Process‐outcome effects in psychotherapy research: an umbrella systematic review and meta‐analysis

Understanding how therapeutic processes are related to treatment outcome is one of the main challenges in psychotherapy research ¹ , ² . Although process‐outcome research has grown exponentially in the last two decades, with several meta‐analyses available, no prior study has synthesized these findings across the whole range of therapeutic processes associated with outcome. We performed an umbrella systematic review of process‐outcome meta‐analyses, classifying them into conceptual categories, and estimating aggregate outcome effects for each type of process.

A systematic search was conducted in PubMed and PsycINFO databases in February 2024, with an updated search in March 2025. We included meta‐analyses focusing on the relationship between process/es and outcome, and reporting at least one correlational effect size. The first title and abstract screening was performed by two independent reviewers (agreement: 94%). On a second step, full manuscripts were checked for eligibility (agreement: 89%). Disagreements at each stage were addressed by consensus. In the included studies, reviewers coded information from both the meta‐analyses and primary studies.

Processes of change were classified in the following categories ³ : relational processes, technical processes, patient processes, and therapist processes. Intersession processes focusing on the time between sessions (such as homework compliance) were not integrated into the systematic search.

We extracted all primary study correlational effect sizes (usually Pearson’s correlations) that represented an association between a psychotherapy process and a post‐treatment outcome (including primary and secondary outcomes). We conducted four‐level meta‐analyses estimating sample variance of the effect sizes (Level 1), between‐effect size variance (Level 2), between‐primary studies variance (Level 3), and between‐meta‐analyses variance (Level 4), using the R package metafor ⁴ . We first transformed correlational effect sizes into Fisher’s z ⁵ . To enhance interpretability, the results were back‐transformed into correlation coefficients (see also supplementary information).

There were 60 meta‐analytic studies that examined 24 different processes of change meeting inclusion criteria for the systematic review. The working alliance was the most studied process of change, with 25 meta‐analyses providing effect sizes for its association with outcome. The other most meta‐analyzed processes were adherence/competence/fidelity/integrity in therapeutic interventions (four meta‐analyses), and patient treatment expectations (three meta‐analyses). The majority of the meta‐analyses examined relational processes (n=39; 65.0%), followed by technical (n=7; 11.7%), therapist (n=9; 15.0%), and patient (n=6; 10.0%) processes. One meta‐analysis explored both therapist and patient processes.

The category of relational processes was dominated by the working alliance ⁶ . It also included further processes related to the alliance, but also having conceptual specificity, such as real relationship, alliance rupture‐resolution, goal consensus collaboration or congruence/genuineness, counter‐transference experiences, group cohesion, patient‐therapist mutuality or nonverbal synchrony.

The technical processes included therapist behaviors related to the adherence, competence, fidelity and integrity to/of theory‐specific interventions or methods. Some of these techniques were evaluated generically (e.g., cognitive‐behavioral or psychodynamic interventions), while in other cases meta‐analyses explored the effect of specific theory‐driven techniques, such as homework assignments and use of psychodynamic interpretations.

The therapist processes category included other, non‐technical therapist behaviors, such as therapists’ cultural/multicultural competences, cultural humility, positive regard, empathy, emotional expression or behaviors seeking patients’ emotional expression.

The patient processes included patients’ experiences within treatment that might be connected to psychotherapy outcome, such as emotional expression, cognitive restructuring, and treatment expectations and credibility.

Out of the 60 meta‐analyses included, 48 reported primary studies information. From those meta‐analyses, we extracted 1,981 effect sizes from 993 unique primary studies.

The four‐level meta‐analytic model suggested an average process‐outcome effect of r=0.238, z=0.243, SE=0.012, 95% CI: 0.219‐0.267, t₁₉₈₀=19.94, p<0.001, with significant heterogeneity among the effect sizes (Q₁₉₈₀=7388.1, p<0.001).

When including types of process as moderators, the aggregated process‐outcome associations of all the categories were significant. Following process‐specific rule of thumbs for effect sizes interpretation ⁷ , the models showed small effects on outcome for technical processes (r=0.102, p<0.001), small‐to‐medium effects for patient processes (r=0.227, p<0.001), medium effects for relational processes (r=0.267, p<0.001), and medium‐to‐high effects for therapist processes (r=0.291, p<0.001).

In the comparison of the effect sizes across the different types of process, the association of technical processes with outcome was significantly lower compared to relational processes (z=0.172, SE=0.022, 95% CI: 0.058‐0.146, t₁₉₇₇=6.95, p<0.001), therapist processes (z=0.198, SE=0.037, 95% CI: 0.126‐0.270, t₁₉₇₇=5.42, p<0.001), and patient processes (z=0.129, SE=0.034, 95% CI: 0.062‐0.197, t₁₉₇₇=3.79, p<0.001).

There were no significant differences between the effect sizes of relational and therapist processes (z=0.026, SE=0.031, 95% CI: –0.035 to 0.087, t₁₉₇₇=0.83, p=0.409), relational and patient processes (z=–0.043, SE=0.028, 95% CI: –0.099 to 0.013, t₁₉₇₇=–1.50, p=0. 135), or therapist and patient processes (z=0.068, SE=0.039, 95% CI: –0.008 to 0.145, t₁₉₇₇=1.75, p=0.081).

This is the first comprehensive systematic review of meta‐analyses examining processes of change in psychotherapy, providing estimates of the effect size for each type of process. Our results indicate that all the processes explored have significant effects, ranging from small to medium‐to‐high sizes. These findings advance our understanding of how psychotherapies work, providing evidence that can be useful to support their enhancement.

Relational, therapist and patient processes were found to have significantly greater effect sizes than technical processes. These findings match empirical and theoretical developments in the field supporting the relevance of therapeutic processes outside the disorder‐ and orientation‐specific framework of specific techniques ⁸ , ⁹ . However, it is important to note that the processes involved in psychotherapy change are not discretely separated, but operate together during treatment. This interaction should be the target of future specific studies.