3.1 Design

The study was designed as a prospective longitudinal study during CBT treatment under naturalistic conditions. All instruments were assessed at the beginning (pre) and end of treatment (post). The treatment took place at two independent German outpatient centers: Mental Health Research and Treatment Center of Ruhr-University Bochum (MHRTC) and the Center for Mental Health and Psychotherapy of Witten/Herdecke University (CMHP). Patients at MHRTC were treated between 2005 and 2010; patients at CMHP were treated between 2016 and 2019. All patients got written and verbal information about the treatment and evaluation of data in the research project and signed informed consent. General inclusion criteria for treatment were: (1) at least 18 years of age; (2) at least one psychiatric diagnosis according to DSM-IV criteria; and (3) sufficient German language skills. Inclusion criteria for analysis were as follows: (4) consensual completion of treatment; (5) individual setting; (6) available post score of at least one outcome variable; (7) at least 10 treatment sessions; (8) realistic scores in the strengths assessment (definition of unrealistic scores: a patient answered all items with either the minimum or the maximum answer). Figure 1 shows the flow chart of patient inclusion and exclusion as well as available data for all instruments.

3.2 Sample description

The total sample consisted of 202 patients (56.44% female, age: M = 42.49, SD = 13.97, range: 18–81). Of these, 127 patients were treated at MHRTC and 75 at CMHP. Mean treatment length was 29.86 sessions (SD = 14.39, range: 10–80). Primary diagnoses were mainly affective disorders (46.08%), anxiety disorders (38.73%), and adjustment disorders (5.88%). More than half the patients (55.94%) fulfilled the criteria for at least two disorders; on average patients had 1.43 diagnoses (SD = 0.81, range: 1–5). The study involved 54 therapists (75.93% female, age: M = 30.34, SD = 5.31, range: 24–51). All therapists had at least a master's degree in psychology. Both licensed CBT therapists and therapists in advanced CBT training (with at least 1 year of clinical experience) took part in the study. On average therapists treated 3.38 patients (SD = 2.92, range: 1–11).

The sample described above was compared to the patients who met exclusion criteria regarding several variables. Table 1 shows sample characteristics and comparisons with the patients meeting exclusion criteria. We found no significant differences in the variables compared, except for marital status. Subsamples from the two treatment centers were compared with regard to the same variables. We found no significant differences, except for two variables: patients at MHRTC showed significantly higher prescores of psychopathology (M₁ = 1.30, SD = 0.62; M₂ = 1.09, SD = 0.64; t(183) = 2.20, p = .03) and therapists in this center were significantly older (M₁ = 31.48, SD = 5.16; M₂ = 28.47, SD = 5.11; t(200) = 4.03, p < 0.001) than their counterparts.

3.3 Procedures at the outpatient centers

The study took place in two university outpatient training and research centers, in which adult patients with various mental disorders receive CBT treatment. Procedures in both centers were parallelized: personal and treatment-specific data of both patients and therapists were managed with a software called AmbOS. Patients interested in treatment first had a phone contact and subsequently a one-session consultation with a licensed therapist. Patients were then listed on an internal waiting list. Therapists contacted them chronologically to invite them to a five-session diagnostic phase, including the Structured Clinical Interview for DSM-IV (SCID; Wittchen et al., 1997). After the diagnostic phase, a CBT treatment according to the German health system began. The total duration of treatment was variable, oriented on individual symptoms and treatment goals (with a maximum of 80 sessions). No planned interventions to systematically use or foster patients' strengths were included in the study therapies (e.g., homework, solution-focused interventions).

3.4 Instruments

3.5 Statistical analysis

Analyses were conducted using R^©, version 3.5.3. Baseline characteristics of the included sample and patients who met exclusion criteria were compared using χ² tests for categorical variables and Student's t-tests for independent samples for continuous variables. The same procedures were conducted to compare patients and therapists from both treatment centers. To test the first hypothesis, paired t-tests with WIRF-pre and -post scores were conducted separately for each subscale (p = 0.05). Effect sizes (Cohen's d with pooled standardization for dependent measures according to Dunlap et al., 1996) as well as 95% confidence intervals (CIs) were calculated. Moreover, to compare possible effects of treatment length, the same procedure was respectively conducted in three subgroups that base on standards in the German health system: short length (<13 sessions, n = 9); moderate length (<25 sessions, n = 47); maximum lengths (25–80 sessions; n = 107). To test the second hypothesis, bivariate correlations (Spearman) between key variables were computed, while relevant correlation coefficients (WIRF-pre subscales associated with BSI-pre and IIP-pre) were compared using Fisher's z-tests for correlations from dependent samples. For the third hypothesis, multilevel regression analyses were conducted to predict treatment outcome. Multilevel modeling is considered the gold standard procedure to handle nested data structures (patients at Level 1 are nested with therapists at Level 2). It was further shown that multilevel modeling is able to counterbalance missing data that is often present in naturalistic psychotherapy designs (Kahn, 2011). In each analysis, one of the outcome variables at post (BSI, IIP, RSES) was defined as a criterion. Each analysis included the same four-step procedure: first, a base model (Model 0) with the random intercept of therapists (Level 2) without any further predictor was determined. In the second step, age, gender, and session number were entered as covariates (Model 1). In the third step, the pre-score of the respective criterion variable was entered (Model 2). At last, one WIRF-pre subscale was entered as a predictor (Model 3). Only WIRF-pre subscales that were significantly correlated with at least two outcome variables were considered in the regression analysis. All metric predictors (age, BSI-pre, IIP-pre, RSES-pre, CrisesS-pre, ProbS-pre) were centered at the group mean before inclusion. The magnitude of additional predictors was assessed by the change in Level 1 pseudo-R² according to Xu (2003). Furthermore, significance of each predictor was tested using Satterthwaite degrees of freedom. Model fit of competing models was tested with likelihood ratio tests for nested models and the Akaike Information Criterion (AIC).

4.1 Descriptive statistics and effect sizes

Table 2 shows the descriptive statistics for all measures. Pre-post comparisons of all WIRF subscales were conducted separately. Paired t-tests showed an increase in EvdayS scores over the course of treatment with a small to moderate effect size (t[161] = −5.38, p < 0.001; d = 0.44, 95% CI: [0.247–0.689]). No pre–post differences were found for CrisesS (t[61] = −1.73, p = 0.085; d = 0.14, 95% CI: [−0.07 to 0.366]), whereas ProbS scores increased over the course of treatment with a moderate effect size (t[162] = −8.96, p < 0.001; d = 0.70, 95% CI: [0.525–0.974]). No differences to these effects were found in the treatment length subgroups, except for EvdayS in the short length group showing no significant change during treatment, p = 0.425.

4.2 Correlation analysis

Bivariate correlations between WIRF-pre subscales and key variables were examined (see Table 2). None of the WIRF-pre subscales was correlated with BSI at the beginning of treatment, ps > 0.05. All WIRF-pre subscales showed significant negative correlations with IIP at the beginning of treatment. ProbS-pre was significantly correlated with RSES-pre, p = 0.03. Fisher's z-tests for dependent samples were conducted to examine differences in the correlation coefficients of the WIRF-pre subscales scores with IIP-pre. Both comparisons were not significant, z(199) = 1.085, p = 0.14 (EvdayS-pre, ProbS-pre, IIP-pre); z(199) = 1.186, p = 0.118 (CrisesS-pre, ProbS-pre, IIP-pre). Significant correlations between CrisesS-pre and all outcome post scores were found, ps < 0.01. ProbS-pre showed significant correlations with IIP-post and RSES-post, ps < 0.01.

4.3 Multilevel regression analysis

CrisesS-pre and ProbS-pre were included separately as predictors in the four-step multilevel regression analyses. Table 3 shows the results of the analyses with CrisesS-pre, while Table 4 shows the results with ProbS-pre as a predictor of outcome. The base model had only a random intercept at the therapists level. Results showed that 0% of the prediction of BSI-post and IIP-post, while 8% of prediction of the RSES-post could be attributed to differences between therapists. Model 1 with age, gender, and sessions as predictors was significant in the prediction of BSI-post, but not in any other analysis. Model 2, including prescores of criterion variables, was found to be significant in overall variance determination and change in pseudo-R² in all analyses, ps < 0.001. The inclusion of the CrisesS-pre in Model 3 led to additional increases in pseudo-R² as well as an improvement of the model fit in the prediction of BSI-post (χ²(1) = 6.292, p = 0.012), IIP-post (χ²(1) = 7.863, p = 0.005), and RSES-post (χ²(1) = 4.441, p = 0.035). CrisesS-pre was a significant predictor in all analyses, ps < 0.05.

ProbS-pre was a significant predictor of RSES-post (p = 0.01), but not of IIP-post when controlled for outcome variables' pre-scores (p = 0.12). The inclusion of ProbS-pre in model 3 led to an additional increase in pseudo-R² as well as an improvement of the model fit in the prediction of RSES-post (χ²(1) = 6.272, p = 0.012).

5.1 Implications

Our first hypothesis was confirmed with significant increases of strengths in everyday life and current problems in the course of treatment. Two interacting explanations are possible for these findings: CBT is a very practical treatment approach involving various techniques to learn new skills and thus add to patients' strengths. On the other hand, the activation of existing strengths was found to act as a change mechanism in psychotherapy (Gassmann & Grawe, 2006; Mander et al., 2013). This activation may lead to higher accessibility of positive and functional aspects and their use in a patient's life. Strengths are increasing particularly in the context of current problems (ProbS), which is to be expected. But there was also a significant increase in the context of everyday life (EvdayS)—this may be interpreted as a more general perception of strengths that patients developed during treatment, even though CBT mostly focuses on current problems. In contrast, strengths perceived in the management of previous crises (CrisesS) seem to be quite stable in the course of treatment. Maybe due to its focus on prior events, it is more resistant to the negativistic perception found in psychopathology.

Our second hypothesis concerning differential associations between strengths subscales and NMH respectively PMH-variables was not supported. All WIRF subscales showed significant negative associations with interpersonal problems, which is explicable as the instrument includes functional interaction patterns. Furthermore, WIRF subscales did not significantly differ in their association with interpersonal problems at the beginning of therapy. Other research has shown that interpersonal problems measured with the IIP are highly stable within individuals (Woodward et al., 2005). In this regard, a person's interpersonal profile may be a trans-situational construct that is less affected by the contextualization of the WIRF. In contrast to the findings of Victor et al. (2019), our results showed that pre-scores of psychopathology were not associated with any WIRF subscale. There is, therefore, indication that persons' with high burden caused by psychopathology may nevertheless be able to perceive their own strengths. We call for other in detail examinations of the relation of symptoms and strengths in psychotherapy patients, to analyze under what circumstances a perception of PMH constructs is enabled (Keyes, 2002).

Multilevel modeling was used to examine predictors of treatment outcome. The analysis of the base model indicated relevant therapist effects for patients' self-esteem, but not for psychopathology or interpersonal problems, in this sample. Baldwin and Imel (2013) showed that approx. 5%–8% of the variance of treatment outcome attributed to different therapists. A systematic review by Johns et al. (2019) further confirmed the relevance of therapist effects for treatment outcome, while on the other hand showing a large heterogeneity across studies and different outcome measures. The effect on posttreatment self-esteem in this study is comparable to average findings in these reviews. This result indicates that some therapists might use better implicit and explicit strategies to foster their patients' self-esteem. Possible associated strategies like the use of capitalization, attribution of mastery experiences, and structuring of the working alliance should be researched in future studies. Regarding our third hypothesis, we found that strengths in previous successful crises management (CrisesS) were significantly correlated with all outcome measures and incrementally predicted outcome. The additional variance determination by CrisesS-pre in all outcome variables is comparable to other nonclinical predictors shown in meta-analytic evidence (Levy et al., 2018; Vall & Wade, 2015). This strengths context may, therefore, function as a complementary predictor beyond patients' problems while at the same time not changing in the course of therapy. However, what may make the perception of strengths in previous crises relevant for treatment outcome? Persons with high CrisesS-pre scores may have meta-knowledge of helpful aspects as well as access to prior experiences of mastery or resilience in critical life events (Chmitorz et al., 2018; Davydov et al., 2010). Such a self-concept may be beneficial, as Flückiger et al. (2009) showed that in-session resource activation is mostly induced by patients speaking about their positive and functional aspects. A better knowledge of relevant strengths may lay the ground to better therapeutic starting points for resource activation in treatment. Concurrently, both therapists and patients naturally focus on negatives and problem solution in the therapeutic setting. A longitudinal diary study further showed that high levels of anxiety and depressive symptoms lead to limited reactivity to positive aspects/situations in everyday life (Carl et al., 2014). An explicit consideration and use of patients' strengths in therapy that goes beyond the simple identification of these aspects, therefore, seems necessary. Several strengths-based treatment approaches were successfully developed to increase in-session resource activation with specific techniques or priming (Franklin et al., 2017; Munder et al., 2019; Padesky & Mooney, 2012; Rashid, 2015). In addition to strength-based interventions, our results indicate that clinicians should focus on the assessment of patients' strengths from the beginning of treatment on. Self-reports may give important feedback, as they actively draw the patients' attention on their strengths. With the WIRF, strengths can be compared between the contexts within a patient to identify positive aspects otherwise possibly neglected due to the high problem focus.

Otherwise, our results also should be discussed critically because they indicate that patients with a higher pre-existing perception of own strengths benefit more from therapy (see YAVIS-concept; Schofield, 1986). From this point of view, some patients (with more subjective strengths) have better possibilities than others. This is an important point for our results, especially as CrisesS seem to be a stable variable during psychotherapy. However, recent research discussed resilience as a process factor rather than only a trait construct (Chmitorz et al., 2018). Therefore, more systematic and strengths-based work with patients' adversities in prevention and psychotherapy is needed.

Our results showed that strengths in the context of current problems (ProbS) were only predictive concerning patients' self-esteem at the end of treatment. Patients experience high burden because of their symptoms at the beginning of treatment. ProbS is directly associated with these problems and feelings of low manageability early on, which may overwhelm positive influences of this context on NMH during treatment. Nonetheless, ProbS were a specific predictor of the positive outcome variable in this study, giving evidence for the differentiation of PMH and NMH constructs proposed in the dual-factor model of mental health (Iasiello et al., 2020; Keyes, 2002). In line with our results, Bos et al. (2016) showed that currently perceived strengths might have a specific predictive value for PMH in persons with high psychopathology.

5.2 Limitations

This study has several limitations. Data of the study were collected under naturalistic conditions without a priori inclusion criteria. This may have led to bias in the sample or overestimation of effects. Furthermore, the naturalistic design led to missing data in various variables. Although the statistical approach of multilevel modeling is quite robust to missing data, a more controlled design is needed in future studies. The increase of strengths over time found in our study might not causally arise from therapy effects and should be tested in a waitlist-controlled examination. Moreover, the outcome variables were only assessed at the beginning and ending of therapy. The absence of repeated measures aggravates that specific slopes in these variables could be determined and facilitates regression to the mean. We only included completers in the analysis, which is a common approach in prediction studies (Amati et al., 2018). However, possible moderation effects of completion/drop-outs cannot be completely excluded. The conceptualization of PMH in dual-factor models of mental health includes a wide range of aspects. Our operationalization of PMH via the WIRF and the RSES encompasses only part of this construct. More sufficient instruments like the Mental Health Continuum-Short Form (Keyes et al., 2008) or the PMH-scale (Lukat et al., 2016) should be used as process and outcome measures in future clinical studies. In addition, strengths were only assessed via self-report. Finally, replications in other contexts and by other researchers are needed.

5.3 Future directions

Based on our findings the connection of self-perceived strengths and treatment outcome should be explored more deeply. Prospective studies may focus on possible moderation effects of pretherapy strengths on treatment processes leading to better outcomes. Observer ratings and session questionnaires may be used to examine whether patients with higher self-perceived strengths show more starting points for resource activation in psychotherapy. Other analyses may investigate whether characteristics of patients (disorder subgroups, personality traits, etc.) are associated with differential strengths perception during treatment. In addition, more frequent assessments during therapy may contribute to a deeper understanding of changes in self-perceived strengths and its connection to other treatment processes. Deliberate training for psychotherapists targeting the perception of patients' strengths and mastery experiences should be developed and examined (cf. Willutzki & Teismann, 2013).