Depression is one of the most prevalent disorders among adolescents [1]. An estimated 16% of adolescents between ages 12 and 17 reported at least one major depressive episode in 2019 [2]. Many of these adolescents experience additional depressive episodes during their lifetime, and the consequences (i.e., occupational, social, psychological) are long-lasting [3]. Considering the pressing need for intervention, it is concerning that current treatments for adolescents have demonstrated limited success in preventing the recurrence of depressive episodes [4]. To provide earlier and more effective treatment of depression in adolescence, we need a better understanding of underlying vulnerabilities implicated in its development. To this end, the present study focuses on rumination and inhibitory control weakness—two proposed cognitive vulnerabilities for depression—and investigates how they relate to one another in individuals with and without prior major depressive episode(s) during a crucial developmental stage (i.e., adolescence) when cognitive control abilities are still solidifying [5,6,7]. In addition, cognitive vulnerabilities for depression are influenced by the presence of other risk and resilience factors in the adolescent’s environment [6, 8]. Thus, the second focus of this study was to assess whether common risk factors (e.g., childhood maltreatment) and resilience factors (e.g., supportive family environment) moderate any associations between rumination and inhibitory control.

Rumination

Rumination is the habit of thinking repetitively, recurrently, uncontrollably, and passively about a negative past or current event [9, 10]. Rumination as a thinking style is thought to emerge first in early adolescence around the time of puberty and has lasting negative effects across development [5, 8, 11]. Importantly, higher rumination has been found to predict onset and more severe symptoms of a depressive disorder in adolescents with a family history of mental health disorders [12], in currently depressed adolescents, and in adolescents with elevated neuroticism [13]. In particular, the “maladaptive brooding” component of rumination, as opposed to relatively more adaptive, “reflective pondering” over stressful events, has been linked to elevated depression levels in adolescent community samples [14, 15]. Moreover, adolescent and adult females tend to show elevated rumination levels compared to males [11, 16], thus conferring greater risk for developing depression. Overall, rumination is a robust predictor of a depressive course in adolescents.

Inhibitory Control

Poor inhibitory control (IC) is a hypothesized vulnerability for both depression and ruminative thinking, though the effect of rumination on IC and depression may be indirect [17, 18]. IC is one of various higher-level cognitive processes, grouped under the term cognitive control, which regulate lower-level cognitive processes, such as motor control [17]. IC refers to the process of (1) disregarding, or (2) removing unwanted/irrelevant stimuli, or (3) refraining from engaging in a prepotent behavior [19, 20]. IC is measured with various paradigms, such as behavioral accuracy in a Go/No-Go task, which measures prepotent response inhibition [21, 22]. Development of IC begins at about age 6 [7, 23]. IC abilities continue to grow and become more efficient throughout adolescence, plateau and stabilize in adulthood, and decline again with later age [7, 24, 25].

Models of Inhibition and Rumination

Multiple theoretical models suggest that maladaptive rumination stems from cognitive control difficulties. One such theoretical framework, first proposed by Linville (1996) [26] and later expanded on by Joorman (2010) [27], places impaired inhibitory control mechanisms at the root of ruminative thinking. Here, impaired inhibition interferes with the ability to prevent the entering of as well as the ability to remove negative information from working memory, resulting in easier and faster retrieval of negative content from memory [26, 27]. As a result, the affected individual remains focused and ruminates on the negative information that is stuck in their memory. Other theories have been proposed that suggest differing cognitive processes, such as set-shifting, as invoked by the impaired disengagement hypothesis. According to Koster and colleagues (2011) [28], rumination is elicited by prolonged processing and the inability to disengage attention away from negative information that concerns the self. In a similar model by Whitmer and Gotlib (2013) [29], negative mood and/or individual differences in attentional scope limit the variety of thoughts and actions that are activated in working memory and available from long-term memory. This narrowing of the attentional scope can lead to negative information remaining stuck and can explain why the individual has difficulties breaking away from the loop of rumination and negative mood [29]. Across all frameworks, certain cognitive weaknesses result in greater risk of relying on or being unable to stop maladaptive rumination [24].

While studies on adults have shown a weak association between higher rumination and less IC in adults with and without depression [22, 30], especially when presented with negative stimuli [31, 32], it is unclear whether an association between the two processes will replicate on a non-affective Go/No-Go task in an adolescent sample. Studies have indicated some evidence of IC weaknesses in adolescents with depression [21], which are also present after remission in young adults [33], but findings are not consistent across the literature and cast doubt on observing an association in adolescent samples [34]. In the large Adolescent Brain Cognitive Development Study on pre-adolescent youth, executive functioning, of which IC is a core component of, predicted general psychopathology two years later, but not internalizing symptoms [35]. Other studies on pre-adolescents [36] and late adolescents [37] did not find support for an association between inhibition and rumination. Findings from another community-sampled study indicated that inhibitory control weakness at age 3 was predictive of elevated rumination levels at age 9, but only in the context of having a highly angry temperament in early childhood [38]. In one adolescent community sample, rumination was associated with inhibition weakness during an emotional Go/No-Go task, especially in response to negative information when shifting from negative to positive stimuli, but not during a non-affective task assessing inhibition [39]. Signs of IC weakness on neural levels have also been observed in adolescents with depression, but only in response to sad faces on an affective Go/No-Go task, not to positive stimuli [21]. Taken together, these studies underscore the importance of stimuli valence during the inhibition task when studying cognitive control difficulties in youth. Affective context may also explain why IC weaknesses are not always evident with non-affective inhibition tasks, and may explain why adolescents appear to have difficulties controlling their responses to information that is emotionally stimulating [8].

Environmental Factors

In addition to IC weaknesses and rumination, childhood maltreatment (CM) and perturbations in the family environment are significant risk factors for depression [6, 8, 40]. CM includes any act by a caregiver that causes harm or the threat of harm, such as physical, sexual, and emotional abuse or neglect [41, 42]. Recent national reports estimate that approximately 9 per 1,000 children in the US population were subjected to maltreatment, with the number of referrals to social services being even higher [43]. Self-reported exposure to CM is associated with higher rumination in adult samples [44, 45]. However, few studies have examined these associations within adolescents. Research on the association between CM and IC has been more substantial. For example, adolescents and adults with and without mental health disorders who reported exposure to CM showed delayed cognitive development and poorer executive functioning overall [46] and weaker IC [47,48,49,50].

A supportive family environment is generally considered an important resilience factor that acts against the development of depression during adolescence and adulthood [51,52,53,54], although little empirical research has examined its impact on rumination or IC. In Olson et al.’s (1992) Circumplex Model of family systems, healthy family functioning encompasses a positive communication style, balanced levels of cohesion or the emotional bond between family members, and adaptability in a family when facing a stressor [55]. One study found that children with poor IC ruminated less if they had parents who displayed positive communication and interaction styles compared to children whose parents displayed negative communication styles [38]. However, the overall lack of research on the associations among CM, family support, rumination, and IC in adolescence presents a missed opportunity for research on vulnerability factors for depression. Studying these potential associations in closer temporal proximity to their occurrence could improve our understanding of the order of effects, as well as minimize memory bias effects in self-reporting, and can offer a window into more timely intervention efforts.

Aims and Hypotheses

Prior research has highlighted important aspects of the development of IC and the emergence of rumination during childhood and adolescence, particularly in the context of vulnerabilities for depression. One hypothesis is that cognitive control weaknesses, childhood maltreatment, and family environment (e.g., parenting behaviors, emotional expressiveness) generate distress in the adolescent, who may begin to rely on rumination as an emotion regulation strategy [8, 17]. Over time, this tendency may develop into a ruminative habit, placing the adolescent at risk for depression. Another hypothesis is that IC weakness in children who experienced CM and poor familial support contribute to greater difficulty disengaging from rumination [56].

The first goal of this study was to investigate whether an association between IC and rumination observed in a meta-analysis of adults would replicate and extend to two samples of adolescents with varying vulnerability levels to depression (due to a prior personal history of depression or familial depression) [30]. Hence, the present study assessed adolescents to understand IC and rumination processes in close temporal proximity to cognitive control development, CM exposure, and the family environment. Timely identification of these vulnerabilities for depression enables future work on early intervention. Given that both IC and rumination have been identified as cognitive vulnerabilities for depression, we expected that poorer performance on an IC task would associate with a greater frequency of rumination in two samples of adolescents, some of whom were preselected for high levels of rumination. Understanding whether there is a relation between IC and rumination would allow for targeted enhancement of IC before the rumination habit develops or targeted modulation of rumination in the context of IC weaknesses in adolescents at risk for depression.

The second goal was to investigate whether additional common risk (e.g., CM) and resilience factors such as family support or cohesion related to the development of depression would affect the association between IC and rumination. In line with previous findings related to CM [44, 49, 50], we hypothesized that adolescents who experienced higher levels of CM would show a stronger link between IC and rumination when compared to those who reported less CM. With cohesion broadly understood as a resilience factor [51,52,53], cohesion was expected to buffer the effects of IC weaknesses on rumination levels. Due to limited research on associations among cohesion, IC, and rumination, moderation analyses of cohesion as a buffer were exploratory. Assessing how IC, CM, and cohesion are implicated in the development of rumination may allow for more precise temporal predictive models that can help assess which adolescents are at greatest risk for a depressive episode, thereby informing targeted intervention.

Method

Participants

A total of 281 adolescents aged 11 to 17 were screened across two studies. Of those, 191 participants were ineligible or withdrew, resulting in the present sample of 90 adolescents. Forty-seven adolescents identified as high ruminators were recruited as part of a randomized clinical trial funded by the National Institute of Mental Health (R61MH116080), which assessed the effectiveness of Rumination-Focused Cognitive Behavior Therapy (RF-CBT) in preventing the recurrence of adolescent depression. Of those, 45 adolescents were recruited based on a prior major depressive episode (MDE) currently in remission (1 adolescent did not meet MDE criteria), whereas two were recruited for a subsample of actively depressed adolescents. The remaining 43 female adolescents in this sample had no personal experience of a depressive episode, recruited as part of a larger study examining the intergenerational transmission of risk for depression (F31MH117856), with 60% (n = 26) at an increased risk for depression onset due to maternal depression history. In the present total sample, the majority were female (84%) and non-Hispanic (93%). Average age was 14.6 (SD = 1.8). A detailed overview of the demographic and clinical data of the full sample and by MDE group can be found in Tables 1 and 2, respectively.

Table 1 Demographic and Clinical Data of Total Sample
Full size table
Table 2 Demographic and Clinical Data of Sample by Adolescent’s MDE History
Full size table

Measures

Rumination. Rumination levels were assessed with the total score of the Ruminative Response Scale (RRS) [57], a 22-item self-report scale measuring three subfactors of rumination: reflection, brooding, and depression-related thinking. Items are rated on a 4-point Likert scale from 1 (almost never) to 4 (almost always). Scores range from 22 to 88, with higher scores representing higher rumination levels. For the Rf-CBT study included in this analysis, cut-off scores of higher ruminators for inclusion were adjusted based on age (14–17 years) and gender, ranging from 28 to 31 for males and 35 to 38 for females to account for possible age and gender effects during targeted rumination treatment. Several studies have used the RRS or a modified version to measure rumination in adolescent samples and support its reliability and validity with younger samples [14, 58, 59]. Internal consistency for the scale in the present sample was excellent (Cronbach’s α = 0.95).

Inhibitory Control. IC performance was assessed using the Parametric Go/No-Go/Stop (PGNGS), a computerized task measuring attention, set shifting, and processing speed using non-affective stimuli, such as letters or shapes [60]. Both a shape and a letter version were used in the current study, though the shapes version was used in analyses for most of the participants (PGNGS shapes, n = 76; PGNGS letters, n = 14). During each level, a series of shapes (i.e., circles, diamonds, half-ovals, parallelograms, pentagons, squares and triangles) – or letters in the other version - is presented for 600 ms each with no interstimulus interval. The participant attempts to respond to each target as quickly and accurately as possible by following certain task rules and clicking a set of keyboard keys (i.e., pressing the index finger on the “n” key on the computer keyboard) across three task levels with increasing difficulty for a total of 6 trial types (i.e., 2-target Go, 2-target No-Go, 2-target Stop, 3-target Go, 3-target No-Go, 3-target Stop). The first difficulty level (Go) assesses the participant’s attention and response time to two presented targets (i.e., circles and diamonds) every time they appear (e.g., all-targets rule). The second level (No-Go) assesses sustained attention, response inhibition, processing speed, and response time deviation by measuring the response accuracy and time to two targets (i.e., circles and diamonds) every time they appear without repetition (e.g., non-repeating rule). This means participants should not respond to the same target twice in a row but rather alternate between them. A stop-go version (Stop) is also included as a pilot measure. Participants respond to all targets again, however this time, a classic insertion of a stop signal midway into the presentation of some targets is the cue to stop responding to that specific target (e.g., stop-sign rule). The final difficulty level assesses the participant’s sustained attention, inhibition, set-shifting, and interference following the same all-targets rule, non-repeating rule, and stop-sign rule of the two-target trials, this time with three items in the target set (i.e., circles, diamonds, and triangles) to increase difficulty. Accuracy is measured by the percentage of correct target trials in inhibitory control (PCIT) by dividing the total number of correct target responses by the total possible target responses in only the No-Go conditions. Response time is the average response time for a correct target. Each participant’s accuracy on inhibitory trials (PCIT) during the 3-target trial on the second level (No-Go) was used as the primary dependent variable to estimate IC in this study.

Participants completed this task from home using Pavlovia software [61] or in-person using E-Prime 3.0 software [62], according to current CDC and university mandates for COVID-19 safety. There was no significant difference in accuracy across software presentation method (t(88) = 1.14, p = .26). Given that different versions of the task were used in the two studies (i.e., letters vs. shapes), performances on each subscale were converted to z-scores to ensure comparability. In cases where PGNGS data looked suspicious due to (1) poor performance on both PCIT and low attention as measured by percent correct on target trials (PCTT), set below 60% for the PCTT, or (2) high PCIT performance with low PCTT, only the score from the letter version of the PGNGS was used as the final PCIT score (n = 6). Participants were supervised by a trained research assistant in person during the letter PGNGS version limiting the probability of distractors and decreased attention during the task as opposed to the virtual application of the shapes PGNGS version, which had only limited supervision via video conferencing.

Childhood Maltreatment. The Childhood Trauma Questionnaire – Short Form (CTQ-SF) was used to assess adolescents’ self-reported history of abuse and neglect [63]. The CTQ-SF consists of 28 items that capture five types of maltreatment, namely emotional, physical and sexual abuse, as well as emotional and physical neglect. Items are rated on a 5-point Likert scale, from 1 (never true) to 5 (very often true). Scores range from 25 to 125 and are grouped into four severity levels for each subtype (emotional abuse: none = 5 to 8, low = 9 to 12, moderate = 13 to 15, severe > 16; physical abuse: none = 5 to 7, low = 8 to 9, moderate = 10 to 12, severe > 13; sexual abuse: none = 5, low = 6 to 7, moderate = 8 to 12, severe > 13; emotional neglect: none = 5 to 9, low = 10 to 14, moderate = 15 to 17, severe > 18; physical neglect: none = 5 to 7, low = 8 to 9, moderate = 10 to 12, severe > 13) [64]. Internal consistency in the present sample was good (Cronbach’s α = 0.74).

Family Cohesion. Perceived family cohesion was measured with the adolescent’s ratings of the Family Adaptability and Cohesion Scale-II (FACES-II) [55], a self-report measure assessing family cohesion and adaptability to situational and developmental stressors as well as family communication [55]. Thirty items are rated on a 5-point Likert scale, from 1 (almost never) to 5 (almost always). Cohesion scores range from 15 to 80 and are classified into disengaged (15–50), separated (51–59), connected (60–70), and enmeshed (71–80) cohesiveness, with separated and connected bonds indicative of healthy family functioning [55]. While ratings on family cohesion were collected from both adolescents and their parents, only the adolescents’ self-reports were used in our analyses to reflect their perception of family dynamics and support due to the low correlation between parent and adolescent ratings (r = .23, p = .09). Internal consistency for the full scale (Cronbach’s α = 0.95) and cohesion subscale (Cronbach’s α = 0.93) were excellent, for the adaptability subscale (Cronbach’s α = 0.89) good.

Procedure

Participants were recruited via online postings, hospital mass recruitment letters, flyers, mass transit advertisements, radio advertisements, and listserv postings. Participants were screened for eligibility during initial phone screens and through RedCap surveys. If initial eligibility criteria were met, assent and informed consent were provided by the adolescent and parent or guardian. Next, participants attended virtual or in-person diagnostic screening interviews with an independent evaluator to confirm diagnoses with the Kiddie Schedule for Affective Disorders and Schizophrenia for School-Age Children Present and Lifetime Version (K-SADS-PL) [65] and Lifetime Suicide Attempt Self-Injury Count (L-SASI) [66]. Participants included in the present analyses encompassed adolescents with (n = 46) or without (n = 44) a history of a major depressive episode, based on criteria from the Diagnostic and Statistical Manual of Mental Disorders, 5th edition (DSM-5) [67]. All participants attended virtual or in-person diagnostic, neuro-imaging, and neuropsychological assessment visits. Adolescents were compensated for their time with gift cards ($50 - $60, depending on the study), and legal guardians received cash ($30 - $60, depending on the study) as well as prints of the adolescent’s brain scan. Both studies were approved by the Institutional Review Board at the University of Utah. A list of inclusion and exclusion criteria can be found in Supplemental Table 1.

Data Analysis

SPSS Statistics version 27 for Windows was used to analyze the data (SPSS Inc., Chicago, Illinois). For the main analyses, hierarchical regression models were used to examine the association between IC and rumination levels and to test potential moderation effects of CM and perceived family cohesion (exploratory). To do so, models were built sequentially one main effect at a time, with interaction effects added, and each new model was compared using changes in F. For the first model, IC (PGNGS-PCIT) served as the predictor variable and rumination levels (RRS total) as the outcome variable to determine whether weaker IC was associated with higher levels of rumination in adolescents (Hypothesis 1). Though previous research has highlighted that the brooding component of the RRS is especially indicative of depression levels in adults and adolescents [14, 15], studies investigating the association between cognitive control and rumination in adolescents have largely relied on the Children’s Response Style Questionnaire (CRSQ) [36, 38, 39], which does not differentiate between rumination subtypes as the RRS does. Results from our posthoc analyses of our models with the RRS-Brooding score as the outcome variable were very similar, if not identical. Thus, only the results with the total RRS scores are presented here. Sex assigned at birth, age, and the presence or absence of an MDE were included as covariates in all regression models to account for any confounding effects [11, 68]. In the second model, severity of CM (CTQ) was entered as the moderator into the first regression model to test its potential moderation effects on the relation between IC and rumination (Hypothesis 2). In the second exploratory moderation model, perceived family cohesion (FACES-II cohesion) served as the moderator variable on the association between IC and rumination (Hypothesis 3). An alternative theoretical approach may have been to analyze mediation models. However, prerequisites for mediation (i.e., statistically significant associations between IC and CM or cohesion) were not met (r = − .05, p > .05; r = − .12, p > .05, respectively). A priori power analysis for the regression models estimated a required sample size of 89 and 119 for a medium effect size at α = 0.05 and power = 0.95 for model 1 and models 2/3, respectively [69]. Due to the COVID-19 pandemic, initial recruitment targets to exceed this sample size were unsuccessful.

Prior to analyses, all variables were checked for normality and missingness. Due to positive skewness in the distribution, CTQ total scores were transformed using the logarithm of 10. Missingness affected approximately 27% of the FACES-II cohesion data points, resulting in a smaller sample of 66 for model 3. Little’s MCAR test was significant, indicating that the missing data was not at random (X 2 = 11.00, df = 4, p = .03). Average age and RRS scores were higher for adolescents with missing cohesion data (t(58.6) = -3.9, p < .001; t(57) = -3.0, p = .004, respectively). In addition, Hispanic adolescents and those with a prior MDE were more likely than expected to have missing cohesion data (X2 = 5.26, df = 1, p = .02; X2 = 10.31, df = 1, p = .001, respectively). To reduce any potential Type 2 Error, listwise deletion was employed. As an alternative data analytic approach, imputation was not applied here due to the poor correlation between adolescent and parental ratings of cohesion (r = .23, p = .09) and confirmed patterns of missingness. Intercorrelations among the main variables are included in Table 3; Figs. 1 and 2, 3 and 4. Assumptions of linear regression, including normality of residuals, linearity and additivity between predictor and outcome variables, and outliers, were assessed using visual inspection of the residual values’ distribution, scatter- and q-q plots and were all within acceptable limits. Additionally, multicollinearity did not appear to be a concern in our data (PCIT, Tolerance = 0.74, VIF = 1.35; CTQ-Total, Tolerance = 0.42, VIF = 2.39; Cohesion, Tolerance = 0.49, VIF = 2.03; Sex, Tolerance = 0.83, VIF = 1.20; Age, Tolerance = 0.44, VIF = 2.29; Group, Tolerance = 0.41, VIF = 2.46). Dummy variables were created for participants’ sex assigned at birth (0 = male, 1 = female) and sample group (0 = prior MDE, 1 = no prior MDE). PCIT scores were standardized and averaged across PGNGS versions when multiple scores were available (i.e., Shapes scored with Pavlovia, Shapes scored with E-Prime, letters scored with E-Prime). Tasks with demonstrably poor effort and performance were excluded from final analyses as described in the Methods section (n = 6) [70]. Interaction effects (β) and 95% confidence intervals (CI) are reported for each model. In addition to controlling for sex assigned at birth, each model was tested in the female sample only. Results for the female-only sample did not differ from the full sample and are presented within each table accordingly.

Fig. 1
figure 1

Bar Graph of Average Rumination (RRS) by Age and Sex. Error bars set at 95% CIs

Full size image
Fig. 2
figure 2

Scatterplot depicting the relation between IC (PCIT) and rumination (RRS). No significant linear association was found. Solid line indicates linear regression line with 95% CI (dashed lines).

Full size image
Fig. 3
figure 3

Scatterplot depicting the association between maltreatment (CTQ) and rumination (RRS). Adolescents who reported more maltreatment tended to ruminate more. Solid line indicates linear regression line with 95% CIs (dashed lines).

Full size image
Table 3 Correlation Matrix of Main Variables of Interest
Full size table

Results

Youth ratings on the total RRS averaged 46.9 (SD = 15.0) and on the brooding subtype averaged 11.2 (SD = 3.9) in the total sample, which is higher than has been found in healthy adolescents [11, 13] but similar to other at-risk samples (see Fig. 1 for RRS scores by age and gender) [13]. This is likely accounted for by the fact that half of the present sample was specifically recruited for frequent rumination. In line with our recruitment strategies, groups based on adolescents’ MDE history differed significantly in their rumination levels, in that adolescents with a prior or current MDE reported higher rumination (t(2,88) = 8.14, p < .001 for RRS-Total; t(2,88) = 6.89, p < .001 for RRS-Brooding). Regarding IC, the average PCIT score in the full sample was 57%, lower than in a slightly older sample of young adults with and without an MDE history using a comparable PGNGS version [33]. No significant difference in inhibition performance between MDE groups was observed. Total CTQ averaged 36.5 (SD = 11.1), indicating no to low frequency of CM in the total sample, and was within 1 SD from a normative, large community sample of young adults [64]. Of note, adolescents with a MDE history reported significantly higher CM levels overall and by subtype, except for physical abuse, compared to adolescents without a MDE history (t(2,88) = 5.23, p < .001 for CTQ-Total; t(2,88) = 5.05, p < .001 for CTQ-EA; t(2,88) = 2.06, p = .043 for CTQ-SA; t(2,88) = 5.08, p < .001 for CTQ-EN; t(2,88) = 2.97, p = .004 for CTQ-PN). Total cohesion scores on the FACES-II averaged 59.1 (SD = 11.9) in the full sample, indicating separation within the family [55]. Adolescents with MDE indicated significantly lower cohesion scores than did adolescents without an MDE history (t(2,64) = -2.94, p = .005). Additionally, groups differed in age (t(2,88) = 8.28, p < .001), CDRS scores (t(2,86) = 8.63, p < .001), and Adaptability on the FACES-II (t(2,64) = -3.36, p = .001). Further details on demographic and clinical data, as well as intercorrelations among the variables of interest, can be found in Tables 1 and 2, and 3.

Fig. 4
figure 4

Scatterplot depicting the relation between perceived family cohesion (Cohesion) and rumination (RRS). Adolescents who reported more cohesion reported less rumination. Solid line indicates linear regression line with 95% CIs (dashed lines).

Full size image

Model 1: Association of Inhibitory Control and Rumination

To test whether IC predicted rumination in the present sample, linear regression analysis was performed with PCIT during the 3-targets No-Go task of the PGNGS as the predictor variable, total RRS score as the outcome variable, and sex assigned at birth, age, and sample group as covariates. Results for the full sample and female-only sample are presented in Table 4. There was no main effect of PCIT (IC) on rumination (see Fig. 2 ), although both sex (females) and sample group (MDE) were significant positive predictors. The model for the full sample was overall significant (F(4, 89) = 19.94, p < .001) and accounted for 48% of the variance in rumination levels.

Table 4 Regression Results for Model 1
Full size table

Model 2: Childhood Maltreatment as Moderator

We continued with planned analyses to examine whether a history of CM moderated the association between IC and rumination. Here, CTQ and the interaction term between PCIT and CTQ were added into the first model to test for moderation. Results are shown in Table 5. As in model 1, the effect of IC on rumination levels, accounting for the effects of all other variables, remained below statistical significance in the full moderation model with CTQ and its interaction with PCIT. The effect of CTQ on rumination was significant when controlling for all other variables (see Fig. 3 ), such that those who experienced more CM also reported higher rumination levels. Importantly, the interaction effect between PCIT and CTQ was non-significant, indicating that CM did not moderate the association between IC and rumination levels in the present sample.

Table 5 Regression Results for Model 2
Full size table

Overall, the regression models with CTQ added (F(5, 89) = 18.24, p < .001) and with the interaction term between PCIT and CTQ (F(6, 89) = 15.03, p < .001) were significant. Further, the addition of the CTQ into the model predicted additional variance (∆F(1, 84) = 6.39, p = .01), accounting for 52% of variance in rumination levels. However, the addition of the interaction term of the PCIT and CTQ was not significant (∆F(1, 83) = 0.04, p = .85).

Model 3: Perceived Family Cohesion as Moderator

We also ran a planned analysis to assess whether adolescents’ perceived family cohesion moderated the association between IC and rumination. To do so, the family cohesion score and the interaction term between PCIT and family cohesion were added into the first model. Results are presented in Table 6. Again, the effect of PCIT on rumination levels remained non-significant while controlling for all other variables in the full moderation model with family cohesion and its interaction with PCIT. The effect of cohesion on rumination when accounting for all other effects was significant (see Fig. 4 ), where those who perceived greater family cohesion reported lower rumination levels. Notably, the interaction effect between PCIT and cohesion was again below statistical significance levels, indicating that the adolescent’s perceived cohesion did not moderate the association between IC and rumination levels.

Table 6 Regression Results for Model 3
Full size table

The regression models with family cohesion (F(5, 65) = 14.62, p < .001) and with the interaction term between PCIT and cohesion (F(6, 65) = 11.98, p < .001) were significant. Further, the addition of family cohesion into the model predicted additional variance (∆F(1, 60) = 10.48, p = .002), accounting for 55% of variance in rumination levels. Of note, the addition of the interaction term between PCIT and cohesion into the model was not significant (∆F(1, 59) = 0.00, p = .99).

Discussion

The overarching goal of this study was to elucidate potential risk and resilience factors related to rumination in adolescents at risk for depression. The two aims were: (1) to examine the association between IC and rumination within an adolescent sample with varying levels of risk for depression (due to the presence or absence of a prior personal depressive episode; model 1), and (2) to assess whether common risk and resilience factors of depression affect the association between IC and rumination. For the second aim, CM (model 2) and cohesion (model 3) were tested in planned analyses as potential moderators of the association using hierarchical regression modeling.

For the first model, IC was hypothesized to be associated with rumination, such that those with lower IC performance would report higher rumination levels. Results failed to demonstrate a significant association between IC and rumination. Contrary to prior literature on adults [30] and our hypotheses, IC was not related to how much the adolescent ruminated, even in our relatively large sample of adolescents with and without risk for depression/recurrence. Our non-significant results are perhaps not surprising, considering that others have found only a weak or inconsistent association between IC and rumination [22, 34, 71], or only in the context of affective IC tasks [21, 39], in the context of high anger [38], or in brain connectivity patterns [72]. It is also possible that a more robust and unified approach to measuring IC abilities would have yielded different results. We only used response accuracy (i.e., PCIT score) as a measure of IC and neglected reaction time for this set of analyses. In addition, the letter instead of the shapes version of the PGNGS task was used for a small subsample (n = 14); either due to missing, incomplete or suspicious looking PCIT scores on the shapes version. Despite our attempt to standardize PCIT scores across both task versions to ensure comparability, the slight differences in the task designs, including type of targets and mode of delivery (remote online versus in-person at the study site), likely affected the strength of our findings due to increased variance. Furthermore, the small subsample of PGNGS letter scores may have skewed PCIT z-scores. Multiple factors may have contributed to the current lack of evidence supporting the association between IC and rumination, including the non-affective task design, inconsistent measurement of IC, two study samples, and the relatively large age range.

Still, our results were consistent with and added weight to previous literature that does not show a reliable relation between non-affective IC weakness and ruminative habits in adolescence. It is important to remember that adolescents are in a developmental stage where IC abilities and ruminative tendencies are still emerging and are not yet fully stable and consistent [24]. In addition, neurodevelopment occurs at varying rates for adolescents, especially as they approach their pubertal onset [73, 74]. This developmental instability and heterogeneity are possible threats to finding consistent results across a sample such as ours. However, continuing research on the mechanisms and risk factors that can contribute to the negative cycles of rumination, such as IC weaknesses, while at-risk adolescents form potentially maladaptive emotion regulation strategies into adulthood is crucial. The differences in the adolescent and adult literature highlight a shift in the developmental process between adolescence and adulthood when IC weaknesses begin to relate to rumination tendencies with increasing consistency. Identifying the window in the developmental process of when and for whom this shift occurs requires continued research efforts and ultimately would offer an opportunity for early intervention and training of IC abilities.

Although no main effect of IC was found, results confirmed that female sex and history of depression were significant predictors of rumination. In the present sample, female sex at birth and having a personal history of depression were associated with higher levels of rumination relative to male sex at birth and no personal history of depression. These findings remained significant throughout all other models and support previous research on gender effects in rumination in a non-clinical community sample [11] and on susceptibility to depression [68]. Both are important sample characteristics to consider when studying ruminative tendencies.

Building from the first model, CM, an established risk factor for depression, was hypothesized to moderate the association between IC and rumination. Contrary to our hypothesis that adolescents who report more CM would have a stronger link between IC and rumination than those who reported less CM, there was no significant interaction effect between IC and CM on rumination levels, indicating that the experience of CM neither worsened nor improved the link between IC and rumination. While CM did not moderate our proposed relation between IC and rumination, maltreatment was significantly associated with rumination levels. Adolescents who reported more CM tended to report higher rumination levels, confirming previous research that has shown that CM is a risk factor for rumination in adults [44, 45]. It is worth mentioning here that we did not find a correlation between CM and IC (see Table 3), contrary to other studies [47,48,49,50]. Taken together, our results suggest that the experience of maltreatment and inhibitory control are two separate processes that can affect how much an adolescent ruminates, though our study did not find support for the latter.

In the third exploratory model, cohesion was examined as a potential moderator of the association between IC and rumination. As with the previous model, results did not indicate significant interaction effects between IC and cohesion on rumination levels. Specifically, whether the adolescent perceived a strong or weak cohesion bond within their family did not impact the link between IC and rumination, rendering no support for the hypothesis that higher perceived family cohesion buffers against the effects of inhibitory control functioning on rumination. However, higher cohesion as an independent predictor was related to lower rumination. This finding is consistent with prior work with children and adolescents, indicating that family support and cohesion are resilience factors against the development of the rumination habit [38, 51].

There were two primary take-away findings from the study. As already discussed briefly, CM was a significant statistical predictor of rumination levels in our sample (see model 2), although it did not moderate the association between IC and rumination. In line with previous research on adults [44, 45], there was a strong positive relation between CM and rumination levels, such that those who reported higher CM levels tended to ruminate more (see Fig. 3 ). As O’Mahen et al. suggest, maltreatment is often unpredictable and uncontrollable, especially for adolescents who may rely on or live in close proximity to the source of adversity [75]. Living in such a stressful environment may encourage the adolescent to adopt coping strategies, such as rumination, as a way of processing or avoiding the experience and emotions [75]. Though intended to help them face adversity, these coping and emotion regulation strategies may then grow into more maladaptive or passive ruminative habits, which place the adolescent at risk for depression and more severe depressive symptoms [10, 12, 13, 76]. As adolescents emerge into young adult developmental roles and spaces where it is adaptive and expected to approach and problem solve, avoidance, passivity and rumination may be a hindrance to adaptive functioning. Indeed, our results demonstrated that adolescents who had endured frequent and more severe CM tended to ruminate more, placing them at an increased risk for a maladaptive rumination habit and ultimately, depression. Thus, clinical assessment of adolescent depression would benefit from increased screening efforts of maltreatment history for early intervention. Similarly, our findings support the need for treatment protocols for rumination and depression that are especially sensitive and targeted to adolescents who may have experienced abuse and neglect.

The present study also showed that the adolescent’s perceived family cohesion was associated with their rumination levels (see model 3). As can be seen in Fig 4. , adolescents who perceived a stronger cohesive bond within their family reported lower rumination levels. It is possible that having a cohesive family bond signals to the adolescent that a support system is available and prevents them from engaging in ruminative thoughts and the opportunity for other adaptive and collaborative habits. Those without a perceived social support network, on the other hand, may tend to ruminate more because they are not presented with alternative distraction and problem-solving skill learning opportunities. They may also lack encouragement from others to engage in more adaptive coping strategies [51, 77] or may find them to be incongruous in low cohesion families. Furthermore, those who tend to ruminate may also seek out less support from their family because they perceive their family dynamics more negatively and less supportive in the first place [51]. Our findings stressed the importance of identifying family dynamics and level of support as they play a significant role in an adolescent’s emotion regulation tendencies in protective and potentially detrimental ways. As such, assessment of the adolescent’s family bonds can be used to inform treatments for rumination and depression in multiple ways: fostering a cohesive family environment and integrating the family’s support into the treatment protocol, addressing and building stronger family dynamics when perceived support is lacking, and practicing more adaptive problem-solving skills.

Strengths and Limitations

This study investigated potential risk and resilience factors for rumination within a sample of adolescents at varying levels of risk for depression. Adopting a developmental approach to studying the emergence of rumination and depression allows for early identification of those who are at heightened risk for depression. Incorporating construct-based aspects of cognitive functioning, like IC, can potentially help us to understand how to train, modulate or scaffold protective skills.

Limitations of this study include a relatively smaller sample size for some of our models, limiting the power to detect significant effects with hierarchical regression. Since sufficient data was only available for 90 adolescents for model 2 and for 66 adolescents for model 3, due to COVID-19 adjustments, the estimated required sample size of 119 for adequate power in these analyses was not met according to a priori power analyses. The demographic and clinical characteristics, as well as the study-specific recruitments methods of the samples employed here also present a limitation for the generalizability of the findings. As seen in Table 1, recruited adolescents were largely White, female, and with a limited range of reported CM, limiting the ability to generalize the current findings in males, underrepresented minorities, and adolescents with greater childhood maltreatment. In addition, family cohesion demonstrated significant missingness patterns, limiting our current understanding of family cohesion to the current full-data sample’s younger age range and primarily non-Hispanic make-up with generally lower ruminative patterns and no prior MDE. Replication of greater family cohesion associated with lower rumination is needed in older youth and especially in youth with high ruminative habits and prior MDE history.

The validity of this study may also have been improved by including more complex, multifaceted measures of the variables of interest. Rumination levels, CM, and cohesion were assessed with subjective adolescent self-reports only, introducing the potential for correlated biased reporting. For example, a recent meta-analysis on retrospective and prospective assessments of CM showed poor agreement between two types of CM measures [78], and we noted a discrepancy between family and adolescent measures of cohesion. IC was measured with one non-affective computer performance task in a mostly virtual environment, measurement was limited to behavioral performance only, and utilized one summary score (accuracy) that did not incorporate other aspects of IC (e.g., reaction time). Incorporating multiple measures-based reports, observation, and performance tasks across research settings would strengthen the study design.

Future Directions

The rumination habit is one of many vulnerability and risk factors for depression. The goal of the present study was to investigate the association between a potential vulnerability factor for rumination and depression (i.e., IC) and rumination itself and to examine potential moderating effects (i.e., CM, cohesion). This study on a unique sample of adolescents at risk for depression enriches the existing literature on rumination and IC by showing that the association did not replicate in a non-affective IC task. Furthermore, higher CM, lower cohesion, female sex, and history of a major depressive episode were related to higher rumination levels. These relations emerged early, even in 11-year-olds with a maternal family history of depression. Therefore, future research should continue to study these risk and resilience factors for rumination in earlier age ranges to inform research and interventions for those at risk of experiencing depression, including younger, preteen at-risk samples. Efforts to prevent depression recurrence in adolescents through treatments focused on the ruminative habit are currently underway [79, 80], and similar strategies for prevention are a high priority area.

Summary

Many individuals experience a depressive episode during adolescence. Rumination, the habit of thinking repetitively and passively about a negative past or current event, is one well-studied vulnerability for depression and may be linked to inhibitory control weaknesses. The roles of other potential risk and resilience factors are less clear. Thus, the goals for this study were twofold: (1) to replicate the association observed in adults between inhibitory control and rumination in a sample of adolescents, and (2) to examine putative moderating roles of childhood maltreatment (i.e., risk) and perceived family cohesion (i.e., resilience) in a sample of adolescents at risk for depression due to a prior personal history. Ninety adolescents aged 11 to 17 (M = 14.6, SD = 1.8), and their legal guardians participated. Adolescents completed validated self-report scales of rumination, childhood maltreatment history, and family cohesion, and performed a non-affective computerized task assessing inhibitory control. Hierarchical regression models showed no significant relation between inhibitory control and moderator variables on rumination. However, childhood maltreatment and family cohesion directly predicted rumination (β = 27.52, 95% CIs [5.63, 49.41], β = -0.40, 95% CIs [-0.65, -0.15], respectively). Adolescents who reported higher levels of childhood maltreatment and who perceived lower family cohesion tended to indicate higher levels of ruminative habit. These findings demonstrate an alternative understanding of factors that increase depression onset risk and recurrence in adolescents. As such, these findings inform identification and targeted intervention efforts to reduce the likelihood of depression recurrence in adolescence.