The transdiagnostic approach stands out within the comprehensive models for mental disorders developing in recent decades (Barlow et al., 2004; Harvey et al., 2004; Ehring and Watkins, 2008). This approach emerges as a response to the deficits of categorical classification systems (Sandín et al., 2012; Dalgleish et al., 2020). It posits that mental problems are caused by processes that exist in the population on a gradient from normal to dysfunctional and that these psychopathological processes are common to more than one disorder, which would explain the high comorbidity and transitions between diagnoses across the lifespan (Nolen-Hoeksema and Watkins, 2011; Pascual-Vera and Belloch, 2018). The transdiagnostic approach gains increasing support from psychiatry and psychology thanks to advances in research on psychopathological processes and the development of effective psychotherapeutic intervention programs (Norton and Paulus, 2016; Pascual-Vera and Belloch, 2018; Heckendorf et al., 2019; Dalgleish et al., 2020; Eustis et al., 2020; Leonardo et al., 2021).

Transdiagnostic processes correspond to nuclear dysfunctional processes that are shared by different mental disorders with high comorbidity, which can be explained by similar theoretical models and respond to the same psychotherapies or treatments (Pascual-Vera and Belloch, 2018; Dalgleish et al., 2020). After decades of research in this area, Nolen-Hoeksema and Watkins (2011) pointed out that further investigation of certain elements of the transdiagnostic model is required. One of them is the multifinality of the transdiagnostic process, i.e., how a transdiagnostic process manages to express itself in various disorders, together with divergence, referring to how it is that a person with a transdiagnostic process develops a specific type of disorder and not another. These aspects remain to be fully defined until now (Liew et al., 2023).

Repetitive negative thinking is a cognitive process characterized by recurrent thoughts and self-focused attention in the face of negative events, whose effects can be adaptive or maladaptive (Cova et al., 2019; Lawrence et al., 2021); the most well-known repetitive thinking is the maladaptive one, usually referred to as RNT, that includes negative rumination, and worry, while on the adaptive side is reflection (Nolen-Hoeksema and Watkins, 2011; Cova et al., 2019; Lawrence et al., 2021). Although repetitive negative thinking is employed as a strategy for self-regulation and modulating own affective experiences, it may be counterproductive, and it has been associated with affective disorders (Ferrer et al., 2018; Spinhoven et al., 2018), eating disorders (Palmieri et al., 2021), increased vulnerability to stress-related disorders in those exposed to childhood adversities (Mansueto et al., 2021), difficulties in emotion regulation (Mansueto et al., 2022, 2024; Palmieri et al., 2023), perfectionism (Kummer et al., 2024; Palmieri et al., 2024) and shyness (Palmieri et al., 2018).

Nolen-Hoeksema (1991) defined depressive rumination as passive and repetitive thinking about one’s own depressive symptoms and their possible causes and consequences; she considered it a tendency or style of thinking stable over time that increases and prolongs depressive symptoms, increasing the probability that they become chronic. Over the years, a more detailed analysis of the scale used to study depressive rumination (Ruminative Response Scale, RRS) made it possible to distinguish two components of rumination: reflection and brooding (Treynor et al., 2003). The latter was conceptualized not only as a response to depressive symptoms but also to negative events or stressors in general (Nolen-Hoeksema et al., 2008). To this day, the distinctive elements of brooding include an orientation towards the past and an evaluative component of the experience in a negative way (Watkins and Roberts, 2020). Conversely, reflection was defined as turning inward to perform cognitive problem-solving to alleviate distress symptoms (Treynor et al., 2003). Subsequently, other authors have defined a construct similar to reflection, adaptive rumination, which is characterized as repetitive thinking focused on understanding the negative process experienced and the search for solutions (Watkins, 2008).

Brooding as a thinking style has been widely studied as a transdiagnostic process, accumulating cross-sectional evidence on its relationship with depressive and anxiety disorders, and suicidal ideation (Tucker et al., 2013; Ferrer et al., 2018). Longitudinal studies in the Chilean population have established that negative rumination influences the development of depressive and anxiety symptoms (Cova et al., 2009; Pimentel and Cova, 2011), and international evidence shows the impact of rumination on the development of suicidal ideation at one-year follow-up (Miranda and Nolen-Hoeksema, 2007).

Reflection or adaptive rumination has less empirical support to date (Cova et al., 2019; Lawrence et al., 2021). One element that has been debated is whether it operates as a protective factor against stress since, although it is associated with fewer symptoms in cross-sectional studies, it is usually related to brooding at follow-ups (Cova et al., 2009, 2019). Some characteristics proposed to consider reflection as adaptive are the attributes of being deliberate, controllable, concrete, and self-distancing, although several of these elements have little research (Cova et al., 2019; Lawrence et al., 2021).

Worry is defined as a chain of thoughts and images focused on the potential negative implications of possible negative events, such as being fired, failing an exam, getting sick, having an accident, or even positive ones, such as a promotion, love relationship, etc. (Papageorgiou, 2006). Initially, it was associated exclusively with generalized anxiety (Newman et al., 2013), but over time, evidence of its action as a transdiagnostic process for other anxiety disorders, depression, pain disorder, eating disorders, and psychosis was gathered (Ehring and Behar, 2020). In Chile, a one-month follow-up study of RNT did not establish the influence of worry on the development of anxious symptoms in university students but of brooding (Pimentel and Cova, 2011).

In the last decade, the field of study of RNT has expanded the focus of determining what type or content of negative repetitive thought-ruminating or worrying is associated with which disorders to know which RNT processes make it maladaptive. Among the attributes proposed for the study are intrusiveness, lack of control, duration, and self-immersive or first-person perspective (Cova et al., 2019; Rosenkranz et al., 2020; Lawrence et al., 2021); elements that are usually assessed with Perseverative Thinking Questionnaire (PTQ). Studies show cross-sectional associations between process RNT and social anxiety and depression in adolescents (Klemanski et al., 2017). It has also been observed that the presence of higher RNT allows distinguishing people with different anxiety disorders and depression and differentiating them from healthy controls (Wahl et al., 2019). Factor analyses of the PTQ in patients with social anxiety found that pre- and post-event RNT clustered into a single RNT factor (Wong et al., 2019), leading the authors to postulate that content toward the future (worry) or toward the past (brooding) is less relevant than process RNT as transdiagnostic elements. A five-year longitudinal investigation of people with anxiety and depressive disorders found that process RNT mediated progression between diagnostics, usually from anxiety to depression (Spinhoven et al., 2019).

Traditional RNT studies used cross-sectional designs using questionnaires, which had limitations such as difficulties in establishing causality in relationships and recall biases due to depressive mood, among others (Connolly and Alloy, 2017; Bravo et al., 2019; Hamonniere et al., 2020). Ecological momentary assessment (EMA) overcomes this difficulty; this design involves the repeated sampling of participants’ behaviors and experiences in real-time, in their natural context (Shiffman et al., 2008). In EMA studies, reactive RNT is referred to as an RNT that is assessed moment-to-moment to distinguish it from RNT styles assessed with questionnaires at baseline.

EMA studies have provided prospective evidence with ecological validity on RNT. Connolly and Alloy (2017) developed an EMA study where reactive rumination predicted increased depressive symptoms in the face of stress in a seven-day follow-up study; furthermore, there was an interaction between stress and reactive rumination in predicting increased depressive symptoms. Another study was able to establish the influence of suicide-specific reactive rumination on the completion of a suicide attempt in an 18-day follow-up of patients with suicidal ideation (Rogers et al., 2021). In relation to worry, Newman et al. (2019) conducted an eight-day follow-up study of patients with anxiety and healthy controls, observing that reactive worry is associated with greater anxiety in the following hour. In line with the study of RNT types and process, a follow-up with EMA detected that RNT process, repetition, intrusiveness, duration, and perceived burden influenced the development of anxious-depressive symptoms and stress at 14 days and were more relevant than RNT types (Rosenkranz et al., 2020).

The present study used ecological momentary assessment to evaluate a transdiagnostic model that includes RNT and stress for the one-week prediction of eight groups of psychiatric symptoms. We hypothesized that higher levels of university life stress and reactive repetitive negative thinking would increase symptom development beyond RNT styles and previous-term academic stress. We also expected different effects of moment-to-moment stress and reactive RNT on the different syndromes (divergence).

It had been estimated that a sample of 155 persons was required considering an F-test for multiple linear regression with 10 predictors, with an expected R² of 10% incremental over any previous model, with a significance level of 5% and a power of 80%. In total, 309 students aged 18 to 25 years from various faculties of a Chilean university filled out the baseline questionnaires. Following previous studies in EMA (Rosenkranz et al., 2020), the final sample consisted of 238 participants (77.0%) who met the criteria defined for the analyses: responding to the day 1 baseline assessment, responding to the EMA assessment on at least 3 days, with at least 13 measurements in total. Of these, 121 were female (56.7%) and 117 were male (42.3%), with an average age of 20.9 years (SD = 1.79), 69 (29.0%) had been in mental health treatment in the past year, and 54 (22.7%) had received a diagnosis of mental disorder in the past year.

The study was approved by the Scientific Ethical Committee of the university where it was performed (Certificate number CB1123-2022) and is in accordance with the Declaration of Helsinki. Students were invited to participate on campus via posters and online announcements on social media. In-person, after signing the informed consent protocol, a brief admissibility test was applied on a web page, excluding cases with suicidal risk or psychotic symptomatology. Excluded persons were given feedback and a pamphlet with ways to seek help. Accepted participants were introduced to the EMA app (OURMIND developed at Concepcion University for research use), which they had installed on their smartphone (iOS or Android). Participants received the equivalent of US$24 for their participation upon completing 80% of the assessments.

On day one, participants responded to the SCL-90R, RRS, PSWQ, SISCO II, and a sociodemographic form. The EMA period started on day one, lasted 6 days (days two to seven), and consisted of five daily assessments. Each time, the students answered items about academic and non-academic stress, reactive RNT content, and process. The SCL-90R questionnaire was reevaluated on day eight.

The application data were exported to Excel format and statistical analyses were conducted in R, version 4.3. In this study, missing data were addressed using multiple imputations by Fully Conditional Specification (FCS). This approach is particularly suitable for our dataset, which exhibits a complex pattern of missingness (Liu and De, 2015). The imputation process was carried out through 20 iterations in each of the 20 imputed datasets, a number determined to be sufficient for convergence based on diagnostic checks. Each variable with missing data was imputed conditionally on all other variables in the model, using an appropriate model for each variable type (e.g., logistic regression for binary variables and linear regression for continuous variables). After the imputation process, analyses were performed on each of the 20 datasets separately, and the results were then combined using standard rules for multiple imputation inference. SCL-90 R Anxiety and Phobic Anxiety subscale scores were summed as in a common anxiety scale (Saldivia et al., 2023).

Descriptive analyses of the self-report questionnaires and EMA variables assessed were performed. The reliability of the questionnaire measures – symptom measurements with SCL-90R and RNT styles – was calculated using Cronbach’s alpha. An exploratory factor analysis was used to define academic and non-academic stress dimensions. The number of factors was estimated using Horn’s parallel analysis, using the least squares extraction method and oblimin rotation. To estimate the reliability of the EMA measures, EMA-stress and EMA-RNT, the method proposed by Cranford et al. (2006), was used through intraclass correlation.

To test the predictive power of the study variables in the development of SCL-90R symptoms, seven hierarchical stepwise linear regression models were used, organized according to a temporal criterion defined by the sequence of data collection and theoretical criteria. The first four models respond to variables obtained from the first measurement (baseline, day one): Model 1 considers only sociodemographic; Model 2 adds the baseline psychiatric symptoms (SCL-90R); Model 3 includes the RNT style: brooding, worry and reflection (RRS and PSQW); Model 4 adds the term academic stress (SISCO-II). Models 5 to 7 include the EMA measurements (days two to seven). Model 5 adds EMA- academic and non-academic stress. Model 6 includes the EMA-RNT process, which would arise from the EMA-stress. Finally, EMA-RNT content is added in model 7.

To determine the predictive power of each model, R² values were estimated; to determine the differences in predictive value between each model, Rubin’s D1, analogous to an Anova for missing data (F value), was used as a test. The relative fit of the models to the data was tested using AIC and BIC to determine the best model combining parsimony with the best predictive power, where lower AIC and BIC values reflect a better fit. Since the comparison between models using AIC and BIC must be performed on the same database, it is not possible to choose the AIC or BIC with the lowest average among the imputed bases; therefore, the strategy used was to identify, for each imputed base, which was the model with the lowest AIC and BIC, and in cases where both indicators corresponded to different models, the model with the lowest AIC in most of the bases was chosen (Vrieze, 2012).

Table 2 shows the descriptive statistics of the study scales with the grand mean of each scale. The grand mean of syndromic groups in SCL-90R ranged from 0.58 for psychoticism to 1.58 for obsessions, being equivalent to other Chilean samples of university students (Gempp). In the case of repetitive thinking styles, grand mean were 2.18 for rumination to 3.00 for worry. Although the results for rumination and reflection are similar to previous samples, the grand mean for worry is higher in this sample (Pimentel and Cova, 2011). Academic stress during the semester obtained a grand mean of 2.97, equivalent to a study in a similar sample (Guzmán-Castillo et al., 2022).

Table 3 shows descriptive data of the EMA items with grand means of the EMA-stress and EMA-RNT process and content scales. It is not possible to compare the obtained data with any existing studies. Although, it is possible to observe that the EMA-stress means are less than half of the possible range. In addition, it stands out that the average EMA academic stress experienced by students is higher than EMA non-academic stress. In the case of EMA-RNT content, the frequencies of brooding, reflection and worry content are low compared to the maximum possible, with reflection being the least frequent in the face of stress. In relation to the EMA-RNT process, duration, and intrusion are low in relation to the maximum possible, being the average of EMA-RNT process: duration, higher than that of EMA-RNT process: intrusion.

According to our data analysis strategy, seven hierarchical stepwise linear regression models were utilized to assess the power of the study’s variables to predict changes in SCL-90R symptoms. These models considered: Model 1: sociodemographic information; Model 2: added initial symptom severity (day one); Model 3: added RNT styles (brooding, reflection, and worry); Model 4: added term academic stress; Model 5: added EMA–academic and non-academic stress; Model 6: added EMA–RNT process (duration and intrusiveness); and EMA-RNT content (reactive brooding, reflection, and worry). Table 4 presents the fit data of the hierarchical regression models for depressive, anxiety, and somatization symptoms, considering R², change in R², AIC, and BIC.

In the case of the prediction of depressive symptoms, there are significant differences between models 1 and 2 (p = 0.001), model 3 and 4 (p = 0.002) and between 4 and 5 (p = 0.001), which would indicate the differential contribution of baseline symptoms, semester stress and week stress, respectively. When analyzing the lowest AIC in the different imputed bases, we can observe that the best model is 6 (Table 4), which adds to the baseline predictors, and the EMA-stress, the EMA-RNT process (R² = 0.62). Table 5 presents the coefficients of multiple regression model 6 for the prediction of depressive symptoms. Significant in this model are the regression coefficients: sociodemographic – housing = other relatives or friends, b = 0.19, p = 0.039, baseline depressive symptoms, b = 0.49, p = 0.001, and the coefficient of RNT-process: duration, b = 0.24, p = 0.019.

For the prediction of anxiety symptoms (Table 4), significant differences are observed between models 1 and 2 (p = 0.001) and between 4 and 5 (p = 0.002), which would indicate the differential contribution of incorporating baseline symptoms and EMA-stress, respectively. When analyzing the lowest AIC in the different imputed bases, we can observe that the best model is 5, which obtains the lowest BIC (R² = 0.67). This model adds EMA-stress to the baseline predictors of anxiety symptoms, RNT style and term academic stress. Of the regression coefficients of model 5 (Table 6), the coefficients of anxiety symptoms on day 1 are significant (b = 0.61, p = 0.001), as well as the coefficient of EMA non-academic stress (b = 0.29, p = 0.001).

When analyzing the models predicting somatization symptoms (Table 4), we can observe significant differences between models 1 and 2 (p = 0.001) and between 4 and 5 (p = 0.002), which would indicate a differential contribution of including baseline symptoms and EMA-stress, respectively. When analyzing the lowest AIC in the different imputed bases, we can observe that the best model is 6, where the EMA-RNT process is added to the four groups of predictors of baseline and EMA-stress, with an R² = 0.63. Reviewing the regression coefficients of model 6 (Table 7), the coefficient of somatization on Day 1 (b = 0.691, p = 0.001), and the coefficient of EMA-RNT: duration (b = 0.245, p < 0.001), are significant.

Table 8 contains the fit data of the hierarchical linear regression models of paranoia, psychoticism, and obsession symptoms, including R², change in R², AIC and BIC.

When comparing the seven hierarchical models for predicting paranoia symptoms (Table 8), significant differences exist between models 1 and 2 (p = 0.001) and between 5 and 6 (p = 0.003), which would indicate the differential contribution of adding the baseline symptoms and the EMA-RNT process, respectively. When analyzing the lowest AIC in the different imputed bases, we can observe that the best model is 6, also with the lowest BIC; this model adds to the baseline predictors and the EMA-stress, the EMA-RNT process (R² = 0.55). Regression coefficients for paranoia symptoms in model 6 are presented in Table 9. The coefficient of paranoia symptoms at baseline (b = 0.54, p = 0.001) and EMA-RNT process: duration (b = 0.25, p = 0.001) are significant.

In Table 8, there is a significant difference between models 1 and 2 (p = 0.001), between 4 and 5 (p = 0.006), and between 5 and 6 (p = 0.001) in predicting psychotic symptoms, which would indicate the differential contribution of adding the baseline symptoms, the EMA-stress, and the EMA-RNT process, respectively. Observing the AIC indices in the different imputed bases, we can see that the model with the lowest AIC is 7, which adds the EMA-RNT content to the predictors to the baseline predictors, EMA-stress and EMA-RNT process, with an R² = 0.59. The regression coefficients of model 7 are in Table 10, where the coefficient of psychotic symptoms on day 1 (b = 0.63, p = 0.001), and the coefficient of EMA-RNT duration (b = 0.19, p = 0.024), are significant.

When reviewing the fit data of the hierarchical regression models for the prediction of obsessive symptomatology (Table 8), we can observe significant differences between models 1 and 2 (p = 0.001) and between 4 and 5 (p = 0.001), which would indicate a differential contribution of including baseline symptoms and EMA – stress, respectively. When analyzing the lowest AIC in the different imputed bases, we can observe that the best model is 6, also with the lowest BIC. In this model, the EMA-RNT content is added to the baseline and EMA-stress predictors, with an R² = 0.58. Reviewing the regression coefficients of model 6 (Table 11), we see that the coefficient of obsessive symptoms on day 1 (b = 0.56, p = 0.001), and the coefficient of EMA-RNT duration (b = 0.30, p = 0.023), are significant.

Finally, Table 12 contains the model fit data of hostility and interpersonal sensitivity syndromes, including R², change in R², AIC, and BIC.

In the case of hostility symptoms (Table 12), there are significant differences in the predictive capacity between models 1 and 2 (p = 0.001), between 2 and 3 (p = 0.010), and between 4 and 5 (p = 0.001), which would indicate the differential contribution of including the baseline symptoms, the RNT styles, and the EMA-stress. When analyzing the lowest AIC in the different imputed bases, we can observe that the best model is 5, which adds the EMA-stress to the baseline predictors (R² = 0.48). When analyzing the regression coefficients of model 5 (Table 13), the coefficient of hostility symptoms on day 1 (b = 0.43, p = 0.001), and the coefficient of EMA-non-academic stress (b = 0.22, p = 0.001) are significant.

Observing the models for the prediction of interpersonal sensitivity symptoms (Table 12), significant differences are observed between models 1 and 2 (p = 0.001), which would indicate the differential contribution of incorporating the baseline symptoms. When analyzing the lowest AIC in the different imputed bases, we can observe that the best model is model 6, which adds the EMA-RNT process to the predictors to the baseline predictors, and EMA-stress, with an R² = 0.61. Table 14 presents the coefficients of regression model 6 for interpersonal sensitivity symptoms, where the coefficient of symptoms at baseline (b = 0.70, p = 0.001) and the coefficient of EMA-RNT duration (b = 0.28, p = 0.012) are significant.

This study has several strengths. Firstly, it uses EMA to assess the levels of academic and non-academic stress and RNT that students experience in their daily lives, combined with traditional measures of thinking styles and baseline cumulative academic stress. Additionally, the study evaluated both the contents of the reactive RNT – brooding, reflection, and worry – and the elements of the RNT process, duration, and intrusion. Finally, the study provides evidence about transdiagnostic processes in the Spanish-speaking population, which is usually studied in English speakers (Hall et al., 2021). Despite these contributions, this research is not free of limitations. One of them is that the results of the predictive models of psychotic symptoms may have been affected because, at the beginning of the study, participants with high scores in psychoticism and suicidal risk were excluded. Although this decision is due to reasons of ethical care of the participants, it must be considered when interpreting these results. Another element to consider is that the seven-day follow-up period is relatively short, although it has allowed positive results in similar studies (Connolly and Alloy, 2017; Newman et al., 2019; Hjartarson et al., 2021, among others). Lastly, while the hierarchical regression analysis used in this study allowed us to determine the contributions of different groups of predictors to the development of symptoms, it did not enable us to evaluate interactions or chains of mediation between them. Future studies in the field would benefit from analysis with structural equation modeling (SEM) models such as those used by Smith et al. (2021). Using an EMA design and SEM analysis, this study evaluated the interaction between momentary rumination, negative affect, and binge eating.

The results achieved are helpful for the prevention of mental disorders in university contexts. Academic and non-academic stress in the students’ lives influenced the increase in symptoms of six of the eight syndromes evaluated. A recent systematic review of stress management programs for college students (Amanvermez et al., 2023), including 46 universal prevention studies, showed that universal stress management programs have moderate effects in reducing perceived stress and a large effect in reducing cortisol levels. Indicating that stress management programs can reduce stress and prevent mental health problems in higher education. In this sense, a previous systematic review concluded that universal preventive programs for university mental health problems with proven effectiveness are those with supervised skills training above psychoeducational or unsupervised training (Conley et al., 2015). Another element that emerges from this study is the effect of excessive use of RNT when facing university life stress. The need to intervene preventively for RNT in university students is even greater than in other age groups since recent longitudinal studies have determined that RNT is higher in young adulthood than at any other time in life (Lilly et al., 2023). Furthermore, another systematic review of selective and indicated preventive programs for college students shows the benefits of cognitive-behavioral programs, which often use a transdiagnostic approach (Barnett et al., 2021). In both cases, the reviews point to the convenience of acting preventively in this group, both to prevent mental disorders and to promote adequate adaptation and well-being in this period of life. The results of this study point to the development of preventive intervention programs for those young people with high RNT. In a recent study involving college students with high trait RNT, participants received an RNT-focused intervention (smartphone app-based, 10 days, 15 min per day). They were then exposed to a standardized psychosocial stressor. Those in the intervention group reported less negative affect during the recovery phase. Additionally, they rated their ability to cope with the stressor as higher and reported less RNT during the recovery phase after the stressor (Funk et al., 2024). In the case of university students needing treatment, targeting RNT becomes a necessary element. A qualitative study (Rogiers et al., 2022) of participants in group interventions aimed at managing repetitive negative thinking (PNR) identified several mechanisms perceived as effective: increasing awareness of PNR, experimenting with exposure rather than avoiding stressful situations, living in the “here and now,” thinking in concrete terms and moving from a “mode of thought” to a “mode of action.” Participants reported that these strategies were helpful tools, suggesting that they should be integral components of treatment approaches to coping with RNT. In this sense, there are new brief online interventions (with and without therapist guidance) to reduce RNT and symptoms of anxiety and depression that are achieving promising results (Joubert et al., 2023) and may be helpful for this group.