine changes in responses to naturalistic stressors in addition to responses to the TSST.

In one study that did examine the effect of an intervention on a naturalistic stressor, Gaab and colleagues (Reference Gaab, Sonderegger, Scherrer and Ehlert2006) examined the impact of the CBSM training on college students' reactions to an important exam. Participants in the intervention group had lower levels of anxiety in the weeks leading up to the exam. However, paradoxically and in opposition to the effects found for the acute lab-based stressor, they found that participants in the intervention group had a larger CAR on the days of testing than those in the control group. It is also of interest, however, that there was no association between perceived stress level and cortisol responses in the control group, but for the intervention group 27% of the variation in cortisol levels was accounted for by participants' perceived stress levels immediately before the exam. The authors argue that because of the chronic nature of the stressor (prolonged anticipation of the exam), the control participants' increased anxiety levels may have downregulated their HPA axis activity, such that it was no longer responding appropriately to the presence of challenge. It would have been interesting to know what the typical daily CAR values looked like for these two groups; in this study, the authors only examined the CAR on the day the students were encountering an extreme stressor. It is debatable what an optimal outcome is in terms of a CAR level. One would not want the CAR to be too high, as heightened CARs are associated with risk for MDD, but nor would one want it to be too low, as this increases risk for disorders related to hypocortisolism such as chronic fatigue, burnout, or rheumatoid arthritis.

One study that focused specifically on a burnout population used a cognitive–behavioral intervention in an attempt to increase the significantly lower CAR values found in individuals with clinical burnout (Mommersteeg, Keijsers, Heijnen, Verbraak, & van Doornen, Reference Mommersteeg, Keijsers, Heijnen, Verbraak and van Doornen2006). A combination of cognitive therapy and practical interventions to improve work-related functioning resulted in a significant increase in CAR levels in the burnout group, and significant reduction of symptom complaints. There was, however, no significant association between the degree of increase in CAR values and the degree of improvement in burnout-related symptoms.

In an attempt to resolve the question of what represents a “healthy” CAR, we have previously hypothesized that optimal CAR functioning is reflected in CAR values that are on average moderate in level, with some variability from day to day such that CAR levels are appropriately calibrated to the anticipated demands of each particular day (Adam et al., Reference Adam, Hawkley, Kudielka and Cacioppo2006). Under this “CAR calibration” theory, CAR values that are chronically high, chronically low, and those that do not modulate in response to changing daily demands would be considered maladaptive. In the Gaab et al. (Reference Gaab, Sonderegger, Scherrer and Ehlert2006) study, those who received the intervention had CAR values that were significantly associated with the perceived level of the stressor they were facing. There is, therefore, some sign that cognitive interventions appear to normalize CAR values to a moderate level and perhaps improve the degree to which CAR responses are calibrated to situational demands (as in the Gaab et al., Reference Gaab, Sonderegger, Scherrer and Ehlert2006, study). All told, this particular aspect of HPA axis activity appears to be an important target that is both related to psychological and physical functioning and modifiable by intervention. Nonetheless, other aspects of HPA axis functioning (such as diurnal cortisol slopes and evening cortisol levels and reactivity to momentary stressors) remain of interest and the effects of interventions on these parameters should also be evaluated.

To summarize the above studies, there is a clear indication that the reactivity of the HPA axis can be influenced in a randomized intervention and that the intervention effects occur at least partially by way of altered appraisal processes. There is much work to be done, however, in examining the impact of interventions on reactivity to real-life stressors, both anticipated (like the exam situation used in Gaab et al., Reference Gaab, Sonderegger, Scherrer and Ehlert2006) and unanticipated (perhaps using an experience sampling method approach to capture participants' cortisol responding to everyday events; Adam, Reference Adam2006; van Eck, Berkhof, Nicolson, & Sulon, Reference van Eck, Berkhof, Nicolson and Sulon1996), and in examining the modifiability of basal HPA axis functioning.

Given success of both cognitive–behavioral and social support strategies in altering aspects of HPA axis activity, it appears that there is some convergence between what works in terms of prevention of adolescent depression and the types of interventions that are effective (at least in adults) in altering HPA axis activity. Whether similar strategies would be effective in adolescents remains to be demonstrated. It seems plausible that changes in social support, and changes in threat appraisal/evaluation, might be relevant in modulating adolescent HPA axis activity, particularly given prior correlational evidence that the adolescent HPA axis is responsive to within-person variation in social experience and appraisal processes (Adam, Reference Adam2006).

Many questions remain, however, that can only be answered by the actual incorporation of HPA axis measures into a preventive intervention study for adolescent depression. These include (a) whether or not the HPA axis changes induced by interventions would be effective in reducing the onset of depression in youth, (b) whether premorbid HPA axis differences would predict which youth are most likely to respond to treatment, and (c) whether the degree of change in HPA axis parameters would serve to indicate which youth are responding most to the intervention. An outline of such a study, drawing on best practice knowledge regarding prevention of depression and our growing understanding of the role of the HPA axis in adolescent depression, is provided below.

Prototype for a HPA axis-focused prevention study

In designing an intervention that would benefit from and inform our understanding of the role of HPA axis activity in depression, our suggested approach would (a) capitalize on the design features of existing preventive interventions for adolescent depression that we know to be effective, (b) supplement those successful interventions with strategies that we know to influence HPA axis activity, and (c) include repeated measures of both HPA axis activity and depression, as well as repeated measures of the stress appraisal and social support processes that may mediate the effects of the intervention on the HPA axis and depression. Given that we know targeted interventions to be the most effective, it would be important to start with a high-risk sample, but it would be somewhat premature, at this point, to select that risk sample based on HPA axis characteristics. We would, however, want to measure HPA axis parameters at baseline among the selected sample to examine whether certain profiles of HPA axis functioning at baseline do indeed place certain individuals at higher risk, or whether certain HPA axis parameters provide insights into which adolescents are most responsive to particular intervention approaches. We would also want to continue to measure HPA axis parameters throughout the intervention and follow-up period to examine whether HPA axis changes occur in response to the intervention and are related to the efficacy of the intervention. Based on our review, the important features of a prevention program with the potential to modify risk for youth depression, and to reveal whether this occurs by way of changes in HPA axis activity include the following:

1. 1. Adequate dosage. The effective depression prevention protocols reviewed above were all conducted in small group format over the course of many weeks, for example, Clarke et al. (Reference Clarke, Hornbrook, Lynch, Polen, Gale and Beardslee2001) had 15 1-hr meetings, Arnarson and Craighead (Reference Arnarson and Craighead2006) conducted 14 90-min sessions, and Young et al. (Reference Young, Mufson and Davies2006) had eight 90-min sessions.Footnote ¹⁰ Thus, we would recommend approximately 15–20 hr of intervention time. Based on our review, we would suggest 15 small-group (6–10 adolescents) meetings lasting 60–90 min each.

2. 2. Substantial cognitive component. The research by Clarke et al. (Reference Clarke, Hawkins, Murphy, Sheeber, Lewinsohn and Seeley1995, Reference Clarke, Hornbrook, Lynch, Polen, Gale and Beardslee2001) has demonstrated that challenging negative cognitions can have a substantial effect on reducing risk for depressive disorder. In related work, Gaab and colleagues (Reference Gaab, Blattler, Menzi, Pabst, Stoyer and Ehlert2003, Reference Gaab, Sonderegger, Scherrer and Ehlert2006; also Hammerfeld et al., 2006) has found that primary cognitive appraisals of threat and challenge are partial mediators of the relationship between their CBSM intervention and cortisol reactivity to the TSST. We envision that six to seven sessions of the intervention protocol would focus on the development of cognitive skills, for example, identifying negative thoughts, examining evidence, and challenging negative thoughts.

3. 3. Interpersonal component. Several of the depression prevention programs reviewed above included some social skill-building component. The strongest emphasis on interpersonal relationships is found in the IPT-AST trial conducted by Young et al. (Reference Young, Mufson and Davies2006), and some modified version of that protocol would be implemented here. Additional rationale for inclusion of this component comes from findings that immediate cortisol levels are affected by the presence of social support (e.g., Heinrichs et al., Reference Heinrichs, Baumgartner, Kirschbaum and Ehlert2003). Thus, it seems plausible that improvement in one's social network via the inculcation of social skills (e.g., constructive expressive of feelings, improved listening, and negotiation of a compromise) might affect both depressive psychopathology and HPA response. We would designate five to six sessions for this module.

4. 4. Physical relaxation component. Several of the depression prevention programs reviewed above incorporated physical relaxation into the protocol. Although none of the programs has explicitly looked at the effects of relaxation alone, there may be good reason for its inclusion here. First, over 20 years ago Reynolds and Coates (Reference Reynolds and Coats1986) found that relaxation training was associated with significantly decreased depressive symptomatology versus wait-list control in a moderately depressed adolescent sample. Thus, there is some evidence that relaxation training (albeit at a higher dosage than will be implemented here) can directly affect symptomatology. Second, relaxation has been not only a component of several effective depression prevention protocols, but it is also part of the CBSM program used by Gaab and colleagues (e.g., Gaab et al., Reference Gaab, Blattler, Menzi, Pabst, Stoyer and Ehlert2003, Reference Gaab, Sonderegger, Scherrer and Ehlert2006), which has been associated with changes in cortisol reactivity. We would include two sessions focusing on these skills.

5. 5. Measurement of HPA axis activity at multiple points throughout the intervention process and across the follow-up period. Given the weight of evidence that morning cortisol (and probably specifically the size of the CAR) are the best prospective predictors of MDD in youth, and that elevated evening levels have been shown in at least some research to be a correlate of the presence of depressive symptoms/disorder in adolescents, the HPA axis measurement protocol utilized should capture each of these aspects of HPA axis functioning. At a minimum, a protocol gathering cortisol at waking, 30 – 40 min postawakening, and at bedtime across 2 days would be desired. This would allow measurement of wake-up cortisol levels, the size of the CAR, evening cortisol levels, the slope of the diurnal cortisol rhythm from wake-up to bedtime, and an estimate of the AUC of cortisol levels across the day, and some indication of whether the CAR changes in response to daily demands.Footnote ¹¹ This HPA axis protocol should be implemented at baseline, several times during the course of the intervention, immediately postintervention, and at follow-up 6 and 12 months postintervention.

   Although perhaps less easily put into practice in a large-scale study, several additional HPA axis measures would be desirable. First, a measure of reactivity to acute psychosocial challenge, such as the TSST, measured at baseline and postintervention would also be useful, beyond our basal measures, as a potential candidate marker of risk for depression. In addition, use of a DEX suppression test (or combined DEX suppression-CRH challenge test) at baseline and postintervention would be helpful in examining whether there is any sign of central HPA axis dysregulation at baseline or whether impairments in negative feedback regulation of the HPA axis emerge with the development of disorder.

6. 6. Attention to gender, age, and pubertal stage as possible mediators or moderators. Given the dramatic changes in prevalence rates of depression by age/pubertal stage and gender over adolescence, inclusion of a sufficient range of ages (or sufficiently length of longitudinal follow-up period) would be important to examine the role of age/pubertal stage and gender and their potential relation to HPA axis measures. For example, it would be important to examine whether the “risky” aspects of HPA axis function vary by gender and age or pubertal stage, and whether age and/or gender-related changes in these parameters help to account for age and/or gender-related changes in onsets of MDD.

7. 7. Assessment of early stress exposure as well as current life stress. We noted evidence that early exposure to trauma and/or to maternal postnatal depression may affect the organization of HPA axis function (Halligan et al., Reference Halligan, Herbert, Goodyer and Murray2007; Heim, Plotsky, & Nemeroff, Reference Heim, Plotsky and Nemeroff2004), and that early stress exposure may sensitize individuals to the effects of later life events on HPA axis function (Essex, Reference Essex, Klein, Cho and Kalin2002). Measurement of early stress exposure might help explain individual differences in HPA axis function upon entering the prevention program, and serve as a moderator helping to explain variations in responses to treatment. Reductions in current life stress, changes in appraisals of these stressors, and strengthening of social support resources may have particularly advantageous effects on youth whose HPA axes have been sensitized by prior stress exposure.