Department of General Internal Medicine and Psychosomatics, Heidelberg University, Heidelberg, Germany.
Find articles by Ulrike DingerSimone Jennissen, Department of General Internal Medicine and Psychosomatics, Heidelberg University, Heidelberg, Germany;
Correspondence concerning this article should be addressed to Simone Jennissen, Department of General Internal Medicine and Psychosomatics, Heidelberg University, Thibautstr. 4, 69115 Heidelberg, Germany; ed.grebledieh-inu.dem@nessinneJ.enomiS
The publisher's final edited version of this article is available at J Couns PsycholThis study aimed to investigate change in insight into maladaptive interpersonal patterns over the course of psychotherapy, as well as the specificity of insight as a change mechanism in dynamic treatments. A total of 100 patients received up to 16 sessions of either cognitive or dynamic therapy for major depressive disorder in a randomized clinical trial. Assessments of insight (Insight into Conflictual Relationship Patterns scale) and depression severity (Hamilton Depression Inventory) took place at the beginning of treatment, at month 2, and month 5. Patient insight significantly improved over the course of dynamic treatments. Gains in insight from the beginning to month 2 of treatment were a significant predictor of decreases of depressive symptoms from month 2 to month 5 of treatment in the dynamic, but not in the cognitive treatment group, despite a nonsignificant interaction. Results provide support for insight as a change factor in dynamic therapies. Better self-understanding of dysfunctional interaction patterns could help patients to find more adaptive ways of behaving, to form more satisfying relationships, and to improve their depression.
Keywords: insight, self-understanding, psychotherapy process, major depressive disorder, mechanism of change
Insight is regarded as a key mechanism of change in focal psychodynamic psychotherapies. In distinct contrast to Freud’s original model that focused on uncovering the patient’s repressed traumatic memories (Freud & Breuer, 1895), and in contrast to the later ego psychology of working through defenses (Freud, 1923), brief focal dynamic therapies highlight insight into interpersonal patterns in the patients’ life, as well as in the relationship with the analyst, as curative components of dynamic therapy (Hirsch, 1998; Luborsky, 1984). These therapies assume that when patients gain insight into their maladaptive interpersonal patterns, they are able to confront their difficulties and find more adaptive ways to interact with others, which eventually improves their relationships and reduces psychological distress (Gabbard, 2014; Lemma et al., 2013; Strupp & Binder, 1984). In these treatments, insight is understood as a change mechanism that is part of the psychotherapy process (Crits-Christoph, Connolly Gibbons, et al., 2013; Ulberg et al., 2017). Other conceptualizations define insight as the state of a sudden “Aha!” experience or the trait-like ability to gain such understanding, which has also been termed insightfulness (Connolly Gibbons et al., 2007). Besides being considered a change mechanism, object relations theory and self psychology introduced the viewpoint of insight as an indicator of good outcome (De Smet et al., 2020; Messer & McWilliams, 2007). This paper is focused on insight into maladaptive interpersonal patterns as a potential mechanism of therapeutic change.
Various definitions of change mechanisms and criteria to put these to an empirical test have been proposed (e.g. Cuijpers et al., 2019; Doss, 2004; Kazdin, 2005, 2007; Ulberg et al., 2017). Following the definitions of Doss (2004) and Crits-Christoph et al. (2013), we consider the change mechanism to consist of events that occur within the patient and bring about symptom change. In theory, a mechanism details how therapist interventions change patients’ experiences and how these changes result in symptom improvements (Kazdin, 2007). Regarding insight into maladaptive interpersonal patterns, a depressed patient might report on a recurring experience with their partner, where they long for the acceptance and care of their significant other, but always seem to get rejected, which leads to further withdrawal and self-criticism of the patient. Over the course of treatment, this patient might understand that based on experiences with neglectful caregivers, they have developed a general cautiousness and negative expectations when interacting with others. The therapist’s mirroring, empathic confrontation or disclosure of countertransference may help this patient realize that they often come across as disliking others, to which others respond by turning away. Interpretations of this behavior as a means of protection might enhance the patient’s self-compassion and enable them to slowly modify their behavior towards others. This in turn could lead to new, more positive experiences in their “real-world” relationships, relieve them from the feeling of helplessness and improve their mood.
There are several requirements to put this proposed mechanism to an empirical test. First, insight into maladaptive interpersonal patterns must be assessed repeatedly during treatment. For the most proximal assessment of the targeted mechanism, the assessment should employ a process perspective and focus on what happens during therapy sessions. Assessments of the trait-like capacity for insight (i.e. insightfulness) are more suitable as a baseline predictor or long-term outcome (e.g. Høglend et al., 1994; Kivlighan et al., 2000). Using the repeated assessments, change in insight over the course of treatment can be investigated (Ulberg et al., 2017). Next, the association with outcome must be examined. First, concurrent change of insight and outcome is investigated to assess the covariation of both variables. Next, the temporal sequence is examined to assure that improved insight is not simply a by-product of symptom improvement. To be an agent of therapeutic change, gains in insight must predict later symptom improvements (Ulberg et al., 2017). Lastly, to assess the specificity of the proposed mechanism, the aforementioned steps must be performed in and compared between the treatment that targets the mechanism and an alternative treatment condition (Cuijpers et al., 2019; Kazdin, 2005, 2007). Showing that patients in dynamic treatments gain insight into interpersonal patterns which in turn predicts symptom improvements would speak in favor of the mechanism but would not allow conclusions on whether this is specific to dynamic treatments or a common by-product of psychotherapy. Introducing an alternative treatment condition which targets different mechanisms allows assessing whether insight into interpersonal patterns is a mechanism that is specifically evoked in dynamic treatments or whether it occurs even when it’s not directly targeted.
Empirical research on insight is relatively rare, as shown by a recent meta-analysis that found 22 empirical studies reporting on the quantitative association between insight and psychotherapy outcome and estimated a moderate and significant association (r = .31 Jennissen et al., 2018). Five studies concurrently investigated insight and outcome and found moderate to large associations within sessions (Ambühl, 1993; Cogan & Porcerelli, 2013; Diemer et al., 1996; Kolden, 1991), with one exception of a moderate but nonsignificant finding (Hoffart & Sexton, 2002). The remaining studies either established a timeline, investigated change scores, or both. Eight studies associated changes in insight with later changes in outcome and thus provided a more rigorous test of the potential change mechanism diminishing the risk of reverse causation. The results of these studies are mixed, with effect sizes varying from r = .01 (Connolly et al., 1999) to r = .59 (Johansson et al., 2010). Both within dynamic and cognitive treatments, some studies show significant relations between changes in insight and subsequent changes in outcome (dynamic: Connolly Gibbons et al., 2009; Grande et al., 2003; Høglend et al., 1994; Johansson et al., 2010; Kallestad et al., 2010) (cognitive: Nyklíček et al., 2010) and some show insignificant associations (dynamic: Connolly et al., 1999; Luborsky et al., 1988) (cognitive: Kallestad et al., 2010). Only one study was conducted as an RCT comparing cognitive and dynamic treatments (Kallestad et al., 2010). The study on a sample of N = 49 patients with Cluster C personality disorders demonstrated that gains in insight from session 6 to session 36 predicted two-year improvements in symptoms and interpersonal functioning for patients who received short-term dynamic therapy, but not for patients in cognitive therapy.
Due to only one study so far with a limited sample size, highly specific study population, and no formal test of an interaction with treatment, further research is necessary to determine whether these findings will hold in different samples and treatment settings and assess whether treatments differ significantly. Specifically, we intended to replicate the findings in a larger sample of patients with major depression. Depression is the leading cause of disability with over 300 million people being affected worldwide according to the World Health Organization (WHO). Both dynamic and cognitive therapies have demonstrated their efficacy and effectiveness in the treatment of depression, but may operate through different mechanisms (Cuijpers et al., 2020; Driessen et al., 2015). In theory, cognitive therapy (CT) for depression considers depressogenic cognitions as responsible for the development of depressive symptoms (Beck et al., 1979). Treatment is focused on altering these cognitions, developing compensatory skills to suppress depressogenic cognitions, and engaging in behavioral activation to overcome motivational anhedonia. Supportive-expressive dynamic psychotherapy (SE) on the other hand aims at increasing patients’ insight into their maladaptive interpersonal patterns, their association with past experiences and current symptoms, and the patients’ ability to change these patterns and engage in more adaptive interactions (Luborsky, 1984). Thus, CT and SE target distinguishable change mechanisms and may achieve similar therapeutic effects through different pathways.
To fill the aforementioned gaps in research, we investigated insight as a potential mechanism of change and its specificity to SE and CT for depression at a community mental health center (CMHC; Connolly Gibbons et al., 2016). The CMHC provides an ecologically valid setting that is highly generalizable to “real-world” treatment conditions while randomized allocation to CT vs. SE maximizes the internal validity of conclusions regarding the specificity of the change mechanism. The parent trial found that SE was not inferior to CT in changing depressive symptoms (Connolly Gibbons et al., 2016). The cognitive mechanisms of change have previously been investigated in the CMHC database and support the compensatory skills model of CT (Barber & DeRubeis, 1992), whereby increased positive compensatory skills, such as the ability to generate alternative explanations for upsetting events, subsequently improve symptoms in CT, but not in SE (Crits-Christoph et al., 2017). The current study sought to investigate insight into maladaptive interpersonal patterns as a specific mechanism of change in dynamic psychotherapy. Based on previous results supporting concurrent changes between insight and outcome (e.g. Cogan & Porcerelli, 2013; Diemer et al., 1996) and changes in insight as a significant predictor of subsequent changes in outcome (e.g. Connolly Gibbons et al., 2009; Høglend et al., 1994; Johansson et al., 2010) as well as first tentative results suggesting that these findings hold in dynamic, but not cognitive treatments (Kallestad et al., 2010), we hypothesized (a) that insight changes significantly more over the course of SE than CT, (b) that insight and depressive symptoms change concurrently in SE but not in CT, and (c) that increased insight improves subsequent outcome in SE but not in CT.
The Institutional Review Board of the second and third authors’ university approved all study procedures. Participants provided written informed consent prior to participation. Readers interested in a thorough discussion of methods are referred to the main publication (Connolly Gibbons et al., 2016).
Participants were recruited at a CMHC in Pennsylvania. Patients seeking outpatient treatment during the study period were screened for depressive symptoms using the Quick Inventory for Depressive Symptomatology (QIDS; Rush et al., 2003). Patients between the ages of 18 and 65 and a QIDS score of 11 or above were referred to the research staff for a brief phone screen and potentially eligible patients were scheduled for a baseline assessment at the CMHC. At baseline, patients completed several baseline self-report measures and were interviewed with the Structured Clinical Interview for DSM-IV Axis I (SCID-I; First et al., 1996) and the Hamilton Depression Inventory (HAM-D; Hamilton, 1960). Patients were included in the study if they met criteria for major depressive disorder (MDD) and did not have (a) diagnosis of bipolar disorder, (b) current or past diagnosis of schizophrenia, psychosis, MDD with psychotic features, or seizure disorder, (c) depression due to organic pathology, (d) substance/alcohol abuse requiring immediate referral to substance abuse treatment, (e) referral to partial hospital, or (e) suicidal thoughts judged by the clinic to require more intensive services.
Of the 3,951 patients assessed for eligibility, 1,110 passed the initial QIDS screening and were referred to the research staff for a brief phone screen. Of those, 581 passed the phone screen and completed a baseline assessment at the CMHC. At this stage, 344 patients were excluded (116 used as training case, 100 past or current diagnosis of schizophrenia, psychosis, MDD with psychotic features, seizure disorder, or depression due to organic pathology, 78 no current diagnosis of MDD, 50 due to other reasons) A total of 237 patients met the inclusion criteria and were randomized to treatment (118 SE, 119 CT). This study included patients who attended at least two sessions of psychotherapy and completed assessments at baseline, month 2, and month 5. The resulting study sample consisted of N = 100 patients (54 SE, 46 CT). Patients received weekly sessions of psychotherapy for a maximum of 16 sessions or 5 months (whichever occurred first). Sessions were audiotaped. Patients were assessed with the HAM-D as the primary outcome measure and several self-report measures at baseline, month 1, month 2, month 4, and month 5. This study used HAM-D assessments at baseline, month 2, and month 5. Patients were invited to the monthly assessments regardless of how many sessions they had attended.
The study sample included patients between the ages of 18 and 64 (M = 39.6, SD = 12.5). The majority of patients were female (80%) and not currently in a long-term relationship (60%). Nineteen percent of patients were married or living with a partner. About half of the patients (51%) were members of a minority group (42% black or African American, 2% American Indian or Alaska native, 1% Asian, 1% Native Hawaiian or Pacific Islander, and 5% other). A majority of patients reported a high school degree or less as their highest level of education (57%) and were unemployed at the time of enrollment (56%).
Treatment was provided by clinicians with at least master’s degrees who were employed by the CMHC. Therapists were recruited at the clinic and allocated to treatment group based on their previous training and supervision, theoretical orientation, and preference for either CT or SE. All therapists participated in an initial 8 hr training workshop led by a supervisor with at least 10 yrs of experience in conducting and supervising either CT or SE. Therapists received intensive individual supervision of 1 hr after every 2 hrs of therapy for their first three cases. Therapists were eligible to treat patients after at least 8 hrs of treatment with at least two training cases. Throughout the study period, all study therapists received one hour of group supervision twice a month.
The SE treatment was delivered following the manual for supportive-expressive psychodynamic psychotherapy (Luborsky, 1984) and the supplemental clinical case manual (Book, 1998). The treatment is focused on building a strong therapeutic alliance to facilitate insight into maladaptive interpersonal patterns. It begins with supportive techniques to develop a therapeutic relationship, familiarize the patient with the focus on relationship difficulties, and set goals to explore a specific, currently problematic interpersonal pattern. The therapist then uses expressive techniques such as clarifications, confrontations, and interpretations to increase the patient’s self-understanding of maladaptive interpersonal patterns and develop alternative ways of responding. Interpersonal patterns are formulated using the Core Conflictual Relationship Theme (CCRT; Luborsky et al., 1994) model. The model identifies the patient’s wishes/needs in the relationship with another person, the other person’s typical response, and the patient’s subsequent stereotypic response.
The CT treatment followed the manuals for cognitive therapy for depressive disorders (Beck, 1970; Beck et al., 1979). Treatment consists of a series of structured sessions focused on behavioral activation and a cognitive approach to modify negative automatic thoughts that are assumed to cause and maintain depressive symptoms. Standard interventions are activity scheduling, using thought records to identify and evaluate automatic thoughts, and behavioral experiments. Further in the treatment process, the focus shifts towards examining underlying dysfunctional attitudes and beliefs.
To evaluate treatment fidelity, both CT and SE were assessed for adherence and competence by four independent advanced graduate student judges who were blind to research design, settings, and interventions. A community-adapted version of the Penn Adherence/Competence Scale for Supportive-Expressive Dynamic Psychotherapy (Barber & Crits-Christoph, 1996) was used to evaluate fidelity to SE. The CT subscale of the Collaborative Study Psychotherapy Rating Scale (Hill et al., 1992) and the Cognitive Therapy Scale (Vallis et al., 1986) were used to assess adherence and competence to CT, respectively. Both CT and SE were delivered with adequate fidelity (Connolly Gibbons et al., 2016).
Insight was rated based on audio recordings of psychotherapy sessions. Since we aimed to examine change in insight from the beginning to the middle of treatment as a predictor of change in depressive symptoms from the middle to the end of treatment, insight assessments were matched with the HAM-D assessments at baseline, month 2, and month 5. Change in insight from the beginning to the middle of treatment was assessed using audio recordings from session one and the session that took place closest to and prior to the month 2 assessment. This procedure ensured that change in HAM-D from month 2 to month 5 would always be subsequent to the insight assessments. Furthermore, we intended to explore the maximum change in insight from the beginning to the end of treatment. Because the last session of psychotherapy often differs from the preceding therapeutic process due to wrapping up and saying goodbye to each other, we decided to rate insight at the end of treatment in the second to last session attended.
A total of three advanced graduate students with a B.Sc. or M.Sc. degree in psychology completed the insight observer assessments based on session audio recordings using the Insight into Conflictual Relationship Patterns (ICR; Jennissen et al., 2020) scale. The three judges first familiarized themselves with the scale and manual use and practiced by jointly discussing insight ratings for audiotapes that were not part of the study sample for about 10 hrs. Next, the judges independently rated insight on new training tapes until high interrater reliability (ICC(3,1; (Shrout & Fleiss, 1979) ≥ .90) was achieved after an additional 10 hrs. The judges then began with their independent ratings of study session recordings. Session recordings were randomly assigned to raters and ratings were completed in randomized order. Raters were blinded to treatment condition. During the coding phase, routine calibrations were performed after every tenth rating by jointly discussing ratings that had already been done independently. Independent ratings were not altered after these discussions. For a total of 30 sessions (16 CT and 14 SE), independent ratings were performed by all three raters to calculate their interrater reliability in the coding phase. Interrater reliability for the three independent raters was excellent across both treatment conditions (ICC(3,1)CT = .84, ICC(3,1)SE = .79).
The QIDS-SR (Rush et al., 2003) is a 16-item self-report measure of depressive symptom severity. Patients indicate the severity of their impairment across nine domains of depression on a 4-point scale. The scale has demonstrated good reliability and high convergent validity with the HAM-D in a sample of patients with chronic MDD (Rush et al., 2003). Internal consistency in the present study was Cronbach’s α = .58.
The SCID-I (First et al., 1996) is a semi-structured interview to assess and diagnose mental disorders according to the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV, American Psychiatric Association, 2000). Trained diagnosticians can reliably administer and code the SCID-I for DSM-IV (Lobbestael et al., 2011; Zanarini & Frankenburg, 2001).
The ICR scale (Jennissen et al., 2020) is an observer rating scale based on audio- or videotape recordings of psychotherapy sessions. It measures patients’ understanding of conflicting interpersonal interests, their associations with past experiences and psychological symptoms as well as the defensive function and the ability to change the maladaptive relationship pattern (Connolly Gibbons et al., 2007; Crits-Christoph, Gibbons, et al., 2013). Trained observers rate insight on a 5-point scale using twelve items, where higher scores indicate more insight. The scale has demonstrated high interrater and retest reliabilities as well as adequate convergent and discriminant validity (Jennissen et al., 2020). Internal consistency in the present study was Cronbach’s α =.81.
The HAM-D (Hamilton, 1960) is a measure of depressive symptom severity administered by trained clinical evaluators. Depressive symptoms during the past two weeks are assessed using 17 items with different scaling formats that are converted into a global score. Higher scores indicate more severe depressive symptoms. Meta-analytic results showed good internal consistency and interrater reliability of the scale (Trajkovi et al., 2011). Internal consistency in the present study at month 5 was Cronbach’s α = .78.
Prior to the analyses, we calculated propensity scores to use as a covariate in later regression models. Previous analyses of the dataset (Connolly Gibbons et al., 2019) had shown that differential dropout and treatment outcome for the two treatment groups were predicted by several baseline covariates (age, education, race, physical functioning, psychotic symptoms, drug use, trauma history, emotional lability, interpersonal problems, and quality of life). To control for selection bias, propensity scores were calculated based on these variables and adjusted for in subsequent analyses (D’Agostino, 1998).
First, we assessed changes in insight over the course of treatment in both CT and SE. To account for the nested data structure, we employed a multilevel modeling strategy (MLM). The unconditional model including therapist as a random effect did not converge and yielded a variance estimate of 0 at the therapist level. Including only the patient as a random effect, the model estimated meaningful variability between patients (ICC = .17). We thus modeled a 2-level hierarchical structure (measurements nested within patients) predicting insight from the session number of insight assessment.
Next, we examined concurrent change in insight and outcome across treatment conditions. Two variables were computed specifying changes (i.e. difference scores) in insight and depression severity each from beginning to month 2 and beginning to month 5. We fit a model simultaneously examining changes from baseline to each of the two following assessments. Recent research suggests that the reliability concerns often associated with difference scores are less of an issue when standard deviations differ between measurement occasions (Gollwitzer et al., 2014; Trafimow, 2015). This is the case in most treatment studies where treatment is expected to affect subjects differently. Again, meaningful variability was only found at the patient level (ICC = .51), but not at the therapist level. Change in depression severity was thus predicted from concurrent change in insight in a 2-level hierarchical model.
Lastly, we investigated the sequence of change in insight and subsequent changes in outcome. We specified a linear regression model predicting change in depression severity from month 2 to month 5 from change in insight from beginning to month 2 of treatment.
All models were estimated in a stepwise procedure, introducing main effects as a first step, adjusting for propensity scores and where appropriate, treatment length, as a second step, and testing for specificity by including interaction terms with treatment type as a third step. Treatment type was dummy coded such that values of 0 represent SE and values of 1 represent CT.
Models were estimated using the Maximum Likelihood (ML) estimator. Partial correlation coefficients according to the formula by Lipsitz, Leong, Ibrahim, and Lipshultz (2001) were computed as a measure of the standardized association of each predictor and the outcome variable, controlling for all other model covariates. Analyses were performed using the R packages lme4 (version 1.1–23) and lmerTest (version 3.1–2) in R version 3.6.1.
Table 1 presents means and standard deviations for the HAM-D and ICR at each measurement point. At baseline, there were no significant differences between the treatment groups on the HAM-D, t(98) = 1.3, p = .18, and ICR measure, t(98) = 0.9, p = .39.
Means and standard deviations for insight and depression severity at each measurement point by treatment type
| Measure | Baseline | Month 2 | Month 5 | |||
|---|---|---|---|---|---|---|
| M (SD) | N | M (SD) | N | M (SD) | N | |
| ICR | ||||||
| cognitive | 0.75 (0.32) | 46 | 0.88 (0.40) | 46 | 0.81 (0.36) | 38 |
| dynamic | 0.80 (0.34) | 54 | 1.02 (0.39) | 54 | 1.01 (0.45) | 42 |
| HAM-D | ||||||
| cognitive | 20.11 (5.49) | 46 | 14.96 (6.60) | 46 | 13.61 (7.58) | 46 |
| dynamic | 21.63 (5.73) | 54 | 20.02 (5.93) | 54 | 17.03 (7.86) | 54 |
Note. ICR = Insight into Conflictual Relationship Patterns scale; HAM-D = Hamilton Depression inventory. For the ICR assessments, baseline represents the first session, month 2 represents the session closest before the month 2 assessment, and month 5 represents the second to last session attended.
Table 2 provides the summary of models predicting insight over the course of treatment. Higher session numbers predicted higher insight scores, even after propensity score adjustment. Introducing an interaction term with treatment, the model continued to demonstrate significantly higher insight scores over time, with a marginally significant interaction between session number and treatment type (see Figure 1 ). Because the interaction term is included in the model, the significant effect of session number represents the conditional effect of a 1-unit change in session number when the treatment type is 0 (Brambor et al., 2006; Braumoeller, 2004). Because the dummy code for treatment type was 0 for SE and 1 for CT, the conditional main effect for session number represents the effect in SE treatments, meaning that increasing session numbers were significantly associated with higher insight in dynamic therapies. The partial correlation for the effect of session number on insight controlling for all other variables in the final model was r = .20. As a cross-check, we conducted a model with reverse dummy coding of treatment type (i.e. CT = 0, SE = 1). In this model, the conditional effect of change in insight on subsequent changes in depression was nonsignificant, b = 0.005, SE = 0.009, p = .60, with a partial correlation coefficient of r = .05. Thus, there was no significant effect of session number on insight in cognitive treatments.
Interaction plot illustrating the effect of progressing session numbers on insight according to Model 4, Table 2
Note. Insight was assessed using the Insight into Conflictual Relationship Patterns (ICR) scale.
Fixed and random effects estimates for a series of two-level models predicting insight over the course of therapy
| Parameter | Model 1 a | Model 2 b | Model 3 c | Model 4 d | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Coefficient (SE) | 95% CI | p | Coefficient (SE) | 95% CI | p | Coefficient (SE) | 95% CI | p | Coefficient (SE) | 95% CI | p | |
| Fixed Effects | ||||||||||||
| Intercept | 0.88 (0.003) * | 0.83 – 0.93 | .001 | 0.88 (0.003) * | 0.83 – 0.93 | .001 | 0.88 (0.03) * | 0.83 – 0.93 | .001 | 0.94 (0.04) * | 0.87 – 1.01 | .001 |
| Session Number | 0.02 (0.0005) * | 0.01 – 0.03 | .002 | 0.02 (0.0006) * | 0.00 – 0.03 | .006 | 0.02 (0.006) * | 0.00 – 0.03 | .005 | 0.02 (0.007) * | 0.01 – 0.04 | .001 |
| Propensity Score | −0.08 (0.13) | −0.33 – 0.16 | .510 | 0.03 (0.13) | −0.23 – 0.29 | .018 | ||||||
| Treatment Type | −0.13 (0.06) * | −0.24 – (−0.02) | .827 | |||||||||
| Session Number * Treatment Type | −0.02 (0.01) † | −0.04 – 0.00 | .076 | |||||||||
| Error Variance | ||||||||||||
| Residual | 0.12 (0.35) | 0.10 (0.32) | 0.10 (0.32) | 0.10 (0.32) | ||||||||
| Intercept | 0.02 (0.15) | 0.03 (0.17) | 0.03 (0.17) | 0.02 (0.16) | ||||||||
| Slope (Session Number) | 0.0007 (0.03) | 0.0008 (0.03) | 0.0007 (0.03) | |||||||||
| Model Fit | ||||||||||||
| AIC | 256.36 | 254.09 | 251.69 | 248.06 | ||||||||
| BIC | 270.90 | 275.90 | 277.06 | 280.67 | ||||||||
Note. N = 100; patient ICC = 0.17; parameter estimates based on Maximum Likelihood estimations in lme4, significance values based on lmerTest. Level-1 predictors were centered prior to the analysis.
Predicting change in depressive symptoms from concurrent change in insight, baseline depression severity was the only consistently significant negative predictor (see Table 3 ). Patients with higher levels of depression at baseline improved more over the course of treatment.
Fixed and random effects estimates for a series of two-level models predicting change in depression severity from concurrent change in insight
| Parameter | Model 1 a | Model 2 b | Model 3 c | ||||||
|---|---|---|---|---|---|---|---|---|---|
| Coefficient (SE) | 95% CI | p | Coefficient (SE) | 95% CI | p | Coefficient (SE) | 95% CI | p | |
| Fixed Effects | |||||||||
| Intercept | 6.11 (2.68) * | 0.87 – 11.36 | .022 | 10.71 (3.41) * | 4.04 – 17.38 | .002 | 11.24 (3.41) * | 4.55 – 17.93 | .001 |
| Baseline HAM-D | −0.54 (0.10) * | −0.75 – (−0.34) | .001 | −0.64 (0.11) * | −0.85 – (−0.43) | .001 | −0.64 (0.11) * | −0.85 – (−0.43) | .001 |
| Baseline ICR | 0.92 (1.98) | −2.97 – 4.81 | .642 | 0.29 (1.98) | −3.59 – 4.16 | .885 | −0.15 (1.97) | −4.02 – 3.72 | .940 |
| Change in ICR | 1.66 (1.23) | −0.76 – 4.07 | .178 | 1.41 (1.23) | −0.99 – 3.82 | .248 | 1.23 (1.54) | −1.79 – 4.25 | .424 |
| Propensity Score | −7.68 (2.83) * | −13.22 – (−2.14) | .007 | −5.86 (3.02) † | −11.77 – 0.05 | .052 | |||
| Treatment length | 0.15 (0.14) | −0.12 – 0.43 | .280 | 0.14 (0.14) | −4.42 – 0.57 | .298 | |||
| Treatment Type | −1.93 (1.27) | −4.42 – 0.57 | .130 | ||||||
| Change in ICR * | −0.38 (2.22) | −4.73 – 3.97 | .864 | ||||||
| Treatment Type | |||||||||
| Error Variance | |||||||||
| Residual | 29.79 (5.46) | 29.92 (5.47) | 29.75 (5.46) | ||||||
| Intercept | 17.11 (4.14) | 14.15 (3.76) | 13.57 (3.68) | ||||||
| Model Fit | |||||||||
| AIC | 1204.03 | 1186.40 | 1187.82 | ||||||
| BIC | 1223.19 | 1211.85 | 1219.63 | ||||||
Note. N = 100; patient ICC = 0.51; parameter estimates based on Maximum Likelihood estimations in lme4, significance values based on lmerTest. HAM-D = Hamilton Depression inventory, ICR = Insight into Conflictual Relationship Patterns scale.
Testing whether changes in insight from the beginning to month 2 of treatment predicted subsequent decreases of depression severity, change in insight was a marginally significant negative predictor of subsequent change in outcome, even when controlling for propensity scores and treatment length (see Table 4 ). When interaction terms with treatment type were introduced, gains in insight emerged as a significant predictor of subsequent decreases of depression severity with a nonsignificant interaction between gains in insight and treatment type. Increases in insight were associated with significant subsequent decreases in depressive symptoms for dynamic treatments (see Figure 2 ). The partial correlation for the effect of gains in insight on subsequent reductions of depression controlling for all other variables in the final model was r = −.21 1 . Again, we ran a cross-check model with reverse dummy coding of treatment type (i.e. CT = 0, SE = 1). In this model, the conditional effect of change in insight on subsequent changes in depression was nonsignificant, b = −0.38, SE = 2.50, p = .88, with a partial correlation coefficient of r = −.02. Thus, there was no significant effect of changes in insight on subsequent changes in depression severity in cognitive treatments.
Interaction plot illustrating the effect of change in insight from the beginning to month 2 of treatment on symptom change from month 2 to month 5 according to Model 3, Table 4
Note. Insight was measured by the Insight into Conflictual Relationship Patterns (ICR) scale, depressive symptoms were assessed using the Hamilton Depression Inventory (HAM-D). Change scores were obtained by subtracting earlier from later scores on the ICR or HAM-D, respectively.
Linear regression model predicting change in depression severity from month 2 to month 5
| Parameter | Model 1 a | Model 2 b | Model 3 c | ||||||
|---|---|---|---|---|---|---|---|---|---|
| Coefficient (SE) | 95% CI | p | Coefficient (SE) | 95% CI | p | Coefficient (SE) | 95% CI | p | |
| Intercept | −2.13 (3.34) | −8.77 – 4.50 | .525 | −0.72 (4.31) | −9.29 – 7.85 | .868 | −0.09 (4.37) | −8.77 – 8.58 | .983 |
| Baseline HAM-D | 0.16 (0.13) | −0.09 – 0.41 | .209 | 0.18 (0.14) | −0.09 – 0.45 | .192 | 0.16 (0.14) | −0.11 – 0.43 | .233 |
| Baseline ICR | −3.72 (2.49) | −8.66 – 1.23 | .139 | −3.30 (2.51) | −8.49 – 1.88 | .209 | −2.86 (2.61) | −8.05 – 2.32 | .276 |
| Change in ICR from baseline to month 2 | −3.23 (1.89) † | −6.98 – 0.53 | .091 | −3.20 (1.91) † | −7.00 – 0.61 | .098 | −5.57 (2.60) * | −10.74 – (−0.40) | .035 |
| Propensity Score | −0.98 (3.58) | −8.11 – 6.13 | .784 | −2.78 (3.85) | −10.43 – 4.87 | .473 | |||
| Treatment length | −0.17 (0.17) | −0.51 – 0.71 | .317 | −0.18 (0.17) | −0.52 – 0.16 | .287 | |||
| Treatment Type | 0.86 (1.70) | −2.52 – 4.24 | .615 | ||||||
| Change in ICR * Treatment Type | 5.19 (3.35) | −1.46 – 11.84 | .125 | ||||||
| Model Fit | |||||||||
| R 2 | 0.05 | 0.06 | 0.10 | ||||||
| Adjusted R 2 | 0.02 | 0.01 | 0.03 | ||||||
Note. N = 100; HAM-D = Hamilton Depression inventory, ICR = Insight into Conflictual Relationship Patterns scale.
Overall, our results provide support for the theoretical models of insight as an agent of change in dynamic psychotherapies. In partial fulfillment of our hypotheses, we demonstrated (a) that insight significantly changed during SE and (c) that increased insight predicted subsequent decreases in depression in SE. Contrary to our expectations, there was (b) no concurrent change in insight and outcome and (a +c) no significant interaction with treatment type. Notably, the main effect of gains in insight was marginally significant in the basic model and reached significance only when the interaction term with treatment type was entered into the model. Main effects in interaction models are often misinterpreted as average or independent effects of a variable (Brambor et al., 2006; Braumoeller, 2004). However, these effects represent the conditional effect of one variable on the outcome when the other variables are fixed at 0, meaning that based on the coding of our treatment variable (SE = 0, CT = 1), the significant conditional effects of session number and gains in insight denote the effects in the dynamic treatment group. The conditional effect has a meaningful interpretation even in the absence of a significant interaction term (Brambor et al., 2006). Running a cross-check with reverse dummy coding showed that insight did not improve during CT and gains in insight did not predict subsequent symptom improvement in CT. Note that because of the insignificant interaction terms, we showed that the effects are significant in dynamic and nonsignificant in cognitive treatments, while the difference between cognitive and dynamic treatments itself is not significant.
Our results are well aligned with previous studies supporting the mediating role of insight (Høglend & Hagtvet, 2019; Johansson et al., 2010; Ulberg et al., 2017) and its specificity for dynamic treatments (Connolly Gibbons et al., 2009; Kallestad et al., 2010). The small to moderate effect of change in insight on subsequent symptom changes corresponds with meta-analytic findings (Jennissen et al., 2018). To our knowledge, this is the second study investigating insight as a potential change mechanism in an RCT comparing dynamic and cognitive treatments. Expanding beyond previous investigations, our results provide support for insight as an agent of change in a sample of depressed patients treated in an ecologically valid setting. While we found indicators of specificity for dynamic psychotherapy, further research is necessary to fully support specificity by demonstrating a significant interaction with type of treatment.
From a theoretical perspective, the SE therapists’ focus on interpersonal patterns inside and outside the therapeutic relationship might have enabled patients to gain insight into their own maladaptive patterns (Luborsky, 1984). Employing a distinct focus, short-term dynamic therapies first identify a currently problematic interpersonal pattern. Patients may report on conflicts with their partners, close friends, or children. Through clarification, confrontation, and interpretation of the therapist, they can learn to identify patterns and find alternative ways of responding. Depressed patients with a dependent and demanding way of interacting may understand why they behave in that way and thereby gain more freedom in choosing how to act. They may also recognize what their behavior elicits in others, realize that they might get appreciated more when they demand less, and start behaving accordingly. When they start experiencing more positive interactions, the need for depressive symptoms as a way to express anger and the wish to be cared for could be diminished. Thereby, improved insight into maladaptive interpersonal patterns could reduce depression in short-term treatments. Considering there were no concurrent changes in insight and depression, but a timeline of early changes in insight and later improvements, one might assume that these processes take some time to unfold. By demonstrating that gains in insight predict subsequent decreases of depression in supportive-expressive dynamic therapy, we showed that the target mechanism of SE was not only addressed in treatment (as shown by adherence ratings), but also induced the proposed changes in the patient.
Our findings imply that within short-term dynamic treatments, an explicit focus on maladaptive interpersonal patterns could be beneficial to patients in the way described above. Interpersonal patterns are a specific domain of potential insights and different theoretical schools may emphasize other areas, such as defense mechanisms (Perry & Bond, 2012) or intrapsychic conflicts (Høglend et al., 1994). Within short-term treatments, we consider maladaptive interpersonal patterns a particularly useful focus because all other aspects of psychodynamics such as defense mechanisms or unconscious conflicts are represented and directly experienced in those relationship episodes. Furthermore, working through a distinct interpersonal pattern is a more realistic goal of short-term treatment than restructuring or changes in personality. Lastly, we chose this focus of treatment and insight because the majority of treatments are conducted as short-term therapies (Shapiro et al., 2003). However, it is worth mentioning that other aspects of insight are not only relevant from a theoretical perspective, but have also been investigated previously. For instance, Grande et al. (2003) investigated structural change and found that insight into intrapsychic conflicts and structural vulnerabilities predicted long-term outcome after inpatient psychotherapy. Other areas of insight that have previously been assessed include emotional involvement (Castonguay et al., 1996), psychological mindedness (Nyklíček et al., 2010), or therapeutic realizations (Ambühl, 1993; Kolden, 1991).
The study is limited in generalizability by the sample and setting of treatment delivery. While major depression is the most prevalent mental disorder (Kessler et al., 2007), we do not know how well our findings extrapolate to patients with other diagnoses. The CMHC presents an ecologically valid setting but may differ from treatments conducted in private practice with regard to patient demographics, social and economic problems, and therapist level of expertise. Further research is necessary to expand our knowledge on insight as a process factor beyond the currently investigated study populations.
Furthermore, we investigated short-term treatments with up to 16 sessions, of which patients in our sample attended an average of 10 sessions (M = 9.83, SD = 4.36). This number is only slightly lower than a meta-analytic report on the mean number of psychotherapy sessions for depression (M = 11.79, SD = 5.51; Cuijpers et al., 2013). Our study should thus be comparable with other investigations. Still, results may differ in long-term treatments.
Further, this study was performed on a subsample of 100 out of 237 patients who had participated in the original trial. Early dropout of a large proportion of the sample prevented psychotherapy process analyses in these patients. To counteract the potential selection bias this may have introduced, we adjusted for predictors of dropout and outcome using propensity score adjustment in all models. However, the study sample size prevented us from applying highly sophisticated statistical models. For instance, operationalizing change in a response surface analysis (Edwards & Parry, 1993; Nestler et al., 2019) would help to disentangle the level from the margin of change (e.g. do changes from low to moderate insight affect outcome differently than changes from moderate to high insight?). Latent modeling strategies such as the random intercepts cross-lagged panel model (RI-CLPM; Hamaker et al., 2015) model change without measurement error, differentiate changes between individuals from changes within individuals, and allow to investigate reciprocal effects (how large are the effects of change in insight → change in outcome vs. change in outcome→change in insight?). While this study employed an RCT sample to maximize internal validity regarding conclusions on differential effects in cognitive and dynamic treatments, future naturalistic studies could gather larger sample sizes and perform the aforementioned analyses to deepen our understanding of the mechanism of change.
Lastly, the nature of the examined treatment factor demanded for an observer-based assessment. Self-reports on insight are prone to bias since patients can hardly estimate “what there is to understand” at the beginning of psychotherapy. Although the raters were formally blinded to treatment condition, they could have inferred the treatment type when listening to the tapes, which may have biased their insight scores. While this is a limitation of all psychotherapy process studies relying on session evaluations, it came with the benefit of assessing what actually happened during these psychotherapy sessions, thus providing a more proximal view on the change mechanism than self-reports or interviews conducted after the session has ended. To fully disentangle change process from change mechanism, future studies should investigate the link between therapist interventions and client insight.
To conclude, this study provides support for one important assumed mechanism of change in a modern dynamic psychotherapy of MDD. While both treatments achieved comparable outcomes, CT and SE may operate through different mechanisms. Previous analyses of the cognitive mechanisms of change suggested that acquiring positive compensatory skills is the main change factor in CT (Crits-Christoph et al., 2017). The current paper further elucidates the operating mechanism of dynamic psychotherapy. Patients gained insight into maladaptive interpersonal patterns over the course of SE and this increased self-understanding predicted subsequent decreases of depressive symptoms for patients in SE. While the strengths of this study lie in the repeated assessments of insight and outcome, which allowed us to establish a timeline and demonstrate a dose-response-relationship, tasks for further investigations of insight as a mechanism of change include investigating several potential mechanisms, using therapist interventions to manipulate the insight, and employing different methodologies to evaluate whether these lead to converging conclusions (Cuijpers et al., 2019; Kazdin, 2007).
This study suggests that insight is a potential change factor in dynamic psychotherapy for depression. Understanding problematic relationship patterns may lead to better symptom improvement for patients.
This research was supported by a grant from the Agency for Healthcare Research and Quality (R01HS018440) to Dr. Connolly Gibbons as well as a dissertation scholarship of the Cusanuswerk Bischöfliche Studienstiftung e.V., Germany, to Simone Jennissen, which is gratefully acknowledged. The content is solely the responsibility of the authors and does not necessarily represent the official views of the Agency for Healthcare Research and Quality. We thank the clinicians, interviewers, psychotherapists, participants, and graduate students involved in the study for their support.
The study protocol for the original trial has been published in advance (Connolly Gibbons, M. B., Mack, R., Lee, J., Gallop, R., Thompson, D., Burock, D., & Crits-Christoph, P. (2014). Comparative effectiveness of cognitive and dynamic therapies for major depressive disorder in a community mental health setting: study protocol for a randomized non-inferiority trial. BMC Psychology, 2(47). https://doi.org/10.1186/s40359–014-0047-y). The data reported in this manuscript have partially been published previously. Findings have been reported in separate manuscripts. The parent study ( Connolly Gibbons, M. B., Gallop, R., Thompson, D., Luther, D., Crits-Christoph, K., Jacobs, J., Yin, S., & Crits-Christoph, P. (2016). Comparative effectiveness of cognitive therapy and dynamic psychotherapy for major depressive disorder in a community mental health setting: A randomized clinical noninferiority trial. JAMA Psychiatry, 73, 904–911. https://doi.org/10.1001/jamapsychiatry.2016.1720) is an RCT comparing the effectiveness of cognitive and dynamic therapy for major depressive disorder and focuses on depressiveness as the main outcome variable. The second study (Crits-Christoph, P., Gallop, R., Diehl, C. K., Yin, S., & Gibbons, M. B. C. (2017). Mechanisms of change in cognitive therapy for major depressive disorder in the community mental health setting. Journal of Consulting and Clinical Psychology, 85, 550–561. https://doi.org/10.1037/ccp0000198) details the mechanisms of change in cognitive therapy and focuses on dysfunctional attitudes, ways of responding, and depressogenic schemas. The third study (Gibbons, M. B. C., Gallop, R., Thompson, D., Gaines, A., Rieger, A., & Crits-Christoph, P. (2019). Predictors of treatment attendance in cognitive and dynamic therapies for major depressive disorder delivered in a community mental health setting. Journal of Consulting and Clinical Psychology, 87, 745–755. https://doi.org/10.1037/ccp0000414) explores predictors of dropout and focuses on patient baseline characteristics as well as early alliance and opinions about treatment.
1 As a sensitivity analysis, we also ran these models using residualized change scores for the HAMD outcomes. This did not change the pattern of results.
The authors report no conflict of interest.
Simone Jennissen, Department of General Internal Medicine and Psychosomatics, Heidelberg University, Heidelberg, Germany.
Mary Beth Connolly Gibbons, Department of Psychiatry, University of Pennsylvania, Philadelphia, Pennsylvania.
Paul Crits-Christoph, Department of Psychiatry, University of Pennsylvania, Philadelphia, Pennsylvania.
Henning Schauenburg, Department of General Internal Medicine and Psychosomatics, Heidelberg University, Heidelberg, Germany.
Ulrike Dinger, Department of General Internal Medicine and Psychosomatics, Heidelberg University, Heidelberg, Germany.
