Stress events and Changes in Dysfunctional Attitudes and Automatic Thoughts Following Recovery from Depression in Inpatient Psychotherapy: Mediation Analyses with Longitudinal Data

Different theories have been proposed on depression risk and relapse/recurrence. Among them, endogenous risk hypotheses propose the existence of certain risk factors that are present even before the onset of the first depressive episode, whereas scarring hypotheses surmise each depressive episode evokes accumulating “mental damage or scars” which perpetuate the susceptibility to subsequent episodes (Burcusa & Iacono, 2007). Both perspectives contribute to a better understanding of the onset and maintenance of depression within Beck’s cognitive model of depression. Related to the endogenous risk hypothesis, the cognitive model postulates that depression origins from a “cognitive disorder” (Beck, 1967; Beck et al., 1979; Garratt et al., 2007; Thomsen, 2006; Winkeljohn Black & Pössel, 2015). However, considering the scarring hypotheses, there is increasing evidence that subsequent depressive episodes are associated with an intensification of depressogenic cognitions, which in turn facilitates the risk of a depressive relapse/recurrence (Beck, 2008; Burcusa & Iacono, 2007).

Beck assumes a person’s cognitions are based on attitudes or assumptions (schemas) that are stable cognitive processing patterns formed by experiences in childhood (Otani et al., 2017). In depressed patients, these negative and stereotyped cognitive patterns are called dysfunctional attitudes (DA). In the clinical context, DA are also known as “rules for living” in the form of if–then statements such as “If I am not successful in everything I do, then I must be a failure”, or “If I don’t meet others expectations, then they will reject me” (Beck, 1979; Conway et al., 2015). Related to DA, depressive patients are prone to cognitive distortions (e.g., selective perception, catastrophizing, arbitrary reasoning) which bias their processing of experiences in an often one-dimensional, global, invariant, absolutizing, or irreversible way. As a result, negative evaluations about the self (“I am useless”), the world (“The world is unfair”), and the future (“It will remain so unbearable”) manifest, which Beck calls the cognitive triad. Because of their DA and cognitive distortions, especially in stressful situations, depressive patients’ thoughts are often extreme, negative, categorical, absolute, and judgmental. These automatic thoughts are fast-acting, flashing, subjectively plausible and involuntary occurring cognitions which connect an experience (external or internal) with an emotional sensation (consequences). Automatic thoughts can be both positive and negative (Huffziger et al., 2008; Pössel & Knopf, 2008; Winkeljohn Black & Pössel, 2015).

Commonly, a “sequential relationship” is assumed between DA and automatic thoughts (Alloy et al., 1985). Thereby, DA are viewed as distal trait-like variable, which are indirectly related to depressive symptoms. Distal variables function at a deep level and exert their depressive influence mainly through facilitating proximal variables such as negative automatic thoughts (NAT; e.g., “Nobody understands me”, “What is wrong with me?”) which are highly concomitant with depressive symptoms (Kwon & Oei, 1994; Marchetti et al., 2021; Oei & Sullivan, 1999). Corroborating this, a meta-analysis assembling 140 cross-sectional studies confirmed that individuals reporting more NAT and DA also show more depressive symptoms; however, compared to DA, NAT is more closely correlated to depression (Tang et al., 2020).

Nevertheless, there is ample evidence that DA are associated with the occurrence and intensification of depressive symptoms (Faissner et al., 2018; Smith et al., 2018). Indeed, depressed individuals show significantly more DA than subclinical (Miloseva et al., 2018) and healthy groups (Yesilyaprak et al., 2019). In a randomized controlled trial with 264 remitted depressive patients, presence of more DA at the beginning of the study prospectively predicted depressive relapse/recurrence (Brouwer et al., 2019). It is also known that negative or sad mood can lead to increased access to DA. This phenomenon, called cognitive reactivity (Jeanne et al., 1998), presents a risk factor for disorder relapse/recurrence among remitted depressive patients (Otto et al., 2007; Rojas et al., 2014; Scher et al., 2005; Segal et al., 1999, 2006).

Regarding NAT, there is consistent evidence that clinically depressed individuals report more NAT than non-depressed controls (Chahar Mahali et al., 2020; Dobson & Shaw, 1986; Dozois et al., 2009; Ertekin et al., 2018; Hjemdal et al., 2013; Mohammadkhani et al., 2020; Yesilyaprak et al., 2019), and increase in NAT is viewed as highly relevant for the relapse/recurrence of depression (Pössel & Knopf, 2008; Riley et al., 2017; Winkeljohn Black & Pössel, 2015). NAT are conceptualized as a state variable triggered by situational cues (Hollon & Kendall, 1980; Hollon et al., 1996; Wenzel, 2012). More specifically, stressful events were found to trigger a reactivation of DA which can lead to an increase in spontaneously occurring NAT, subsequently promoting the incidence or relapse/recurrence of depression (Lewinsohn et al., 1999, 2001). Indeed, there is evidence that upon stress, increasing NAT are highly predictive for emerging depressive symptoms (Jacobs et al., 2008; Lakdawalla et al., 2007; Tang et al., 2020). However, other studies found no evidence that DA and related NAT mediate the negative effects of stressful events on depression (Lethbridge & Allen, 2008; Marchetti et al., 2021; Wojnarowski et al., 2019). It was also established that improving depression through psychotherapy accompanies reduction of NAT (Furlong & Oei, 2002; Riley et al., 2017; Tang et al., 2020).

At the same time, improving depression involves an increase of positive automatic thoughts (PAT), for example, “I can achieve everything “ or “I feel good “ (Dozois, 2007; Dozois et al., 2009; Shiraishi, 2005). Accordingly, an inverse relationship between PAT and depressive symptoms is assumed (Ingram et al., 1995; Missel & Sommer, 1983). Indeed, longitudinal studies observed that less occurrence of PAT characterized remitted patients who incurred early relapses after successful psychotherapy (Rojas et al., 2016). In addition, remitted depressive patients experienced less PAT than healthy subjects after negative mood induction, suggesting that PAT may be easily disrupted in remitted depression (Rojas et al., 2016). Accordingly, we expect that less PAT or their decreased occurrence could be a major risk factor for depressive relapse/recurrence. However, as of now, it is unclear if the production of PAT is hindered in individuals with more DA. Moreover, it has not been investigated whether the negative impact of stressful events on depressive relapse/recurrence is mediated through a stress-related decline in PAT occurrence.

To conclude, there is first evidence that stressful events can elicit depressive relapse/recurrence in remitted depressed patients through reactivating DA and subsequently, increasing the occurrence of NAT (Beck & Haigh, 2014; Scher et al., 2005). At the same time, it needs to be investigated to what extent stressful event exposure and DA jeopardize the stability of remission through lowering the occurrence of PAT. Therefore, in this study, we aim to investigate the sequential mediatory role of DA and automatic thoughts in the relationship between stressful events and depressive relapse/recurrence in remitted depressive patients. Thereby, we hypothesized that, (1) stressful event exposure and (2) more DA increase the risk for depression relapse/recurrence after discharge from acute-phase treatment. Additionally, we hypothesized that the negative effect of stress on remission stability is mediated through (3) stress-related reactivation of DA, and (4) stress-related increase in NAT as well as decrease in PAT occurrence. We also tested whether (5) the negative effect of DA on remission stability is mediated through more NAT and less PAT occurrence.

This study uses longitudinal data from a pre-post-therapy study with a 16-months follow-up phase, which comprised four additional measurement time points. All study procedures were approved by the ethics committee of the Medical School of the University of Tübingen (Germany). Data were collected at the Schön Clinic Roseneck (Germany).

The patients’ age, sex, family status, educational level, employment status, and physical health condition were assessed at study enrollment.

Spearman rank correlations for repeated measures were computed (using the R package rmcorr; Bakdash & Marusich, 2016) since majority of variables were non-normally distributed. Friedman tests were conducted to examine whether depressive symptoms and cognitive variables changed across the follow-up. Durbin-Conover tests served as post-hoc tests applying a Holm correlation to adjust p values for multiple testing.

Associations of depressive relapse/recurrence (as binary outcome variable), DA, and automatic thoughts were analyzed with linear and logistic mixed effect models with restricted maximum likelihood (REML) estimators (using the R packages lme4; Bates et al., 2015; and lmerTest; Kuznetsova et al., 2016). Random intercepts (U_0j) were modelled to map the repeated measurement character of the data (i.e., the five time points, t₁: therapy discharge, t₂: 4 months follow-up, t₃: 8 months follow-up, t₄: 12 months follow-up, t₅: 16 months follow-up).

We used a stepwise modelling procedure to test for mediation effects (Baron & Kenny, 1986): (1), depressive relapse/recurrence was statistically predicted by the participants’ cumulated exposure to stressful events at each time point (model: Relapse/recurrence ~ stressful events + U_0j). (2), DA were included as mediator between stressful events and depressive relapse/recurrence (model: Relapse/recurrence ~ stressful events + DA + U_0j). (3), PAT and NAT were entered as mediators between stressful events and DA as well as depressive relapse/recurrence (model: Relapse/recurrence ~ stressful events + DA + PAT + NAT + U_0j). Moreover, (4), DA were predicted by stressful events, and (5), PAT and NAT were predicted by stressful events and DA. We statistically controlled the patients’ history of major depression episodes prior to the index episode leading to the inpatient treatment described above.

All variables were z-standardized for enabling comparability of effects across models. Appropriateness of including additional variables into the models was determined by a relevant reduction of Akaike’s information criterion (AIC). Statistical significance of indirect (mediation) effects was determined by 95% quasi-Bayesian confidence intervals obtained from 5000-times nonparametric bootstrapping (using the R package mediation; Tingley et al., 2014). A sample size of at least N = 19–24 is recommended to test longitudinal mediation effects in data with five measurement points, medium- to large-sized mediation paths, and medium ICCs (0.4–0.6) (Pan et al., 2018).

Self-reported depressive symptoms changed significantly within the 16 months post treatment [Friedman-χ²(4) = 56.27, p < 0.001]. Post-hoc tests (Supplementary Table 1) revealed, depressive symptoms increased substantially within the first four months after discharge and subsequently remained on the stable level (Fig. 1a). Linear mixed effect models indicated the course of depressive symptoms after discharge was not associated with the patients’ sex, antidepressant medication after discharge, and attendance of outpatient psychotherapy (Supplementary Table 2). However, patients with a history of major depression episodes prior to the index episode leading to the inpatient treatment described above showed quicker depression relapse/recurrence after discharge (Supplementary Table 2). Therefore, we subsequently considered lifetime depression as a covariate when modelling depressive relapse/recurrence and depressive symptoms, respectively.

Moreover, seen on average across all patients, their DA [Friedman-χ²(4) = 3.81, p = 0.433, Fig. 1b] and NAT [Friedman-χ²(4) = 5.98, p = 0.201, Fig. 1c] did not alter throughout the observation period. However, the patients indicated their PAT significantly decreased within the first four months after discharge and subsequently remained stable on a comparably low level [Friedman-χ²(4) = 80.67, p < 0.001; Fig. 1d; see Supplementary Table 1 for post-hoc tests].

Correlations of repeatedly assessed variables are reported in Table 3. Negatively and positively connoted stressful events were both associated with more self-reported and clinician-rated depressive symptoms as well as less PAT in the follow-up. To test our hypotheses, as a first step, we modelled the relevance of stressful events and lifetime depression for depressive relapse/recurrence since discharge (Table 4, Model A). Stressful events (b = 1.56, OR 4.8, p < 0.001) and lifetime depression (b = 3.71, OR 41.1, p = 0.043) increased the risk for depressive relapse/recurrence after discharge.

In the second step, DA were added as a possible mediator between stressful events and depressive relapse/recurrence (Table 4, Model B). Adding DA improved the explained variance (R²_marg) of depressive relapse/recurrence from 33.6% to 54.8%. More DA implicated higher risk for depressive relapse/recurrence (b = 2.18, OR 8.8, p < 0.001). DA were relatively stable within individuals across the follow-up period (r_ICC = 0.68; Table 4; Fig. 1b) and did not depend on stressful event exposure (β = 0.06, p = 0.205; Table 4, Model G). Thus, DA did not mediate the association between stressful events and depressive relapse/recurrence (b_indirect = 0.001, p = 0.210; Table 5).

The courses of PAT and NAT were modelled as functions of stressful events and DA (see Table 4, Models H & I). PAT (Fig. 1d) decreased with exposure to stressful events (β =  − 0.50, p < 0.001) but did not depend on the patients’ DA (β =  − 0.02, p = 0.713). Accordingly, the association between stressful events and declining PAT was not mediated through changes in DA (β_indirect = − 0.001, p = 0.790; Table 5). Furthermore, Fig. 1c shows that NAT were relatively stable during the observation period (r_ICC = 0.44), although with higher individual variability than DA (see Table 3). Fluctuation in NAT was not related to stressful events (β = 0.10, p = 0.082); however, patients with more DA also reported more NAT (β = 0.50, p < 0.001).

Considering PAT and NAT as predictor variables alongside stressful events, DA, and lifetime depression explained additional 26.6% of variance (R²_marg) in depressive relapse/recurrence risk (Table 4, Model C; Fig. 2a). DA (b = 2.18, OR 8.9, p = 0.021), PAT (b = − 5.65, OR 0.004, p = 0.021), and NAT (b = 1.48, OR 4.4, p = 0.047) were significantly associated with the risk of depressive relapse/recurrence; however, stressful events showed no direct association with the depressive relapse/recurrence risk (b = 0.87, OR 2.4, p = 0.126; Fig. 2a). Lifetime depression was not a significant predictor of depressive relapse/recurrence (p = 0.067).

Figure 2a provides a graphical summary of the consecutive models. Mediation analyses showed the association between stressful events and depressive relapse/recurrence risk was mainly mediated through the stress-associated decline in PAT (77.5%, b_indirect = 0.136, p = 0.017; Table 5). Furthermore, NAT partially mediated the association between patients’ DA and depressive relapse/recurrence (b_indirect = 0.040, p = 0.049; Table 5).

Note that all analyses were repeated using depressive symptoms (as measured with the BDI-II) as outcome variable which lead to almost identical results. Detailed information of these results can be found in Table 4, Models D–F for linear mixed effect models, and Table 5 for mediation effects. Figure 2b provides a graphical summary of these results.

Our results have relevant implications for therapeutic practice and suggest considering additional therapeutically relevant variables in the cognitive model of depression. First, the maintenance of depressive remission appears to be related to retaining PAT despite of exposure to stressful events after discharge. For patients, it is therefore relevant to derive a realistic perspective on their life circumstances and relevant coping resources. Development of positive or more realistic thoughts can be encouraged through various approaches such as self-induction techniques (Meichenbaum, 2009), positive-psychology interventions (Seligman, 2002, 2011), controlling thoughts techniques (Lewinsohn et al., 2010), as well as mindfulness methods (e.g., MBCT; Segal et al., 2002). Such techniques can help increasing the patients’ awareness of the current situation and their needs, and imparting strategies, mechanisms, and actions to maintain an emotional and physical balance.

Second, sustainably maintaining remission would require effective strategies to consistently reduce DA and associated NAT. Inpatient acute-phase cognitive therapy is likely to be insufficient in altering deep-seated negative cognitive schemas. Autobiographical work, which is typically administered in subsequent outpatient settings, has been proven effective to help patients to elucidate the origins of their DA in their own history. It further enables patients to understand the “functionality” of their DA, and, thereby, increases their willingness to apply learned therapeutic strategies to overcome existing dysfunctional perspectives. Integrating methods from various theories is recommended such as techniques to recognize and change NAT (Beck et al., 1979) and cognitive restructuring to modify DA (e.g., Beck, 1979; Ellis & Dryden, 2007).

Beside cognitive restructuring, patients may learn to detach from their dysfunctional cognitive patterns through “metacognitive” abilities. To this end, MBCT aims to enable patients to become aware of their internal (e.g., dysfunctional attitudes) and external experiences (e.g., stressful events), to accept them through detaching from their negatively biased cognitive patterns, and thereby, to cease worrying and ruminating about these experiences (Segal et al., 2002). In their review, Bockting et al. (2015) argue that MBCT could be specifically beneficial for patients with chronic and recurrent depression (for meta-analyses see Goldberg et al., 2019; Kuyken et al., 2016). Rumination as the uncontrollable and unwanted recurrence of thoughts about negative self-concepts and self-evaluations is associated with an increased risk of relapse/recurrence in depression (Nolen-Hoeksema et al., 2008; Rogers & Joiner, 2017). Consequently, the use of cognitive therapy strategies focused on rumination such as the Response Style Theory by Watkins (2018) is recommended (Marchetti et al., 2021).

Strengths of this study include its multiple follow-up measurements (five times within 16 months), enabling the meaningful analysis of the courses of exposure to stressful events, DA, NAT, PAT, and depression, as well as their interplay. Although the sample size was sufficiently powered to test longitudinal mediation effects (Pan et al., 2018), the study cohort was relatively small and has been recruited using convenience sampling, which is not representative for the patient population. Therefore, replication studies within larger cohorts are required to generalize our findings on patients who differ from the present sample (e.g., in the severity or chronicity of depression, comorbidities, sociocultural influences). Our analysis model is based on the theoretical assumptions of Beck’s theory (1967) and was expanded to include PAT. As an alternative to our meditation perspective, it is conceivable that DA and automatic thoughts moderate the influence of stress on depressive relapse/recurrence (Marchetti et al., 2021). Future studies with larger cohorts could examine moderated mediation effects, and may also consider other cognitive vulnerability variables assumed by Beck (e.g., cognitive distortions, cognitive triad) along with other risk factors (e.g., rumination, maladaptive coping styles) and resilience factors such as positive self-concepts (e.g., functional attitudes, self-efficacy, locus of control, sense of coherence). Moreover, it is commonly surmised that negative cognitive schemas are formed by negative life events, which warrants the investigation of etiological risk factors such as childhood maltreatment and major life events in the prospect of depression relapse/recurrence. Finally, regarding therapeutic methods, it is highly relevant to investigate which therapeutic techniques in outpatient follow-up care (psychotherapy, psychiatry, and counseling) benefit remission maintenance, and whether antidepressant and other psychotropic medication influences cognitive risk variables and benefits the stability of therapy success.