Adolescent depression is considered a global health crisis (Patel, 2013
), provided it is associated with poor psychosocial functioning, poor academic performance, and lower physical and mental health (Andersen & Teicher, 2008
; Fombonne, Wostear, Cooper, Harrington, & Rutter, 2001
, Zisook et al., 2007
). Subclinical levels of depression are present in more than 20% of adolescents (Bertha & Balazs, 2013
) and about 18% of individuals experience full-blown depression before turning 19 year (Lu, 2019
). Hence, examining the cognitive vulnerabilities associated with depressive symptoms in adolescence is of particular relevance since this transition period is characterized by a marked increase of depressive symptoms compared with younger cohorts (Bufferd, Dougherty, Carlson, Rose, & Klein, 2012
; Lu, 2019
). Additionally, early onset of depressive symptoms is associated with more problematic outcomes in adulthood, such as major depression and suicide (Fergusson, Horwood, Ridder, & Beautrais, 2005
). Finally, understanding mechanisms associated with depression in adolescence may provide useful insights for prevention, which is increasingly considered a major priority (Cuijpers, Beekman, & Reynolds, 2012
; Patel, 2013
A variety of cognitive theories have been developed to understand the onset, maintenance, and recurrence of depressive symptoms. The three major theories are the cognitive theory
), the hopelessness theory
(Abramson, Metalsky, & Alloy. 1989
), and the response styles theory
(Nolen-Hoeksema, Girgus, & Seligman, 1992
). Although characterized by different features (see below), these three theories share important similarities. For instance, they all adopt a vulnerability-stress perspective, in that encountering stressors is deemed to be necessary to activate cognitive vulnerability and, in turn, trigger the onset of depressive symptoms. Moreover, they have usually been proposed as sequential theories (Alloy, Clements, & Kolden, 1985
), with some cognitive vulnerabilities being directly linked to depressive symptoms (proximal vulnerabilities
, i.e., automatic thoughts or brooding), while other vulnerabilities being linked to symptoms primarily via other mechanisms (distal vulnerabilities
, i.e., dysfunctional attitudes or negative cognitive style) (Pössel & Knopf, 2011
). We briefly describe these three major theories below.
) describes in his cognitive theory the onset and maintenance of depressive symptoms as being due to a cascade of cognitive vulnerabilities, such as dysfunctional attitudes, cognitive errors, the negative cognitive triad, and negative automatic thoughts. According to Beck (1976
), dysfunctional attitudes are rigid and maladaptive assumptions that substantially alter and steer information processing, when activated by stressors. The influence of dysfunctional attitudes is then exerted through cognitive errors, such as catastrophizing, overgeneralization, and selective abstraction, which have people draw negative and unhelpful interpretations. This leads the person to generate a system of negative beliefs that is characterized by a negative view of themselves, their future, and their surrounding world (i.e., negative cognitive triad). As a result, negative automatic thoughts, namely thoughts that come rapidly and automatically to mind when a person is stressed or upset, dominate mental activity and eventually spur the development of the emotional, somatic, and motivational symptoms of depression.
Similarly to the cognitive theory, the authors of the hopelessness theory emphasize the importance of altered cognitive processes, but their focus is specifically on negative cognitive style (Abramson et al., 1989
). Negative cognitive style refers to attributing a negative event (i.e., stressor) to stable and global causes and drawing negative inferences about the self-worth of the individual and the consequences of the stressor. This renders the individual often unable to resolve their issues which causes hopelessness (negative view of their future; Pössel & Thomas, 2011
). Research shows that the habitual adoption of this type of causal attributions and inferences over time leads to depressive symptoms (Alloy et al., 2000
Finally, according to the response styles theory (Nolen-Hoeksema et al. 1992
), what determines the onset, maintenance, and severity of depressive symptoms is the cognitive response to negative mood. If an individual reacts to stressors with rumination, namely negative, repetitive, self-referential thinking, they are likely to develop depressive symptoms over time. Among the different forms of rumination, the most maladaptive type is brooding, which refers to passively focusing on symptoms of distress and on the meaning of these symptoms (Treynor, Gonzalez, & Nolen-Hoeksema, 2003
). Solid research shows that brooding is strongly associated with concurrent and future depressive symptoms and suicide attempts (Nolen-Hoeksema, Wisco, & Lyubomirsky, 2008
; Rogers & Joiner, 2017
These three cognitive theories have been extensively tested in adult populations and rapidly became the dominant framework for understanding the onset and maintenance of depression (Beck & Haigh, 2014
; Liu, Kleiman, Nestor, & Cheek, 2015
; Nolen-Hoeksema et al., 2008
). Yet, a number of important issues remain poorly investigated and understood.
First, only a handful of studies have investigated vulnerabilities from multiple cognitive theories simultaneously (Lam, Smith, Checkley, Rijsdijk, & Sham, 2003
; Pössel & Knopf, 2011
; Smith, Alloy, & Abramson, 2006
), while exploring the integration between these theories is crucial. In fact, the development of an integrated model could shed light on how a certain mechanism influences and is influenced back by the other vulnerability factors, regardless of the theory that originally proposed it. In turn, this could not only help us gain a deeper understanding of how vulnerability to depression functions, but also strengthen our clinical interventions. In fact, therapeutic procedures often rely on a certain cognitive model in order to modify mechanisms proposed in another model without any theoretical justification (i.e., technical eclecticism; Lampropoulos, 2001
). For instance, in the context of the Penn Resiliency Program (Gillham, Jaycox, Reivich, Seligman, & Silver, 1990
), adolescents are trained to question their inferences (i.e., hopelessness theory), by evaluating their accuracy and generating alternative inferences, as proposed in Beck's cognitive model. Second, most of the studies were conducted in adult samples, rather than adolescents (for exceptions, see Pössel & Pittard, 2019
; Winkeljohn Black & Pössel, 2015
). This is unfortunate, as previous studies showed that the transition from adolescence to adulthood is characterized by complex patterns of symptoms change (Costello, Copeland, & Angold, 2011
). Third, little is known about whether the interplay among cognitive vulnerabilities in adolescence is stable or if it is subject to change due to the transitory nature of this developmental phase (Hankin et al., 2009
). While cognitive vulnerabilities are posited to be relatively stable after late childhood (Hankin & Abramson, 2002
), some have emphasized that vulnerabilities for depressive symptoms may change with brain maturation in adolescence (Davey, Yücel, & Allen, 2008
Network analysis represents an alternative and innovative avenue to investigate the interplay between vulnerabilities from multiple cognitive theories. Based on graph theory, statistical network analysis is an exploratory approach, which conceives of psychological phenomena as causal systems where the constituting elements mutually influence and reinforce each other (Borsboom & Cramer, 2013
). Crucially, network analysis provides important benefits over other approaches. First, it is a bottom-up approach, which does not require strong prior assumptions about the phenomenon under investigation (van den Berg et al., 2020
). Although this does not constitute an added value per se, this feature can turn out very helpful when modelling many elements with no strong theoretical predictions about interrelations between variables or when mutually exclusive models are reported in the literature. Second, this approach generates a specific hypothesis about the possible network of causal links (called edges) among the constituting elements (called nodes) of the system (Dalege, Borsboom, van Harreveld, Conner, & van der Maas, 2016
). By doing so, network models can easily model feedback loops (i.e., cyclic models), which are likely present in the context of psychopathology (i.e., bidirectional influence between rumination and sad mood) (Borsboom & Cramer, 2013
). Third, network analysis can unveil previously unknown links among factors as well as identify which elements may play the most central role in the network (van den Berg et al., 2020
). In turn, these pieces of information can both inform our knowledge of the phenomenon under investigation and propose new targets for clinical interventions. It is also important to mention, however, that network analysis shows several limitations, among which the fact that it does not model random error, it usually generates hundreds of parameter, whose relevance is evaluated in a subjective fashion, and the information elaborated at the group-level may transfer to the individual-level to a limited extant (Forbes, Wright, Markon, & Krueger, 2017a
Although traditionally used to investigate symptoms and attitudes (Borsboom & Cramer, 2013
; Dalege et al. 2016
), network analysis has recently been applied to unveil the interplay among psychological constructs (i.e., structural network analysis; Epskamp, Rhemtulla, & Borsboom, 2017
; Guyon, Falissard, & Kop, 2017
), such as vulnerabilities to depression in adults (Faelens, Hoorelbeke, Fried, De Raedt, & Koster, 2019
) and in remitted depressed individuals before and after treatment (Hoorelbeke, Marchetti, De Schryver, & Koster, 2016
; Hoorelbeke, Van den Bergh, Wichers, & Koster, 2019
). In the context of adolescence, previous studies have adopted network analysis to decipher the interrelationship among depressive symptoms (Mullarkey, Marchetti, & Beevers, 2019
), emotional and behavioral problems (Boschloo, Schoevers, van Borkulo, Borsboom, & Oldehinkel, 2016
), and posttraumatic stress symptoms (Russell, Neill, Carrión, & Weems, 2017
). Recently, Bernstein and colleagues (2019
) applied network analysis to investigate the temporal relationship between specific components of cognitive style, as derived from the hopelessness theory, and depressive symptoms in adolescents. Interestingly, this study showed that the development of stable and global negative attributions predict future depressive symptoms. However, no study has so far tested the structure and longitudinal (in)stability of multiple cognitive vulnerabilities derived from different theories of depression, stressors, and depressive symptoms in adolescents using network analysis.
In our longitudinal network analysis study, we measured several cognitive vulnerabilities derived from the three cognitive theories described above along with depressive symptoms and stressors (i.e., stressful life events). We recruited a large sample of adolescents and assessed them four times (wave 1 to wave 4) across a period of one year (i.e., one assessment session every four months). By doing so, we could model the relationship among cognitive vulnerabilities, stress, and depressive symptoms at cross-sectional level, and also investigate the stability of their interplay longitudinally. To meet these goals and ensure the trustworthiness of our results, we performed standard network analysis along with stability and accuracy checks. Finally, for the first time to our knowledge, we investigated the possible moderating role of stressful events (i.e., vulnerability-stress perspective) in the context of moderated network analysis (Haslbeck, Borsboom, & Waldorp, 2019
Cognitive vulnerabilities to depressive symptoms in adolescents is increasingly attracting the attention of both scholars and clinicians, given that at least 27% of adults with major depression have had their first episode in adolescence (Kessler, Petukhova, Sampson, Zaslavsky, & Wittchen, 2012
). Although many vulnerabilities have been identified, only limited efforts have been done to clarify how different vulnerabilities in concert lead to depressive symptoms. In our study, we adopted a network analysis approach to shed light on the specific interplay among major cognitive vulnerabilities in adolescence.
The analyses revealed that all the vulnerabilities considered were included in the network (i.e., no node was expelled), although they did not cluster in ways that were anticipated or specified in the original theories. For instance, negative cognitive style (hopelessness theory; Abramson et al., 1989
) appeared to potentially exert an influence on depressive symptoms, mostly through automatic thoughts and cognitive errors (cognitive theory; Beck, 1976
), while brooding (response styles theory; Nolen-Hoeksema et al., 1992
) emerged to possibly play a mediating role between dysfunctional attitudes (cognitive theory) and depressive symptoms. Taken together, these findings clearly demonstrate that the vulnerabilities from the three original cognitive theories are related to each other and provide suggestions on how these cognitive vulnerabilities jointly impact depressive symptoms. Our study represents a preliminary step in the direction of a more comprehensive understanding of cognitive risk for depression.
Furthermore, our study also confirmed that the architecture of cognitive vulnerabilities to depressive symptoms is likely to be sequential (Alloy et al., 1985
), with some cognitive vulnerabilities acting as proximal and others as distal vulnerabilities. Among the proximal cognitive vulnerabilities were automatic thoughts, the negative cognitive triad, and brooding, which could explain, across the four waves, about 17%, 14%, and 8% of depressive symptoms variance, respectively. In keeping with these results, previous studies have identified negative automatic thoughts as surface level cognitions (Kwon & Oei, 1994
) that highly covary with depressive symptoms (Oei & Sullivan, 1999
). Moreover, although Beck (1976
) does not consider the negative cognitive triad as a direct contributor to depressive symptoms, but only through automatic thoughts, our study showed that having a negative view of the self, the future, and the world did have a direct relationship with depressive symptoms. This is in line with previous research showing that cognitive vulnerabilities impact depressive symptoms via both direct and indirect pathways (i.e., partial mediation) (Pössel & Winkeljohn Black, 2014
). Finally, in accordance with the response style theory, brooding had a significant role in explaining depressive symptoms, although less than automatic thoughts and negative cognitive triad (i.e., 8% vs 17% and 14%, respectively).
Our study also highlighted three distal factors, namely dysfunctional attitudes, cognitive errors, and negative cognitive style, which were loosely connected with depressive symptoms. In keeping with the results of the network analysis, relative importance analysis showed that, when considered together, these distal vulnerabilities could account for only about 12% of depressive symptoms variance. It is perhaps worth mentioning that these cognitive vulnerabilities refer to cognitive contents and processes that are not expected to influence depressive symptoms directly, but via proximal factors. In fact, dysfunctional attitudes primarily refer to rigid and unrealistic attitudes about performance perfectionism and approval by others (i.e., cognitive content; Cane, Olinger, Gotlib, & Kuiper, 1986
), while cognitive errors and negative cognitive style capture a biased way to elaborate negative information (i.e., cognitive process; Giuntoli et al., 2019
). Hence, it is reasonable to presume that these distal vulnerabilities probably function at a deep level (Kwon & Oei, 1994
) and their influence on depressive symptoms is primarily exerted through cognitive contents and processes that are expressed in the above mentioned proximal vulnerabilities, such as automatic thoughts, negative cognitive triad, and brooding.
Automatic thoughts emerged as the strongest node of the network and this result held very consistently across the four waves. Although it is tempting to equate node strength with causality, caution with this interpretation is recommended (Dablander & Hinne, 2019
; Rodebaugh et al., 2018
). In fact, on the one hand, in their seminal work Beck, Rush, Shaw, and Emery (1979
) proposed that negative automatic thoughts act as mediator of change in cognitive therapy, hence implying that these cognitions play a causal role in eliciting depressive symptoms. On the other hand, a recent systematic review of the literature showed that only in about half of the studies the mediating role of the automatic thoughts was empirically confirmed (Lemmens, Müller, Arntz, & Huibers, 2016
). This latter point raises the possibility that automatic thoughts may not always be the source of depressive symptoms, but in some instances reflect a by-product of other cognitive vulnerabilities (i.e., negative cognitive triad or brooding) or depressive symptoms per se. In the light of our study, future studies should clarify the causal status of the automatic thoughts by means of direct experimental manipulation (Lemmens et al., 2016
To our knowledge for the first time in the context of network analysis, we tested whether experiencing stressful events could interact with cognitive vulnerabilities when impacting depressive symptoms (i.e., moderated network analysis). Although our analysis revealed no evidence supporting the vulnerability-stress model, we recommend caution in interpreting this result (“absence of evidence is not evidence of absence”; Altman & Bland, 1995
), as many reasons, among which statistical, methodological, and theoretical ones, could account for this result.
First, our study may have failed to detect significant interactions due to power issues (i.e., Type II error). Although possible, this explanation is unlikely given that previous studies have shown that high level of true-positive rate (sensitivity) and true-negative rate (specificity) are reached in networks with features similar to ours (i.e., less than 26 nodes, normally distributed nodes, gamma parameter equal to 0.5; Epskamp, 2016
). Second, we only evaluated the occurrence of external stressors (e.g., arguments with the parents or receiving a failing grade), while internal stressors of mental (i.e., spontaneous thoughts, feelings, and mood; Beck et al., 1979
; Marchetti, Koster, Koster, & Alloy, 2016
) or organic (i.e., neuro-inflammation; Maes et al., 2009
) nature were not considered. Third, not all the theories considered in this study attribute the same importance to the role of stressful events. For instance, the role of stressors is central for the activation of the vulnerability factors for the hopelessness theory and for some of the vulnerability factors in Beck’s cognitive theory, while their role is less vital for the response styles theory. Moreover, cognitive theories often stress that the individual’s vulnerability is reactive only to specific types of stressors (i.e., sociotropy and interpersonal stressors; Hammen, Ellicott, Gitlin, & Jaminson, 1989
). In our study, we did not consider stress-vulnerability type matching. Fourth, we only considered contemporaneous networks in our study, an aspect which might have obscured the joint impact of stressful events and vulnerability factors on depressive symptoms over time (i.e., temporal networks; Epskamp et al., 2018
). However, at the theoretical level, our results may also be interpreted as being in line with the alternative etiologies of cognitive vulnerabilities (Parry & Brewin, 1988
). According to this model, no interaction between stressors and vulnerabilities is required, as they both impact depressive symptoms, but in an independent fashion. In sum, although our moderated network analysis did not provide any evidence supporting the diathesis-stress approach in adolescence, many interpretations could be proposed. Hence, future studies should further explore this crucial aspect of vulnerability to depression and depressive symptoms with appropriate research designs and more sensitive measures for both stressors and vulnerabilities.
Finally, our study revealed that the interrelationship among cognitive vulnerabilities, depressive symptoms, and stressors is markedly stable at about 15 years old, in that no difference in the network structure was detected at this age during a period of 12 months. Our findings are in line with previous research (Hankin et al., 2009
), showing that already during middle adolescence (i.e., 15–17 years old) cognitive vulnerabilities are somewhat stable as compared to childhood or early adolescence (Hankin, 2008
; Romens, Abramson, & Alloy, 2009
). Moreover, again in line with previous research (Hankin et al., 2009
), small-to-negligible reductions in the mean levels of the constructs investigated were found. Although difficult to explain, these fluctuations have been usually attributed to suboptimal psychometric properties of the instruments and regression toward the mean (Hankin, 2008
; Romens et al., 2009
). Overall, our study suggested that network structure and intensity of cognitive vulnerabilities and depressive symptoms has reached a point of marked stability already in middle adolescence, and perhaps even earlier.
Our study is characterized by both strengths and limitations. Among the former is that, first, we investigated the interrelationship among several cognitive vulnerabilities by relying on network analysis. Although previous studies have attempted an integration of the cognitive theories of depressive vulnerability (Lam et al., 2003
; Pössel & Knopf, 2011
; Smith et al., 2006
), different and sometimes mutually exclusive models have been proposed in the literature. Hence, considering that there is no consensus as to how the different mechanisms are interrelated, in our study we did not impose any a-priori constraint on the statistical model and took full advantage of the exploratory nature of network analysis. We recommend that future studies complement our results with confirmatory techniques, such as structural equation modeling. Second, in this study, we considered the major cognitive vulnerabilities as described in the three major cognitive theories of depression, and we followed them up in a longitudinal manner. Third, we ran sophisticated analyses, such as moderated network analysis and cross-validation analysis, in order to increase the informativeness and replicability of our findings. Fourth, our results held, even after changing the measure of depressive symptoms (i.e., from CES-D to CDI), hence implying a high degree of trustworthiness.
Among the limitations is, first, the sole reliance on self-report questionnaires. While different methods are available for capturing depressive symptoms and life stressors, only self-report questionnaires are available for measuring cognitive vulnerabilities (Gotlib & Neubauer, 2000
). Hence, we preferred relying on the same method, in order to avoid discrepancy among the different nodes of the network. Second, we only followed the adolescents for a total period of about 12 months, which might have prevented detecting changes in the network structure that require a longer period to emerge. It is worth mentioning, however, that substantial stability of cognitive vulnerabilities was also found in a 7-year longitudinal study on adolescents (Romens et al., 2009
). Third, it is not clear to what extent our sample was representative of the students at the school, where this study was carried out, and of the general adolescent population. Future studies may want to consider a more stratified recruitment to ensure as much as possible that participants were demographically representative of the student population. Moreover, about 16% of participants dropped out across the four waves. Although not negligible, the attrition rate is in line with previous studies carried out in adolescents (i.e., Hankin, 2008
) and, importantly, almost identical results were obtained when missing data was treated with imputation procedure. Fourth, given the presence of only four timepoints, it was not possible to estimate temporal networks, such as graphical vector autoregressive modelling (Epskamp, Waldorp, Mõttus, & Borsboom, 2018
) and cross‐lagged panel network (Rhemtulla, van Bork, & Cramer, 2017
). Future studies could directly address this important research question, by implementing longitudinal studies with multiple assessment points over the period of one year or more.
Our study paves the way for future studies. For instance, although several cognitive vulnerabilities were considered simultaneously, only about 57% of the variance of depressive symptoms across the four waves could be accounted for. Note that this estimate is likely to be inflated, as it is highly unlikely that all the cognitive vulnerabilities investigated actually exert a causal influence on depressive symptoms. Hence, future studies should consider adopting a multilevel and multifactor approach (Hankin et al., 2009
), where other crucial mechanisms are included, such as cognitive biases (Marchetti et al., 2018
), temperament (Compas, Connor-Smith, & Jaser, 2004
), social support from families, friends, and teachers (Pössel et al., 2018
), neural pattern (Davey et al., 2008
), and genetic substrate (Fox & Beevers, 2016
). Furthermore, our study was silent with respect to the temporal dynamics that generated the network of cognitive vulnerabilities and their trajectories after adolescence.
The findings of this study have important implications, in that they could help calibrate clinical interventions in the context of depressive risk in adolescents. Our results indeed suggest that, on the one hand, targeting automatic thoughts, cognitive triad, and brooding may reduce depressive symptoms in an effective way. Interestingly, this datum spurs for the integration of techniques derived from different theories, such as the “ABC” technique for detecting and changing automatic thoughts (Beck’s cognitive model; Beck et al., 1979
), the cognitive restructuring for modifying the cognitive triad (Beck’s cognitive model; Beck et al., 1979
), and the rumination-focused cognitive therapy for reducing brooding (response style theory; Watkins, 2018
). On the other hand, clinical or experimental interventions on inferences, dysfunctional attitudes, and cognitive errors could be helpful in hindering the onset and recurrence, but they could have limited impact on depressive symptoms per se. In sum, our study provides an emerging framework for guiding clinicians in their work with adolescents with depressive symptoms.
In conclusion, our study shows that, at the age of 15 years old, the architecture of cognitive vulnerability is already markedly stable, with some vulnerabilities being more central and proximal (i.e., automatic thoughts, cognitive triad, and brooding) and others being more distal (i.e., negative cognitive style, dysfunctional attitudes, and cognitive errors).