In the last decades, psychotherapy research has consistently addressed the importance of the working alliance between therapists and their clients (Horvath et al., 2011; Flückiger et al., 2018; Tschuschke et al., 2020). In general, studies tend to show that the quality of the therapeutic relationship is a strong predictor of therapeutic outcome (O'Connor et al., 2019).

Process-outcome research has increasingly focused on personality characteristics of therapists and clients: Studies have addressed therapists’ and clients’ attachment styles (Marmarosh et al., 2014; O'Connor et al., 2019), therapists’ professional experience (Goldberg et al., 2016), therapists’ effectiveness (Wampold & Brown, 2005; Baldwin & Imel, 2013; Berglar et al., 2016), therapists’ and clients’ interpersonal skills and verbal capabilities (Wampold et al., 2017), the role of treatment adherence (Tschuschke et al., 2015; Kivlighan et al., 2019), clients’ initial symptom distress profiles (Tschuschke et al., 2015; Kivlighan et al., 2019; Uckelstam et al., 2019), or the working alliance congruence between client and therapist (Kivlighan et al., 2014; Zilcha-Mano et al., 2018; Tschuschke et al., 2020).

Although many researchers have assessed the working alliance from either the therapist’s or client’s perspective (Kivlighan et al., 2019), there are reasonable arguments for investigating the quality of the working alliance as a collaboration between clients and therapists. Research questions have been raised in recent years regarding methodological limitations and potentials in measuring the alliance experience (Tryon et al., 2007; Falkenström et al., 2017; Zilcha-Mano, 2017). Limitations concerned measuring the alliance mainly and solely on the side of the client and only for a few sessions, mostly at the beginning of treatment and not throughout the entire treatment. Therefore, this study aims at investigating ratings of the alliance by both client and therapist across treatment duration in order to identify which side is converging or disconnecting with the alliance impression on the other side. There has also been increased research interest in the last 10 years in alliance ruptures and their repair (Coutinho et al., 2014; Eubanks et al., 2018; Horvath, 2018).

Researchers found that clients’ severity of psychological distress had a strong impact on the quality of the therapeutic alliance. Also, more successful treatments were characterized by therapists’ ability to adapt their own sense of therapeutic alliance by approaching their clients’ level of alliance ratings as treatment progressed (Tschuschke et al., 2020).

We used an indirect self-report method, a version of the naturalistic observation paradigm, by tracking the natural occurrence of alliance ruptures and resolutions and examining their relationship to outcome (Eubanks-Carter et al., 2010). Most previous studies concentrated on measuring clients’ subjective experience of the alliance quality which we think is only half of the truth. We therefore looked at both sides of the alliance experience. We thus used clients’ and therapists’ alliance rating differences at the end of the session. We built on results from previous research and ended up with the following hypotheses:

FormalPara Hypothesis 1

Initial psychological distress, the chronicity of clients’ psychological problems, and therapists’ effectiveness significantly predict the emergence of alliance ruptures.

FormalPara Hypothesis 2

Alliance ruptures significantly predict treatment outcomes.

Methods

Participants

The data were derived from a nationwide naturalistic process-outcome psychotherapy study in Switzerland from 2007 through 2013 (PAP-S study) (Von Wyl et al., 2013). Each participating therapist was asked to work according to his or her usual practice routine. Starting from a time point in 2007, therapists were requested to ask all clients entering psychotherapy to participate in the study voluntarily. Each client was assured of having the right to not participate in the study and to receive treatment from the same therapist. Clients who participated signed a written informed consent form. A research application was submitted to the ethical committees in the relevant seven Swiss cantons (states) before the start of the project; the ethical committees approved all of the applications (Von Wyl et al., 2013).

Outcome Measures

The outcome battery was administered by independent, trained psychotherapists (not identical with clients’ therapists and not involved in the study). Three tests were completed by the clients: The first was the Brief Symptom Inventory (BSI) (Franke, 2000), which consists of 53 items comprising a broad range of psychological symptoms and nine subscales. We used the Global Severity Index (GSI) as a global measure for psychological distress. The scales have satisfactory high internal consistencies, ranging between 0.70 and 0.89 for the GSI. Convergent and concurrent validities were established by high positive correlations with different clinical self-rating scales (Geisheim et al., 2002).

The second outcome measure, the Outcome Questionnaire (OQ-45.2) (Lambert et al., 2002), measures symptom load, interpersonal relationship functioning, and the quality of the social integration. The German version has internal consistencies ranging from 0.59 to 0.93 for the different scales (Cronbach’s alpha). Validation studies showed convergent and concurrent validities between 0.45 and 0.76.

The third outcome measure was the German version of Beck’s Depression Inventory (BDI-II) (Hautzinger et al., 2006). All scales have an excellent internal consistency (Cronbach’s alpha ≥ 0.84) and the retest reliability is 0.78 after 3 weeks as well as after 5 months. Convergent and discriminant validities range between 0.68 and 0.89, depending on the depression measures used.

Global Outcome and Quality of Treatment Outcome

Outcomes were operationalized using the strategy of multiple outcome criteria (Fishbein & Ajzen, 1975; Ajzen & Fishbein, 1980). Rather than use a single outcome criterion, we combined several outcomes from the three different outcome measures to do justice to the complexity of therapeutic effects. For this, T-score transformations for each score of each outcome measure (BSI-GSI, BDI-II, and OQ-45.2) at each measurement point were made. T-scores were then summed up across the three outcome measures, each at pre (t1), post (t2), and follow-up (t3), and the total at t3 was subtracted from the total at t1, resulting in a final ‘outcome T-score.’ T-score sum at premeasurement (t1) served also as a measure of the client’s initial severity of psychological problems prior to treatment.

Quality of treatment outcome was defined by using both statistical reliability and clinical significance (reliable change index [RCI] and cutoff score) (Jacobsen et al., 1984; Jacobsen & Truax, 1991). Treatment success was defined by a change score greater than RCI and a score less than the cutoff score of the respective measure (remission) as well as a change score greater than the RCI and a final score greater than the cutoff score (responder). Treatment failure (no change) was defined by a change score less than the RCI and still greater than the cutoff score of the particular test, or by a deterioration.

Therapists’ Effectiveness

Therapists’ effectiveness was calculated via differences of the T-transformation scores between pretreatment and post-treatment/follow-up, which were based on the three outcome measures. A factor analysis of the change scores was carried out as in Blatt et al. (1996) to obtain factor scores (eigenvalue > 1) that served as a composite measure of the therapist’s effectiveness. The total of the resulting scores for each therapist was then subjected to a hierarchical cluster analysis in order to find clusters of more effective and less effective therapists (Berglar et al., 2016). Two clusters, labeled as ‘more effective’ and ‘less effective,’ were found.

Process Measure

The therapeutic relationship was rated after each fifth session using the Helping Alliance Questionnaire (HAQ) (Alexander & Luborsky, 1986; Luborsky et al., 1996; De Wert-van Oene et al., 1999). The questionnaire consists of 11 items and comprises two scales: The first scale measures the therapeutic alliance as experienced by the client (and in the therapist version, the alliance as experienced by the therapist). Scorings ranged from 1 (not applicable) to 6 (fully applicable) for each item. The higher the experienced alliance, the higher the rating on each item. Scale 1 scores were calculated by summing up the ratings on the six items and dividing the sum by 6. The second scale measures treatment satisfaction, (5 items, again, in a client and a therapist version). This approach can be considered as an indirect self-report approach, as it uses clients’ and therapists’ alliance ratings of their subjective impressions of the quality of their working alliance in the preceding session (Eubanks-Carter et al., 2010).

We administered the HAQ in the client version (HAQ-P; α = 0.88) and in the therapist version (HAQ-T; α = 0.89). We applied the factorial solution by De Weert-van Oene et al. (1999), which divides the 11 items into two subscales: alliance (client version [HAQ-A-P; α = 0.90], therapist version [HAQ-A-T; α = 0.87]) and treatment satisfaction (client version [HAQ-TS-P; α = 0.79], therapist version [HAQ-TS-T; α = 0.80]). We used the alliance subscale as a measure for the quality of the therapeutic relationship (alliance) as experienced by clients (HAQ-A-P) and therapists (HAQ-A-T) (see also Tschuschke et al., 2020, p. 57).

Data Analyses

Mixed model analyses were calculated to identify variables that predicted alliance rupture. The mixed model analyses were calculated with different relevant variables as fixed factors (mixed model analysis and fixed effects). Because of the nested data structure (some therapists treated different clients), therapists were included in the analyses as a random factor. T-tests, univariate analysis of variance, factor analysis, hierarchical cluster analysis, correlations, and linear mixed model analysis were all calculated using IBM SPSS Statistics Version 27 from 2020.

Results

The data used in this study are based on a total of 177 clients treated by 60 therapists. Clients with Axis I diagnoses (anxiety, affective, adjustment, eating, sexual, posttraumatic stress disorders) and Axis II diagnoses (cluster A, B, and C) of the DSM-IV were included; clients with substance abuse problems and clients with schizophrenic spectrum disorders were not included in the study. There were no age restrictions on client inclusion. Two thirds of the clients entered treatment for the first time, whereas approximately one third was labeled ‘chronic,’ as they had been treated one or more times before in inpatient or outpatient settings (Table 1). The client sample can be considered as relatively well educated, with approx. 60 percent having at least a university entrance diploma. Only 3.4 percent of the client sample was unemployed at the beginning of treatment.

Table 1 Clients—demographic data—n (%)
Full size table

Clients’ age ranged from 17 to 72 (M = 40.1, SD = 11.2). The sex distribution mirrors the typically found 2:1 distribution: 70% of the clients were women and 30% men. Surprisingly, more than 70% of all clients lived alone (as single, separated/divorced or widowed). Regarding DSM-IV diagnoses, 81% of the clients had an Axis I diagnosis and approximately 34% had an Axis II diagnosis.

Therapists were affiliated with 10 different theoretical concepts, including psychodynamic approaches (psychoanalysis, analytical psychology, newer psychoanalytic concepts), humanistic concepts (Gestalt Therapy, Bioenergetic therapy, Transaction Analysis), Integrative Body Psychotherapy (IBT) as developed by Jack Lee Rosenberg, Expressive Arts Therapy (EAT), also referred to as intermodal therapy, following Paolo Knill, Existential Psychotherapy and Logotherapy as developed by Viktor Frankl, and Process-Oriented Psychotherapy (an integration of Jungian analytical psychology, humanistic psychology, and physics following Arnold Mindell).

Table 2 Assignment of therapists and their treatments with respect to alliance rupture
Full size table

More than 70% of the therapists were women. Therapists’ age ranged from 35 to 79 (M = 54.1, SD = 8.0). Therapists were very experienced (nearly 14 years of professional experience on average). The clustering of therapists’ effectiveness into two groups (more and less effective) is described elsewhere in detail (Berglar et al., 2016). This sample comprised 124 successful and 53 unsuccessful treatments treated by 40 more effective and 20 less effective therapists. The 10 different theoretical orientations of the therapists were clustered in four main theoretical orientations (humanistic, psychodynamic, body oriented, and integrative). About 75% of the total sample were treated following either body oriented or humanistic approaches; 21% were treated with a psychodynamic and 4% with an integrative treatment approach.

Treatment Outcome

Therapists provided approximately 2 to 5 clients on average; clients per therapist ranged from 1 to 8. The general treatment outcome across the sample of this study (N = 177) (effect sizes [ES]) in the BSI were 0.93 at post-treatment and 1.22 at 1-year follow-up (Cohen’s d), the effect size for the OQ-45.2 was 1.04 at post-treatment and 1.53 at 1-year follow-up, and the effect size for the BDI-II was 0.96 at post-treatment (missing values at follow-up). As the effect sizespre-FU showed, clients continued to improve substantially after treatment had ended. All therapeutic gains were achieved by an average of 58.2 therapy sessions (s = 37.3). Thus, the effect sizes were consistently in the upper range compared to values reported in the corresponding literature. Conceptual orientations/psychotherapy approaches did not differ substantially in treatment outcome (Tschuschke et al., 2015).

Alliance Ruptures and Distance in Alliance Ratings

To identify significant shifts in the alliance ratings—compared to minor fluctuations (Stiles et al., 2004)—we calculated the mean of the differences between clients’ and therapists’ alliance ratings across treatments for the whole sample similar to the study by Strauss et al. (2006). However, we took the peaks of the differences in the alliance ratings between therapist and client in each treatment and calculated the mean of these 177 peaks. The mean was 1.37 (SD = 0.52). We ended up with 82 treatments with peaks in the alliance differences that were well beyond the mean of the whole distribution (46.3%) in at least one session. Thus, 82 treatments out of 177 showed alliance ruptures at some point during treatment.

Twenty therapists out of the total of 60 therapists (not identical with the less effective group of therapists) treated 85 clients and had alliance ruptures in 55 treatments (64 percent), whereas 40 therapists had a total of 27 treatments with alliance ruptures (29.3 percent) and 65 treatments without ruptures in the working alliance (70.7 percent) in the remaining 92 treatments. Twenty-five (45.5 percent) of the 55 therapies with alliance ruptures ended prematurely within 15 sessions after a rupture in the alliance experiences of both therapist and client had occurred. Thus, an alliance rupture was not necessarily followed by a premature ending of the treatment, and the rupture was repaired, so that the therapy could be continued in 30 of the 55 treatments (54.5%).

The 85 treatment cases of the 20 therapists with more alliance ruptures were significantly less effective compared to the 92 treatment cases of the 40 therapists with few ruptures in their working alliances (Scale HAQ-A-T; T =− 2.036; df = 175; p < 0.043).

In total, 95 treatments (53.7%) had alliance difference ratings under the critical difference score of 1.37 and were taken as treatments without an alliance rupture. This seemed to us to be a conservative approach, as we took only the highest alliance difference ratings of each of the 177 therapies as the basis for further calculations. Therapists were considered ‘distant’ to their clients if the average level of all their alliance ratings was beyond the critical distance score of 1.37 throughout treatment.

The return to an alliance rating difference lower than 1.37 for at least three further sessions (which in fact were based on a minimum of 15 more sessions) served as the criterion for a rupture repair. No recovery in the alliance rating difference and a discontinuation of the treatment within the next three session ratings (at least 15 treatment sessions) was defined as a premature ending and a demolition of the treatment. Table 2 shows that 17 therapists who treated 40 clients were consistently ‘distant’ in their alliance ratings, whereas only 1 client of the study sample was distant in her alliance ratings. Thus, some therapists cannot catch up with most of their clients alliance level.

Alliance Ruptures and Clients’ Characteristics

Clients’ degree of psychological distress at treatment entry as well as the degree of their chronicity of psychological problems did not predict alliance rupture. This was the case also for demographic variables and diagnostic classifications.

Alliance Ruptures and Therapists’ Characteristics

Treatments with interruptions in the working alliance that were repaired did not occur significantly more often in treatments by effective therapists than in treatments by less effective therapists. However, more effective therapists had fewer premature treatment terminations (10.3% of their treatments) than less effective therapists (21.7% of their treatments).

Prediction of Alliance Ruptures

Table 3 shows results of a mixed model analysis. Ruptures in the therapeutic alliance were not predicted by the extent of clients’ psychological problems, their degree of chronification, or therapists’ effectiveness, thus disproving hypothesis 1.

Table 3 Mixed model analysis (dependent variable: alliance rupture)
Full size table

Likewise, these variables did not predict discontinuation of treatment, as calculated in another mixed model analysis with premature treatment termination as dependent variable. However, as Table 3 shows, therapists’ distances in alliance ratings, in contrast to ratings by their clients, highly significantly predicted alliance ruptures. And alliance ruptures in turn highly significantly predicted premature termination of treatment (Table 4; p < 0.003).

Table 4 Mixed model analysis (dependent variable: termination of treatment)
Full size table

Alliance Ruptures and Prediction of Treatment Outcome

As the results of a further mixed model analysis shows, treatment outcome was predicted by clients’ severity of psychological problems at treatment entry (p < 0.000) and by therapists’ effectiveness (p < 0.000). Again, the nested data structure had no influence on the results; the person of the therapist per se did not play an important role. Treatments with alliance rupture or with repaired alliance ruptures did not predict treatment outcome, thus refuting hypothesis 2. But another variable was found to be slightly predictive in psychotherapy: The higher therapists’ distances in their alliance ratings were compared to their clients’, the less effective these treatments tended to be (p < 0.075).

Discussion

The results of this study provide insights into a complex picture of the therapeutic process. Client variables did not play a major role in ruptures in the therapeutic working atmosphere in our sample of 177 treatments. Based on our data, therapists’ conceptual orientation did not play a significant role in the causation of breaks in the therapeutic alliance in psychotherapy or regarding treatment outcome.

We also found that diagnostic categories did not contribute to an understanding of ruptures in the working alliance in therapy. As discussed also in other studies (Safran, 1993; Horvath, 2000; Strauss et al., 2006), our findings support the suggestion that ruptures in the therapeutic alliance do not necessarily worsen treatment outcomes.

Breaks in the therapeutic alliance were not predicted by clients’ chronicity of their psychological problems or their current degree of psychological distress. However, we found that clients’ initial psychological distress predicted treatment outcome highly significantly (Tschuschke et al., 2015), but the degree of clients’ psychological burden at treatment entry did not predict breaks in the working alliance. Also, therapists’ effectiveness did not predict ruptures in the alliance or discontinuation of treatment.

Regarding alliance ruptures, we found one therapist feature to have a significant effect: The only therapist variable predicting ruptures in the therapeutic working alliance was a specific attitude on the part of some therapists, namely, throughout treatment, alliance ratings > 1.37 below ratings by their clients. In this study, 25 of the 177 treatments ended prematurely. Of these 25 treatments, 12 cases were due to therapists’ distance in alliance ratings right from the beginning of treatment, whereas only one case was due to a client’s distance ratings. One might wonder whether this inability to bond sufficiently with their clients right from the beginning of treatment was due to an insufficient attachment style on the part of these therapists. Although Marmarosh et al. (2014) did not find an association between attachment styles of clients and therapists in the early alliance, our results are in favor of the hypothesis that therapists who seem to have difficulty establishing a good enough attunement early in therapy run an elevated risk for a burdened working atmosphere or even treatment failure. At least some studies support the hypothesis that securely attached therapists have skills that might help them handle alliance ruptures or even help them to avoid ruptures (Malinckrodt, 2000; Schauenburg et al., 2010; Strauß & Petrowski, 2017).

The results of this study are completely in line with previous findings of our larger research project. Successful treatments were conducted more often by therapists who showed significant convergence of their alliance ratings with their clients’ ratings over time, whereas discrepant alliance ratings correlated significantly with unsuccessful treatments (Tschuschke et al., 2020). In the present study, therapists with low and, compared to their clients, distant alliance ratings have highly significantly more treatment ruptures and run risks for premature treatment endings (in our study, 12 of 25 premature treatment terminations).

Limitations and Strengths

Weaknesses of this study concern mainly the sometimes very small subsamples, so that the generalizability of the findings may not be warranted. The present results should be taken as an attempt to generate hypotheses on the complex interrelationship between presumably relevant variables in the therapeutic process, which is based on the working atmosphere between therapist and client.

Strengths of the study can be seen in several aspects. The results are based on 177 treatments that were carried out by 60 very experienced therapists coming from 10 different theoretical orientations in a naturalistic clinical setting. First, the results cannot be traced back to particular theoretical affiliations. Second, the large sample with a typically wide range of psychological problems provides a solid ground for assumptions that should be investigated in further research. Third, the results are based on detailed process-outcome research that includes objective ratings of a subsample of therapists’ true intervention behavior across the whole treatments and continuous ratings of the working alliance on both sides of the therapeutic dyad.

Implications for Future Research

The results of this study are in favor of the idea that the person of the therapist plays a far more important role in psychotherapy than has long been assumed (Baldwin & Imel, 2013; Barkham et al., 2017; Wampold et al., 2017). Our findings point to the differential effectiveness of psychotherapists beyond their theoretical orientations and regardless of clients’ symptoms (Berglar et al., 2016). Recent research results suggest that more successful treatment outcomes in psychotherapy are due to therapists’ ability to adjust their orientation on relatedness or self-definition to their clients’ predominant personality configuration or to their clients’ sense of the working alliance (Werbart et al., 2018; Tschuschke et al., 2020). Research in this domain has to address such topics as therapists’ competence (whatever it may look like) and their capability to attune to their clients’ feelings and experience (keyword ‘attachment’).