RESEARCH DOSSIER
Integrative Psychotherapy and Treatment Matching
Abstract. Integrative psychotherapy and treatment matching investigate how therapeutic theories, procedures, and care pathways can be combined and selected for specific people, problems, clinical contexts, and changing treatment responses. This fundamental research monograph traces the history of psychotherapy integration and defines theoretical integration, technical eclecticism, assimilative integration, common-factors approaches, modular treatment, and transdiagnostic psychotherapy. It critically examines prospective treatment-matching trials, Project MATCH, randomized modular youth interventions, the Unified Protocol, the Personalized Advantage Index, therapist-to-patient matching, stratified depression care, clinical feedback systems, and the 2023 systematic meta-analysis of personalized psychological interventions. Methodological sections distinguish prognostic prediction from differential treatment-effect moderation, mediation from causal change mechanisms, retrospective model fit from prospective policy benefit, and individual preferences from statistically supported indications. The synthesis evaluates clinical outcome measures, longitudinal adaptive interventions, cultural accessibility, risks of bias, external validation, informed consent, patient safety, equity, implementation, and artificial decision support. Evidence supports selected structured personalization strategies under defined conditions; differences in intervention type, study quality, population, and healthcare organization prevent a universal matching rule. The research identifies priorities for randomized policy evaluation, replicated interactions, adaptive sequencing, interpretability, and durable functional outcomes. The work is a critical narrative evidence synthesis, with no new clinical trial, original patient data, registered systematic-review protocol, or independent meta-analysis.
Keywords: integrative psychotherapy; treatment matching; psychotherapy integration; technical eclecticism; theoretical integration; assimilative integration; common factors; transdiagnostic psychotherapy; modular psychotherapy; personalized psychotherapy; Personalized Advantage Index; heterogeneous treatment effects; treatment moderators; patient preferences; stratified care; measurement-based care; adaptive intervention; therapist matching; psychotherapy effectiveness; clinical decision support; Artificial Era
Introduction and Research Questions
Integrative psychotherapy and treatment matching constitute a research domain concerned with how psychological interventions are combined, selected, sequenced, and revised for particular persons and clinical contexts. Psychotherapy integration refers to principled coordination of explanatory models, intervention procedures, or change processes across established therapeutic traditions. Treatment matching asks whether a patient's pretreatment characteristics, preferences, risks, and ongoing response justify choosing one viable intervention, therapist, intensity, or care pathway rather than another. The central unit of analysis is a treatment decision conditional on a person and context, not the average efficacy of a therapy in isolation. The scientific challenge is to demonstrate that a decision rule improves outcomes over a credible alternative policy. The literature on precision mental health clarifies that a variable predictive of generally poor outcome need not identify which therapy is better for that individual. Cohen & DeRubeis (2018) Nye et al. (2023)
The research questions are correspondingly layered. Which forms of integration are conceptually coherent and clinically reproducible? What evidence establishes benefit from modular, transtheoretical, or transdiagnostic arrangements? Which clinical characteristics are prognostic across treatments and which are genuine treatment-effect moderators? How should patient preferences and therapist competencies enter selection? When should selection be revisited in response to new measurements? What methodological designs demonstrate that a proposed matching rule has real clinical utility? These questions connect psychotherapy process research to comparative effectiveness, causal inference, behavioral measurement, implementation, ethics, and service organization. They also require separating a clinician's justified flexibility from untested combinations of procedures. Kazdin (2007) Lutz et al. (2022)
This work is a critical narrative research monograph drawing on original randomized trials, historical and methodological scholarship, meta-analyses, prospective implementation studies, and an official clinical guideline. It does not claim to be a registered systematic review, an individual participant data meta-analysis, or an experiment involving newly recruited patients. Quantitative findings appear with their study designs, samples, and comparators; exploratory or post hoc analyses are explicitly distinguished from prospective validation. The evidence base is concentrated in depression, anxiety, youth emotional and behavioral presentations, and outpatient services, while examples from alcohol treatment show how early matching hypotheses were empirically evaluated. Clinical applications are discussed as scientific problems, with individualized care decisions reserved to qualified professionals and patients. Project MATCH Research Group (1998) Weisz et al. (2012) NICE (2022)
Definitions and Levels of Integration
Psychotherapy integration has several recognized routes. Theoretical integration attempts to combine concepts and mechanisms from more than one orientation into a coherent explanatory account. Technical eclecticism selects interventions on grounds of practical fit and evidence without requiring a unified theory of mind. Assimilative integration maintains one principal theoretical language while deliberately incorporating techniques originating elsewhere. Common-factors integration organizes clinical attention around processes shared across treatments, including collaboration, expectancy, emotional engagement, and contextual conditions for learning. The boundaries overlap in clinical practice, but the distinctions help specify what is being integrated: theories, techniques, interpersonal processes, or treatment packages. An historical survey of integration practitioners identifies these categories as substantive routes used by therapists rather than merely retrospective labels. Norcross et al. (2005) Norcross & Goldfried (2005)
Integration should also be distinguished from transdiagnostic and modular intervention design. A transdiagnostic treatment targets processes that maintain problems across conventional diagnostic categories; it may use a unified theoretical framework. A modular therapy is built from identifiable treatment elements and an explicit rule for their selection, sequencing, or modification. A therapist can be integrative without following a modular protocol, and a modular protocol can be developed almost entirely within a cognitive-behavioral tradition. The practical scientific advantage of explicit modules is that investigators can specify doses, decision points, fidelity, and component exposure. The challenge is that complex combinations introduce interactions and implementation demands that are not captured by the evidence for each component separately. Weisz et al. (2012) Barlow et al. (2017)
Treatment matching encompasses at least four different decisions: choosing between treatment modalities; selecting among active psychotherapies; allocating patients to clinicians with relevant competencies; and choosing the intensity, sequence, or components of care over time. These decisions involve different causal contrasts. A preference for psychotherapy rather than medication is not the same as a statistically demonstrated advantage of cognitive therapy over interpersonal psychotherapy. A therapist's historical performance with a given problem domain is not interchangeable with the average effect of a named treatment school. Likewise, a warning that someone may deteriorate identifies risk and monitoring needs without automatically specifying which alternative will prevent deterioration. Precision requires stating the candidate treatments and the outcome one intends to optimize. Constantino et al. (2021) Schwartz et al. (2021) Cohen & DeRubeis (2018)
Historical Emergence of Psychotherapy Integration
The development of distinct psychodynamic, humanistic, behavioral, and cognitive schools created both a vocabulary of competing explanations and an enduring practical problem: real patients often present difficulties that cut across each school's preferred objects. Integrative ambitions emerged through clinicians' attempts to articulate what treatments share, reconcile seemingly incompatible concepts, and borrow effective techniques from neighboring traditions. Historical synthesis is best understood as a set of interacting professional and research movements rather than a single founding event. The second edition of the Handbook of Psychotherapy Integration, edited by Norcross and Goldfried in 2005, records how the field had developed institutional training, programmatic debates, and multiple distinguishable integration strategies. Norcross & Goldfried (2005)
The evolution of outcome research changed the terms of the debate. With controlled comparative trials, the question of whether an intervention produces improvement became empirically tractable. Subsequent meta-analyses and comparative research emphasized that meaningful average benefits could coexist with heterogeneity of individual response, dropout, inadequate recovery, and adverse outcomes. This shifted emphasis from the abstract search for the universally best psychotherapy toward the more difficult question of what characteristics determine when a particular treatment decision has value. Cohen and DeRubeis's historical synthesis of depression treatment selection describes this transition from general prediction to multivariable individualized counterfactual recommendations. Cohen & DeRubeis (2018) Cuijpers et al. (2023)
Another historical current is the movement from fixed manuals to treatment systems with decision rules. Youth services dealing with anxiety, depressive symptoms, and conduct problems offered a testing ground for modular arrangements that reorganize recognizable evidence-based components in response to changing presentations. In parallel, clinical feedback systems began to transform outcome monitoring into a possible basis for revising plans rather than merely evaluating them after discharge. The historical trajectory thus joins integrative theory, evidence-based practice, care personalization, and data-informed clinical governance. Its common ambition is defensible responsiveness; its central scientific burden is to establish that responsiveness changes outcomes, not merely that flexible treatment feels more individualized. Weisz et al. (2012) Lutz et al. (2019) Lutz et al. (2022)
From Average Treatment Effects to Individual Contrasts
Average treatment effects answer a population question about the expected difference between interventions under a defined trial design. They rarely identify the best choice for any named patient. If two psychotherapies produce similar mean outcomes, one may still be preferable for some individuals, while the evidence for such variation must be shown rather than assumed. Treatment-effect heterogeneity is the difference between a person's potential outcomes under competing options; because only one potential outcome is observed for each person in ordinary care, the individual contrast requires causal assumptions, randomized comparisons, or carefully justified statistical inference. The personalized advantage framework attempts to estimate this counterfactual difference from baseline characteristics and treatment interactions. DeRubeis et al. (2014) Cohen & DeRubeis (2018)
A prognostic factor predicts outcome regardless of which treatment is given. Baseline symptom severity may predict an unfavorable result across modalities without establishing a relative advantage for one. A prescriptive factor or treatment moderator interacts with assignment such that comparative benefits vary by characteristic. This distinction is critical: selecting treatment based on prognosis alone can optimize intensity or monitoring while leaving the modality choice unsupported. Multivariable prediction can integrate multiple small and interdependent effects, but flexible models also risk overfitting and spurious interactions. The clinically meaningful test is how a prespecified policy performs when used prospectively on new patients in settings that resemble intended practice. Schwartz et al. (2021) Nye et al. (2023)
Outcomes further complicate selection. A policy optimized for short-term symptom reduction might differ from one optimized for sustained functioning, adverse events, dropout prevention, patient-reported quality of life, or service equity. The time horizon matters because early response can alter later treatment availability and choices. A difference on a symptom inventory may be statistically precise yet trivial to an individual facing cost, travel burden, cultural discordance, or lack of a particular therapist. Selection therefore combines causal evidence about expected outcomes with the person's own priorities and realistic service constraints. Clinical guidelines for adult depression explicitly incorporate preferences, prior treatment experience, need, risks, and timely monitoring into decisions rather than framing matching as a purely computational exercise. NICE (2022) Windle et al. (2020)
Scientific Scope, Taxonomy, and Boundaries
The dossier's primary object is the conditional design and allocation of psychological treatment. Research on psychotherapy mechanisms is incorporated only when it informs the choice or combination of procedures; a complete theory of alliance or emotion regulation belongs in its own scientific inquiry. Comparative psychotherapy effectiveness supplies the available options, while individualized matching examines differences between people and decision rules applied to those options. Measurement-based care supplies repeated observations that can modify a selection made at intake. These are connected research programs with separate inference targets, and evaluating one by the evidence standards of another invites category errors. Kazdin (2007) Lutz et al. (2022)
Clinical populations are not homogeneous decision units. Presentations may include overlapping diagnoses, chronic physical conditions, histories of treatment response, trauma exposure, family obligations, and barriers to attendance. Demographic categories such as age, gender, or ethnicity can correlate with access and experience without being valid stand-alone indicators for assigning a particular psychotherapy. The legitimate use of individual information requires a mechanism, a supported interaction or validated decision rule, and an assessment of fairness. Broad claims that people from a named population require one school of psychotherapy often confuse cultural adaptation, accessibility, treatment preference, and differential causal efficacy. Windle et al. (2020) NICE (2022)
The conceptual border also includes human–artificial systems used to support assessment or decision-making. A prediction model may rank expected outcomes, flag likely dropout, or suggest an evidence-supported treatment module. Such a system is a decision aid whose recommendations remain empirical hypotheses about outcomes, subject to dataset shift, measurement error, transparency requirements, and accountable clinical oversight. The Artificial Era adds a new locus for modeling and organizing care decisions, while the scientific test remains prospective comparative benefit and the ethical standard remains respect for human patients. Assertions about the psychology or personhood of artificial agents are separate philosophical questions and do not follow from successful treatment prediction. Lutz et al. (2019) Schwartz et al. (2021)
Theoretical Integration, Coherence, and Clinical Formulation
Theoretical integration begins with an account of what psychological processes are believed to sustain a person's difficulty and how interventions alter those processes. An integrative formulation might characterize avoidance, dysfunctional beliefs, interpersonal threat, and habitual emotion regulation as mutually reinforcing contributors to distress. The central research issue is whether these explanatory claims are sufficiently specified to generate interventions whose effects can be observed and distinguished. An integrative vocabulary does not itself establish a mechanism: constructs must be operationalized, tied to measurable temporal changes, and tested against explanations that fit the same outcome data. Kazdin's account of psychotherapy mechanisms remains fundamental here because an associated change in a putative mediator may follow symptom improvement, share a common cause, or reflect measurement artifacts. Kazdin (2007)
Assimilative integration resolves part of the problem of conceptual coherence by retaining a primary model while incorporating a technique that addresses a clinically important process. For example, a clinician working within a cognitive-behavioral framework may attend explicitly to interpersonal losses or patterns of therapeutic engagement. The coherent question is why the imported procedure is required, how it is expected to alter the mechanism, and whether the original treatment components retain fidelity. A technique can have different meanings and effects under different timing, therapeutic relationships, and expectations. Manualized fidelity, therapist competence, and patient response therefore remain relevant even when practice is deliberately flexible. Theoretical pluralism should make its rules of combination more visible rather than allowing every apparently useful action to count as evidence-based care. Norcross & Goldfried (2005) Kazdin (2007)
The context of treatment is itself part of the explanatory structure. Credibility, agreement on goals and tasks, interpersonal safety, and opportunities to practice new behavior can influence the uptake of specific procedures. Research on the alliance demonstrates a robust association with outcomes across approaches, while improved adjustment for patient and process characteristics does not fully settle how much of the association is causal. Integrative design therefore treats relational processes as potentially active ingredients and as conditions under which specific techniques operate. It should not infer that all methods are interchangeable merely because common relational processes matter. Controlled trials, dismantling studies, process measures and prospective randomization are needed to identify which combination has added value. Flückiger et al. (2018) Flückiger et al. (2020)
Technical Eclecticism and Evidence-Based Responsiveness
Technical eclecticism concerns the selection of interventions primarily on their demonstrated or plausible utility for a presenting process. Its attraction is practical: many clients encounter overlapping problems that call for more than one clinical technique across time. Its weakness is that treatment packages provide evidence for combinations evaluated in studies, not for every imaginable rearrangement of isolated techniques. Borrowing exposure from one approach and an emotion-processing exercise from another creates a new intervention sequence whose dose, timing, assumptions, and side effects may differ from each source protocol. A reasoned plan therefore documents the proposed target, available comparative evidence, known contraindications, and the criterion that would prompt revision. Norcross & Goldfried (2005) Barlow et al. (2017)
Evidence-based responsiveness differs from retrospective justification. After a client improves, it is easy to declare that the therapist correctly anticipated what the client needed; after deterioration, an untested narrative can similarly assign blame to resistance or poor motivation. Research requires the decision rule to be specified in advance and evaluated against plausible alternatives. Prespecification also helps distinguish sensible therapeutic adaptation within a protocol from unrestricted clinician improvisation. The most informative studies ask not simply whether clients who received a certain procedure got better, but whether an explicit strategy for selecting that procedure generated better expected outcomes than standard practice. DeRubeis et al. (2014) Nye et al. (2023)
The practical limits of eclectic combination are especially important in treatment involving trauma, acute risk, severe impairment, or complicated comorbidity. Clinical supervision and guideline-concordant assessment establish guardrails for changes in a treatment plan. Integration cannot bypass evidence on treatment indications, provider training, monitoring, or prevention of harm. A patient may value flexibility and still reasonably expect that the clinician can explain which components are evidence-supported for the presenting condition and how their use will be evaluated. Shared decisions require intelligible options and accessible explanation of uncertainties, not an unlimited menu of theoretically attractive techniques. NICE (2022) Lutz et al. (2022)
Transdiagnostic Models and Process-Based Organization
Transdiagnostic psychotherapy develops treatment procedures around processes that recur across symptom categories, such as avoidance of emotional experience, rigid threat appraisal, or behavioral withdrawal. Its attraction is both theoretical and organizational: when co-occurring presentations share maintaining mechanisms, clinicians may learn and administer a smaller set of flexible procedures. The Unified Protocol for Transdiagnostic Treatment of Emotional Disorders provides an influential example, with modules addressing emotional awareness, cognitive flexibility, avoidance and emotion-driven behavior, and emotion exposure. The theoretical assertion of a shared target is distinct from the empirical question of whether the transdiagnostic treatment performs as well as competing diagnosis-specific care for the populations studied. Barlow et al. (2017) Farchione et al. (2024)
In a randomized trial of 223 adults with anxiety disorders, Barlow and colleagues compared the Unified Protocol with diagnosis-specific protocols and a waitlist. The trial reported equivalent reductions in principal-diagnosis clinical severity for the two active approaches at posttreatment and six-month follow-up, within its prespecified equivalence framework, with higher completion for the Unified Protocol. These findings justify considering a process-organized approach in conditions represented by that trial. They do not establish that all transdiagnostic packages are superior, that every emotional disorder has the same mechanisms, or that a process-focused approach is the best initial option for every individual. The sample, therapist expertise, treatment setting, and choice of comparators limit inference. Barlow et al. (2017)
Process-based integration becomes most scientifically informative when treatment elements are linked to measured functional targets. One might hypothesize that a reduction in avoidance precedes improvements in social functioning, while change in self-criticism precedes improved engagement. Repeated measures can test temporal ordering, but temporal precedence alone still leaves unmeasured common causes. Experimentally varying a component, testing different module sequences, and observing the hypothesized mediator are stronger routes to causal interpretation. Contemporary transtheoretical models connect cognitive, emotional, motivational, behavioral, interpersonal, and systemic levels, yet their integration remains a research program whose separate claims require comparative validation. Kazdin (2007) Lutz et al. (2024)
Modular Psychotherapy as an Experimental Architecture
Modularity means that a treatment comprises recognizable elements whose delivery can be governed by an explicit decision logic. A module can target a specific symptom domain or mechanism, while the decision rule can respond to initial diagnostic presentation, functional impairment, symptom trajectory, and practical barriers. In theory, the approach combines replicability with clinical adaptation. In practice, module libraries must specify procedures, permissible combinations, clinician competencies, and stopping conditions. Without such details, 'modular' becomes a label for unspecified flexibility and generates little reproducible knowledge. A modular architecture also makes it possible to study the additional value of a component rather than relying solely on contrasts between whole branded schools of therapy. Weisz et al. (2012) Collins et al. (2007)
Weisz and colleagues' randomized effectiveness study compared standard manualized treatment, a modular approach, and usual care among 174 youths receiving treatment for anxiety, depression, or conduct problems. The modular condition integrated components derived from established treatments and permitted more responsive selection across problem domains. Its central contribution was the empirical comparison of treatment organization under community-relevant conditions. The trial should not be read as proving that any locally invented module combination will work: protocol training, quality assurance, decision procedures, and the particular presenting problems define what was tested. Subsequent applications also need independent replication across services with different therapist training and resources. Weisz et al. (2012)
Factorial experimentation offers a related method for optimizing modules. The Multiphase Optimization Strategy distinguishes component screening, refinement, and confirmation, while a sequential multiple assignment randomized trial can test alternative subsequent treatments at prespecified decision points. These designs offer better evidence for adaptive sequences than post hoc comparisons of clients who happened to receive different combinations in routine care. They also require careful specification of randomization stages, tailoring variables, outcomes, and estimands. The complexity of study design is justified when a dynamic question cannot be answered by a single baseline randomization. Collins et al. (2007)
Common Factors, Alliance, and Responsive Treatment
Common-factors approaches draw attention to processes that appear across orientations, including the working alliance, treatment expectancy, empathic attention, and opportunities for corrective learning. Their role in integrative treatment design is to identify conditions that may amplify or diminish the delivery of technique-specific procedures. The alliance is especially important because agreement about treatment tasks and goals provides a practical mechanism for recognizing a poor therapeutic fit before dropout occurs. A synthesis of 295 independent studies involving more than 30,000 patients found a reliable association of alliance with outcome across face-to-face and internet-based psychotherapy. This is a statement about a predictive relationship, not a randomized demonstration that increasing an alliance score by a fixed amount produces a specific improvement. Flückiger et al. (2018)
A subsequent meta-analytic analysis of 60 independent samples and 6,061 participants examined alliance-outcome associations adjusted for intake and concurrent treatment-process variables. The reported correlations remained moderate across adjustment categories, supporting the independent predictive role of alliance while leaving the causal ordering problem open. A therapist may adapt the pace, explanation, or module selection in response to weak task agreement. Research should still distinguish whether such adaptation changes alliance, whether alliance changes outcome, and whether early symptom improvement causes more favorable alliance reports. Repeated within-person methods help separate stable between-patient differences from dynamic changes in a therapeutic relationship. Flückiger et al. (2020) Zilcha-Mano & Fisher (2022)
The practical connection to treatment matching is iterative. A treatment may be well supported on average yet fail to engage a particular patient because tasks appear irrelevant, burdensome, inaccessible, or inconsistent with informed preferences. The appropriate response might involve negotiating goals, adapting delivery format, increasing support, offering an alternative active treatment, or changing therapist. Each option rests on a different evidence base, and the clinician should avoid equating an alliance problem with a necessary failure of the treatment mechanism. Integrative responsiveness becomes scientifically accountable when such decisions can be observed, timed, and linked to both patient-reported experience and clinically meaningful outcomes. Windle et al. (2020) Constantino et al. (2021) NICE (2022)
Project MATCH and the Logic of Prespecified Treatment Interactions
Project MATCH remains a landmark attempt to test whether characteristics of people seeking treatment for alcohol problems determine comparative benefits from alternative psychosocial approaches. The multisite trial randomized participants to cognitive-behavioral coping-skills therapy, motivational enhancement therapy, or twelve-step facilitation, with prespecified client-by-treatment hypotheses and longitudinal alcohol outcomes. Its importance lies as much in its methodological design as in any particular successful interaction. Explicit hypotheses permit a scientific distinction between genuine replication and a chance relationship uncovered after many candidate variables have been examined. An adequately designed matching study must also compare how the same baseline variable predicts outcomes under each active treatment. Project MATCH Research Group (1993) Project MATCH Research Group (1998)
The reported three-year follow-up found fewer robust differential matching relationships than the field's original enthusiasm might have suggested. Client anger showed a comparatively persistent interaction: among participants in the higher anger stratum, motivational enhancement therapy was associated with better drinking outcomes than the two other tested treatments. The original report also described a later interaction involving social support for drinking, whereas an earlier psychiatric-severity interaction did not persist to the three-year analysis. These results were conditional on the specific three therapies, clinical setting, measurements, and time horizon. They resist simplification into a universally applicable alcohol-treatment matching rule. Prognostic variables such as readiness to change had separate significance even when they did not supply dependable comparative treatment indications. Project MATCH Research Group (1998)
The general lesson is an inference discipline. A treatment assignment variable can have theoretical face validity and still fail in a randomized interaction test. A significant subgroup result can also lose stability at later follow-up or in independent replication. Many hypotheses create a multiplicity problem, especially when outcome domains and time points are numerous. Contemporary precision psychotherapy builds on this lesson by estimating multivariable contrasts, restricting the degrees of freedom available for post hoc selection, and evaluating models on participants not used to fit them. Retrospective predictive discrimination must never be confused with evidence that deploying a matching rule prospectively improves care. Project MATCH Research Group (1998) DeRubeis et al. (2014)
The Personalized Advantage Index
The Personalized Advantage Index, introduced in a 2014 demonstration by DeRubeis and colleagues, estimates the expected difference between two possible treatments for a person using pretreatment variables. In the reported depression dataset, 154 randomized participants had been assigned to antidepressant medication or cognitive behavioral therapy. The investigators used leave-one-out prediction so that a person's own outcome was not included in the model used to predict that person's alternative-treatment outcomes. Five baseline variables contributing to differential response were entered into regression models. The index was expressed in units of the Hamilton Rating Scale for Depression. The inferential ambition was to translate a set of moderator associations into a decision-oriented contrast rather than leaving clinicians to interpret unrelated predictors independently. DeRubeis et al. (2014)
The method is attractive because a clinically meaningful recommendation depends on both predictions. A patient with poor predicted outcome under every treatment might require greater intensity or monitoring, but their absolute risk does not identify the least harmful option. The index instead estimates whether one treatment is expected to outperform another for that same profile. Its performance depends on model specification, the reliability of baseline measures, the chosen loss function, the adequacy of randomization, and the distribution of patient profiles. A model trained on one comparison cannot legitimately select among therapies it has never observed. Furthermore, leave-one-out prediction mitigates some in-sample optimism while stopping short of validation in an independent new health system. DeRubeis et al. (2014) Cohen & DeRubeis (2018)
A policy based on a continuous predicted advantage also needs an action threshold. The clinical significance of a small modeled difference may be outweighed by preference, cost, safety, or the available therapist. Error bands are important: two predicted outcomes with wide uncertainty may be indistinguishable for decision purposes even if their point estimates differ. Treatment selection needs to consider calibration and actual benefit among those receiving the policy's recommended care. An assessment of model usefulness should therefore report uncertainty around decision value and how often the recommended treatment meaningfully differs from an appropriate default, as well as ordinary measures of symptom prediction accuracy. Cohen & DeRubeis (2018) Schwartz et al. (2021)
Selection Algorithms Compared With Clinical Judgment
An important empirical question is whether statistical selection adds value to the clinical reasoning already used by experienced practitioners. A depression analysis comparing personalized statistical recommendations with therapist judgment reported that participants receiving the statistically indicated treatment had lower depression severity than those receiving their statistically nonindicated treatment, whereas an analogous association did not appear for the clinician-recommendation comparison. The authors emphasized that the statistical predictions were not externally validated and that the clinicians' reasoning processes had not been recorded. The study therefore raises a consequential hypothesis about the comparative reliability of formal prediction and intuitive decision-making; it does not demonstrate that a deployed algorithm would outperform clinicians in a new prospective randomized service. van Bronswijk et al. (2021)
Direct comparisons must recognize that clinical judgment can use information unavailable to an algorithm, while algorithms can integrate interactions that humans evaluate inconsistently. Both may be influenced by systematic distortions. Clinicians can be guided by vivid recent cases, professional allegiance, stereotypes, or incomplete outcome feedback; prediction models can reproduce selection patterns and inequities embedded in historical referral and treatment data. Clinical judgment is also not a uniform comparator: an expert integrating supervision and outcome feedback differs from an unstructured intake recommendation. A fair test specifies the information available to each strategy, the choice set, and who ultimately decides when patient preferences conflict with the recommendation. Cohen & DeRubeis (2018) Lutz et al. (2022)
A practical research agenda compares several policies prospectively: treatment as ordinarily selected, evidence-informed shared choice, selection by a locked statistical model, and an integrated model in which clinicians receive transparent recommendations together with uncertainty and monitoring requirements. Patient-centered endpoints should include improvement, burden, retention, function, satisfaction, and subsequent care. A randomized comparison of policies can estimate total usefulness, including the consequences of clinician nonadherence and a patient's informed refusal. Accuracy metrics alone cannot capture those downstream effects. When a trial is infeasible, external validation in a genuinely new setting and careful observational emulation can advance knowledge, but the remaining confounding must be reported. Schwartz et al. (2021) Nye et al. (2023)
Out-of-Sample Performance and Routine-Care Selection
Routine-care datasets offer scale and ecological relevance while introducing confounding by indication, therapist assignment, missing assessments, and variation in treatment delivery. Schwartz and colleagues analyzed a disorder-heterogeneous outpatient sample of 1,379 people treated with cognitive behavioral or psychodynamic therapy and modeled differential response using machine-learning and regression methods. Their model performed better in the derivation sample than in the independent holdout sample. The overall holdout comparison between individuals receiving predicted optimal and nonoptimal treatments was nonsignificant. A subgroup with larger predicted advantages showed an apparent differential benefit, but subgroup selection and calibration require independent validation before clinical deployment. Schwartz et al. (2021)
This example illustrates why model performance is not a single number. A classifier could predict deterioration well while providing little evidence about which of two treatments is causally preferable. A treatment model could improve mean outcomes for a small fraction of patients even if its overall signal is weak. Conversely, selecting the subgroup with the most dramatic retrospective modeled differences may amplify estimation error, because extreme scores are often disproportionately influenced by noise. The research task is to demonstrate that the chosen threshold, model, measurement schedule, and selection policy are stable on future patients. Reporting both overall and selected-subgroup results prevents a promising local signal from being mistaken for universal effectiveness. Schwartz et al. (2021) Cohen & DeRubeis (2018)
The availability of high-dimensional clinical records and machine learning does not remove the need for randomized allocation or credible causal identification. Models may accidentally learn insurance coverage, practice site, or therapist access as if those variables were biological or psychological moderators. Changing the available treatments also changes the counterfactual choice problem. In a service where neither recommended therapy is accessible, a technically calibrated model has little immediate use. Deployability therefore requires local data quality, comprehensible outputs, retraining governance, referral capacity, and a policy for cases outside the training distribution. Prospective studies should evaluate clinician trust and patient autonomy alongside symptom measures. Lutz et al. (2022) Cuijpers et al. (2026)
Therapist-to-Patient Matching as a Distinct Intervention
Matching patients to therapists is different from matching them to treatment modalities. Clinicians vary in expertise, style, interpersonal skills, and performance with particular presenting problems. A patient receiving a named therapy may therefore experience different outcomes depending on who delivers it. Constantino and colleagues conducted a randomized clinical trial comparing measurement-based matching to usual therapist assignment. Forty-eight therapists and 218 outpatients participated; matching used therapists' historical effectiveness for patients' problem domains. The matched allocation produced significantly greater improvements in several symptom and impairment measures during follow-up than assignment as usual. The prospective randomized design is stronger evidence for the assignment policy than retrospective comparisons of therapists judged effective after seeing outcomes. Constantino et al. (2021)
The trial also leaves questions about transportability. It depended on an existing database of therapist outcomes and problem-domain classifications, which many services lack or record inconsistently. Estimates of therapist strengths can be unstable when caseloads are small, presenting problems are complex, or outcomes are missing for people who disengage. Public or administrative use of clinician rankings may change referral patterns, expose providers to perverse incentives, and systematically reduce access for high-need patients if adjustment for case mix is inadequate. Privacy and fairness become part of the intervention's evaluation. No matching system should be generalized to all settings without examining the reliability of its therapist-level estimates. Constantino et al. (2021) Lutz et al. (2022)
Even when a match is empirically plausible, personal preference and continuity have independent clinical value. A person may decline a proposed clinician for reasons involving language, location, prior experiences, or relational safety. A transparent matching service can present more than one acceptable option and monitor how assignment rules affect wait times and inequity. Scientifically, the outcomes of patient-to-therapist matching should be analyzed separately from the effects of modality choice, since improvement could arise through case-specific therapist skill, alliance, engagement, or other unmeasured processes. Factorial or sequential trial designs could test whether combining the two forms of matching produces additional benefits. Constantino et al. (2021) Windle et al. (2020)
Systematic Synthesis of Personalized Psychotherapy Trials
Nye, Delgadillo, and Barkham's 2023 systematic review and meta-analysis directly addressed whether personalized psychological interventions outperform standardized comparators. Seventeen studies involving 7,617 participants met the review criteria; nine studies with 5,134 participants contributed to the quantitative synthesis. The pooled effect favored personalized treatment, with a standardized difference of 0.22 and a 95% confidence interval from 0.05 to 0.39. When high-risk-of-bias studies were excluded, the reported effect was smaller, 0.14 with a 95% confidence interval from 0.08 to 0.20. The review classified only one included study as low risk of bias, underscoring why a favorable average effect must be interpreted alongside evidence quality. Nye et al. (2023)
The review compared a family of personalization strategies rather than one standardized intervention. Personalization can mean matching treatment type, stratifying intensity, flexibly selecting modules, responding to progress feedback, or accommodating preferences. If the mechanisms and control conditions differ, the pooled estimate is best understood as evidence that some personalization strategies may help on average. It cannot identify which rule is optimal for a particular disorder or person. Heterogeneous interventions also make it difficult to ascribe improvement to the matching algorithm itself rather than greater therapist attention, additional resources, or enhanced monitoring. Mechanistic claims require designs that separate decision information from associated treatment intensity and implementation support. Nye et al. (2023) Cuijpers et al. (2026)
The strongest synthesis is conditional: evidence supports continued development and careful testing of personalization, with promising effects in some randomized programs and important uncertainty concerning generalizability. A systematic review's pooled effect is not a guarantee that every local implementation will reproduce it. Because several studies are at risk of bias, apparently precise estimates may not fully represent the uncertainty associated with selection and measurement. Transparent public scientific communication should display null findings, long-term follow-up, and safety outcomes alongside favorable subgroup estimates. Selection algorithms that are difficult to interpret require higher standards of external validation because users cannot readily identify when the model is operating beyond its evidentiary domain. Nye et al. (2023) Schwartz et al. (2021)
Stratified Care and Treatment Intensity
A service often chooses not only among treatment schools but between levels of support. Stepped care begins with a lower-intensity intervention and escalates when progress is inadequate. Stratified care allocates intensity at the outset according to assessed risk or likely need, seeking to avoid delays for people whose presentations are unlikely to respond to lower-intensity services. These strategies are operationally distinct from choosing between cognitive and psychodynamic psychotherapy: the primary contrast can concern whether a person should receive low- or high-intensity support. A severity predictor may be useful for this problem even if it provides no evidence that one theoretical modality is better. The decision also affects waiting lists, staffing, cost, and equity. Delgadillo et al. (2022)
Delgadillo and colleagues reported a cluster randomized clinical trial of stratified versus stepped care for depression among 951 adults in psychological therapy services. Stratified allocation using a predictive model produced better depression outcomes than the comparison care strategy and was considered cost-effective within the study's economic evaluation, albeit with additional resource use. A cluster design addresses aspects of organizational implementation but requires attention to clustering, service variation, recruitment, and contamination. The result is evidence for the evaluated allocation policy in the sampled service context; it is not a license to classify individuals permanently as high- or low-intensity patients. Needs and risk can change rapidly, making regular reassessment important. Delgadillo et al. (2022)
A service that adopts stratification should specify its objective function. Maximizing short-term symptom improvement is different from reducing inequalities, maintaining continuity, or minimizing the time a patient spends receiving ineffective care. The cost of a false low-intensity recommendation can be substantial for someone at high risk of deterioration; the cost of excessive intensity includes unnecessary burden and constrained availability for others. Decision thresholds therefore require analysis of both statistical accuracy and consequences. A fair implementation study should report who is screened, who declines, who is excluded by missing information, and whether model performance differs across relevant groups and service sites. Lutz et al. (2022) Delgadillo et al. (2022)
Adaptive Care, Early Response, and Feedback
Baseline selection is only the first stage of an unfolding treatment episode. After therapy begins, symptom trajectory, alliance, attendance, side effects, and changing goals supply new evidence about the current plan. An adaptive treatment strategy states what should be assessed, at which decision points, and which modifications are triggered by particular observations. It can involve adding a module, changing intensity, switching treatment, arranging specialist consultation, or revisiting the treatment agreement. Mere collection of symptom scores is not equivalent to an effective adaptive system; the measurement must lead to decisions that are feasible, timely, and evaluated for benefit. Collins et al. (2007) Lutz et al. (2022)
Measurement-based care uses repeated standardized outcome instruments together with clinical evaluation. It can reduce the gap between therapist impression and observed trajectory and may identify clients who are at risk of deterioration. The Trier Treatment Navigator illustrates the attempt to connect pretreatment predictions, dropout-risk assessment, and adaptive feedback during care. Its development paper analyzed a clinical sample and described predictive and feedback functions, while a subsequent prospective evaluation examined clinical decision support in psychological therapy. The critical question is whether sharing this information leads to useful behavioral changes in clinicians and, ultimately, better patient outcomes, rather than whether the dashboard produces impressive predictive metrics. Lutz et al. (2019) Lutz et al. (2022, prospective evaluation)
Repeated measurement presents technical and relational issues. Patient-reported symptom scores can vary with recall periods, rapport, language, motivation, and changes in the meaning of a questionnaire item. Sudden changes may reflect events outside therapy, medication changes, or ordinary measurement error. A feedback system should distinguish signal from noise, explain uncertainty, and retain room for patient narrative. Data collection also has an opportunity cost; burdensome forms can reduce engagement precisely among people whose outcomes most require attention. An accountable adaptive framework specifies safe escalation procedures, who reviews alerts, what constitutes a clinically meaningful deviation, and how the patient participates in revising the treatment. Lutz et al. (2022) NICE (2022)
Sequential Decisions and Dynamic Treatment Regimes
A dynamic treatment regime is a sequence of decision rules that map observed histories to treatment actions. Early improvement is one tailoring variable; poor adherence, emerging comorbidity, or changing goals may also matter. A single-arm observational record cannot determine which alternative action would have been better at the same decision point, because patients who receive rescue treatment often have worse prognoses at the moment they are switched. Sequential randomization can address this problem by reallocating eligible patients at specified stages. The SMART design described by Collins, Murphy, and Strecher offers one methodological model for studying adaptive intervention sequences. Its central value is causal identification of policy options across multiple stages. Collins et al. (2007)
Dynamic analysis also requires clarity about treatment history. A person who has received eight sessions of exposure therapy differs from someone newly starting therapy even if current scores are identical. Previous response, skill acquisition, alliance, adverse experiences, medication adjustments, and events outside treatment can affect both subsequent decisions and outcomes. Models that ignore these temporal dependencies may estimate biased effects. Time-varying mediation and longitudinal causal methods can represent the changing system, but their assumptions are demanding; observed associations among successive processes do not automatically identify the consequence of intervening on one process. Kazdin (2007) Lutz et al. (2024)
The translation into ordinary services raises a policy question: how much flexibility can be supported with available data? An adaptive protocol with numerous branches may fit the development sample but be too complex for consistent clinical use. A smaller rule based on a few reliable variables may be more transportable, even when it appears less sophisticated in simulation. Evaluation should include fidelity to decision points, the time necessary to review feedback, acceptability to patients, and errors introduced by missing observations. A successful program must demonstrate that the whole policy works under realistic workforce conditions, rather than that an idealized sequence can be optimized in historical data. Lutz et al. (2022, prospective evaluation) Collins et al. (2007)
Outcome Measurement, Clinical Significance, and Treatment Burden
Treatment matching studies frequently operationalize improvement as change on depression, anxiety, or global distress scales. These instruments make comparison possible but represent only selected dimensions of recovery. A person can report fewer symptoms while still struggling with relationships, employment, sleep, quality of life, or ability to maintain gains after treatment. Conversely, meaningful functional progress may occur before a symptom score crosses a categorical threshold. A robust assessment system therefore connects disorder-specific outcomes with patient-defined goals, functioning, treatment burden, and long-term follow-up. Outcome definitions must be chosen before examining matching subgroups to avoid selecting whichever endpoint produces the most compelling positive effect. Delgadillo et al. (2022) NICE (2022)
Clinical significance differs from statistical significance. A large sample can make a very small group difference highly precise, while a favorable average can conceal poor outcomes among a vulnerable subgroup. Decision studies should report absolute rates of response and deterioration when available, alongside standardized differences and confidence intervals. Prediction studies must separate model discrimination from calibration and demonstrate net clinical benefit at plausible decision thresholds. The same principle applies to negative outcomes: treatment-related distress, dropout, delayed access to an effective modality, and deterioration are clinically relevant even if a trial's primary endpoint is average improvement. Nye et al. (2023) Schwartz et al. (2021)
The preferred outcome also depends on the stage of care. An initial treatment choice might prioritize access and early symptom relief; a later decision may prioritize functional recovery and sustained remission. For patients with complex or chronic presentations, successful care may involve measurable improvement in life participation even when full symptom remission is not immediately attained. Shared choice and feedback permit goals to be revisited, but changing the target after results are observed can obscure comparative evidence. Researchers should distinguish prespecified endpoints used for causal inference from evolving individualized goals that remain essential to practice. A transparent integrative system can respect both without conflating their purposes. Lutz et al. (2022) Windle et al. (2020)
Comparative Effectiveness and the Choice Set
The value of matching depends on the treatments available to match. Comparative effectiveness research estimates how alternative approaches perform for defined populations and outcomes. Large syntheses of adult depression trials find that several psychotherapies have efficacy, with varying study quality and control conditions. Cuijpers and colleagues' 2023 synthesis included 409 trials and more than 52,000 participants in comparisons involving cognitive behavioral therapy, other psychotherapies, medication, and combined treatment. These broad findings help establish the pool of scientifically plausible options. They cannot identify individualized superiority simply by ranking average effects. A selection model needs direct or adequately connected evidence about the particular alternatives it recommends. Cuijpers et al. (2023)
A choice set is constrained by clinician competence, local availability, treatment format, cost, language, safety, and patient willingness. An algorithm restricted to two therapies can look successful even when a third accessible treatment would better serve a subgroup. A model may also perform poorly if services change treatment protocols or eligibility rules. This is one reason externally validating a matching rule is not a routine recalibration exercise; investigators must ask whether the treatment versions and outcomes are equivalent across settings. When an option was unavailable in the development data, the model has no empirical basis for estimating its individual effect. Cohen & DeRubeis (2018) Schwartz et al. (2021)
Comparative trials should also distinguish the added value of a combined treatment from the added value of combining psychotherapy theories. For some depressive presentations, clinical guidelines include combined psychological and pharmacological treatment among viable options, but the evidence for medication–psychotherapy combinations cannot be transferred automatically to any integrated package of psychological techniques. Similarly, a trial showing that a transdiagnostic manual performs comparably to diagnosis-specific manuals demonstrates an alternative organizational solution, not that adding arbitrary extra modules increases efficacy. Integration is evaluated as a concrete intervention strategy with a defined comparator, rather than a general promise of greater comprehensiveness. NICE (2022) Barlow et al. (2017)
Moderators, Mediators, and Causal Identification
The central statistical distinction in treatment matching is between variables that predict outcome across care pathways and variables that modify a comparative treatment effect. A baseline depression score can forecast average symptom burden after treatment, while a treatment-by-severity interaction would indicate that relative benefit changes with severity. These propositions require different estimands and different validation. An investigator who discovers that severe depression predicts lower remission in both cognitive and interpersonal treatment has learned something about prognosis, but has not identified a preferred modality. Conversely, a true interaction can be clinically useful even when a moderator has weak general prognostic influence. Randomization is especially important because clinicians may direct patients with more severe or complex presentations toward particular therapies. Cohen & DeRubeis (2018) DeRubeis et al. (2014)
Mediation describes how treatment might produce change, whereas moderation identifies for whom or under which conditions treatment effects differ. Treatment-induced changes in avoidance, beliefs, emotion regulation, or social behavior can be candidate mediators; pretreatment avoidance or a history of previous therapy can be candidate moderators. Some variables play both roles at different times, making temporal ordering essential. Demonstrating that change in a candidate mediator correlates with symptom improvement provides an initial clue, but sequential causality requires addressing confounding, reverse causation, measurement overlap, and treatment effects on variables that influence both mediator and outcome. Kazdin's methodological critique argues that credible mechanism research depends on theory, precise measurement, temporal sequence, specificity, and experimental tests whenever feasible. Kazdin (2007)
For treatment selection, discovery of a mediator does not automatically imply an individual indication. A procedure may reliably affect a process without that process identifying the people for whom the procedure is comparatively preferable. An interaction involving an unreliable baseline questionnaire may also fail to replicate even if its theoretical mechanism is plausible. Researchers should state a causal diagram or substantive model before examining interaction terms; compare alternatives using appropriately adjusted analyses; distinguish exploratory discovery from confirmatory testing; and ask whether the resulting decision rule survives external validation. The purpose is to convert a clinically compelling story into a falsifiable account of what happens under competing assignments. Schwartz et al. (2021) Nye et al. (2023)
Statistical Modeling, Validation, and Calibration
The construction of an individualized treatment rule often begins with candidate variables from intake assessments: symptoms, functioning, comorbidity, previous treatment, preferences, and service constraints. A simple regression with interaction terms can provide interpretable estimates, whereas flexible machine-learning methods may represent nonlinear relations and higher-order interactions. The apparent complexity of a model is not evidence of its usefulness. High-dimensional modeling increases vulnerability to overfitting when the number of treatment comparisons and observed outcomes is small relative to the number of predictors. Researchers should separate model development from final evaluation, predefine treatment contrasts, use nested procedures for tuning, and report how missing data are handled. Schwartz et al. (2021) Cohen & DeRubeis (2018)
External validation asks whether a decision rule retains meaning when confronted with genuinely new patients, clinicians, and service conditions. Cross-validation within a single trial is useful for estimating internal predictive stability but leaves many transport problems untouched. Recalibration may be needed when baseline severity, therapy availability, or clinician expertise differs from the derivation sample. Calibration concerns whether predicted outcomes correspond to observed outcomes, ideally examined separately by assigned treatment and relevant subgroups. A model that correctly ranks who will do poorly can still misstate the magnitude of benefit from switching modalities. Such miscalibration matters whenever clinical decisions depend on a threshold in predicted benefit. DeRubeis et al. (2014) Schwartz et al. (2021)
A treatment decision rule is finally evaluated at the policy level. The relevant contrast is the outcome distribution under assignment by the new rule versus assignment by an appropriate existing policy, not simply correlation between predicted and observed symptoms. Prospective randomization to decision strategies offers direct evidence, especially when some recommendations are intentionally overridden for patient reasons. Treatment-policy analyses should report deviations and their causes rather than counting every deviation as model failure. In routine practice, decision support is successful only if clinicians can interpret it, patients can contest it, service availability supports its recommendations, and outcomes improve at an acceptable cost. Constantino et al. (2021) Delgadillo et al. (2022) Lutz et al. (2022, prospective evaluation)
Study Design and the Evaluation of Adaptation
Randomized trials of fixed therapies establish comparative mean effects, but many adaptive questions require additional design features. A factorial trial can examine whether two techniques have additive, interacting, or null contributions under a defined delivery schedule. Dismantling studies compare a treatment package with versions lacking a hypothesized active component. A sequential multiple assignment randomized trial can randomize participants again when a clinically meaningful decision point is reached, permitting comparison of alternative rescue and maintenance strategies. Each design answers a different question, and each becomes more informative when the intended target, eligibility criteria, timing, and primary outcome are clearly specified in advance. Collins et al. (2007) Kazdin (2007)
A modular psychotherapy trial may compare a prespecified adaptive system with a standard evidence-based protocol or treatment as usual. When the control condition differs in supervision, session intensity, monitoring, or clinician enthusiasm, a favorable effect cannot be assigned solely to the algorithm governing modules. Cluster randomization can capture real organizational implementation but introduces imbalances and correlations within services. Designs with repeated outcome assessments may detect earlier shifts than conventional pre-post trials, yet repeated measurements are affected by missingness, expectations, and changing interpretation. Evaluation should address fidelity, contamination, therapist clustering, recruitment, dropout, and the reproducibility of decision rules across settings. Weisz et al. (2012) Delgadillo et al. (2022)
Retrospective analyses of routine-care records remain valuable for hypothesis generation, service profiling, and model testing, provided their limitations are explicit. Patients are not allocated at random; clinician choice is influenced by unmeasured severity, urgency, availability, and patient preference. Sophisticated propensity or longitudinal methods can address measured confounding but cannot guarantee that all relevant causes of treatment assignment were recorded. For a novel matching rule, the strongest program proceeds from theoretical specification to exploratory models, external validation, prospective comparison, and ongoing postimplementation monitoring. Treating retrospective fit as final clinical proof confuses model development with intervention evaluation. Schwartz et al. (2021) Nye et al. (2023)
Evidence Quality, Replication, and Publication Bias
Personalization research contains a distinctive multiplicity problem. Studies may measure dozens of baseline characteristics, compare several therapeutic approaches, and examine numerous endpoints at multiple times. Even if each hypothesis test were individually calibrated, uncorrected searching increases the chance that seemingly convincing matching signals arise by accident. Selective publication of favorable subgroups can further distort the literature. Prespecification, transparent reporting of all candidate variables and analyses, multiplicity control where appropriate, and independent replication are safeguards against turning exploratory heterogeneity into a prescriptive clinical rule. The mixed history of Project MATCH shows why impressive theoretical rationales need carefully designed comparative tests. Project MATCH Research Group (1998) Cohen & DeRubeis (2018)
Meta-analysis can improve statistical precision, but the pooled result is interpretable only in relation to the component studies. Personalized psychotherapy reviews include different interventions, populations, and comparator conditions. The systematic synthesis by Nye and colleagues reported a small average advantage, with an attenuated effect after removing high-risk-of-bias studies. This pattern demonstrates the importance of distinguishing confidence intervals describing sampling variability from uncertainty arising through study quality, clinical heterogeneity, and selective reporting. A pooled favorable value should stimulate rigorous implementation studies, not eliminate the need to assess the particular matching policy a service proposes to use. Nye et al. (2023)
Replication has multiple levels. A treatment interaction can be re-estimated in the same randomized dataset with alternative specifications, tested in an independent trial of the same therapies, and evaluated in a different population or healthcare system. The first is a robustness test; the latter two are progressively stronger tests of generalization. A decision rule's performance may change when therapists become more experienced, services alter their intake procedures, or patients can access a newly available modality. Evidence quality therefore includes reproducible algorithms and documented treatment definitions as well as the familiar indicators of randomization, blinding of outcome assessment where feasible, attrition, and selective outcome reporting. DeRubeis et al. (2014) Schwartz et al. (2021) Constantino et al. (2021)
Cultural Context, Accessibility, and Equity
Treatment fit is shaped partly by social conditions. Language access, cultural interpretations of distress, disability accommodations, geographic distance, childcare, treatment costs, migration experiences, and trust in institutions can all influence what interventions a person can actually receive. These factors should be represented as determinants of feasibility and patient experience, not automatically converted into stereotypes about which psychological theory is appropriate for members of a social group. A person's stated preference or previous experience may be more informative than a demographic proxy. In an integrative framework, culturally responsive formulation helps identify the meaning of problems and the acceptability of procedures, while comparative treatment evidence remains necessary for claims of differential clinical efficacy. Windle et al. (2020) NICE (2022)
An algorithm trained on a service's historical records may learn that people with certain characteristics had worse outcomes because they received fewer sessions, attended overburdened clinics, or encountered inaccessible appointment times. If that association is treated as an intrinsic characteristic of the patient, the system may perpetuate the service's existing inequalities. Fairness analysis therefore examines both predictive performance and the consequences of allocation. Relevant measures include who appears in training data, whose outcomes are missing, which options are realistically available, and whether a decision rule alters access and waiting time. A formally identical rule can have unequal consequences when capacity is distributed unevenly. Schwartz et al. (2021) Lutz et al. (2022)
Accessibility itself can be an object of personalization. A person may prefer remote therapy because travel is burdensome, or in-person care because privacy at home is inadequate. Translating a protocol to another language involves more than literal wording when examples, distress concepts, and social expectations differ. However, a favorable adaptation cannot be assumed to preserve the efficacy of every source treatment component. Cultural responsiveness is most credible when goals, procedures, engagement, and outcomes are examined collaboratively and empirically. Researchers should distinguish adaptation for access from evidence that one adapted model outperforms an otherwise equivalent nonadapted model. NICE (2022) Windle et al. (2020)
Clinical Governance, Consent, and Potential Harm
Psychotherapy integration involves clinical responsibility for both the selected intervention and the process by which it is chosen. Before recommending a combination, clinicians should consider established evidence for the presenting condition, severity and acute risk, the patient's history and preferences, provider competence, expected burden, and plausible adverse outcomes. Treatments with evidence from controlled trials may still be inappropriate when implemented without the necessary training or safety procedures. Adding more techniques may increase confusion, extend ineffective care, or obscure which component caused an adverse response. Integrative practice therefore requires explicit goals, documentation of rationale, appropriate supervision, and monitoring that can detect deterioration. NICE (2022) Kazdin (2007)
Consent in personalized treatment has two connected dimensions. Patients need an understandable explanation of the proposed therapy options and the grounds for selecting among them. When a model contributes to the recommendation, they also need to know which information is used, what degree of confidence is reasonable, and whether human review and alternative choices remain available. A probabilistic prediction should not be communicated as an individualized certainty. The right to decline a recommendation is compatible with evidence-based care, and treatment selection should not transform the patient into a passive recipient of an opaque optimization process. Windle et al. (2020) Lutz et al. (2022)
Research quality and patient safety intersect when studies omit people at high acute risk, experiencing major social instability, or presenting complex comorbidities. Findings in an outpatient depression trial cannot automatically be applied to psychiatric emergencies or to situations requiring specialized medical and safeguarding decisions. Clinical protocols must retain procedures for assessment and referral outside the decision aid's validated scope. Reporting of deterioration, discontinuation, and differential access is essential because an apparent improvement in average scores can coexist with clinically important harm for particular groups. A responsible scientific synthesis explicitly marks the boundary between evaluated treatment populations and future clinical generalizations. Delgadillo et al. (2022) Nye et al. (2023) NICE (2022)
Depression as a Test Case for Integrative Selection
Depression is a central research setting for treatment matching because several established intervention families are available and response is heterogeneous. Cognitive behavioral therapy, interpersonal approaches, behavioral activation, other structured psychotherapies, pharmacotherapy, and selected combined treatments all enter comparative decisions, depending on severity, access, history, and preference. The scientific problem is not simply to determine whether psychotherapy works; it is to estimate the consequences of choosing one viable plan over another for a particular presentation. A broad meta-analysis of cognitive behavioral therapy included 409 trials and 52,702 participants across comparison categories and illustrates the strength and complexity of the mean-effect evidence base. Large aggregate evidence supports treatment choice, while individualized selection still requires credible moderator and policy-level evaluation. Cuijpers et al. (2023) Cohen & DeRubeis (2018)
The Personalized Advantage Index demonstrated how baseline variables from randomized depression treatment comparisons could be translated into expected differences between cognitive therapy and antidepressant medication. Its importance was methodological: potential treatment outcomes were modeled for an individual and the expected difference was used to identify the comparatively favorable option. This formulation places the relevant alternatives at the center of decision-making. It also highlights a transfer limitation, because a model derived from one treatment pair and a selected research sample does not automatically apply to a different psychotherapy, a new medication, or a patient whose characteristics were rare in the source trial. A clinical decision process must distinguish validated knowledge about average benefits from tentative person-specific estimates. DeRubeis et al. (2014) Cohen & DeRubeis (2018)
Evidence-informed integrative work with depression may target depressed mood, behavioral avoidance, social withdrawal, interpersonal loss, self-criticism, sleep disruption, and barriers to engagement. Treatment may be organized around one coherent manual or a structured sequence of procedures justified by the clinical formulation. A scientifically adequate formulation specifies why an intervention is expected to help and how progress will be checked. NICE guidance on depression in adults recognizes shared decision-making and an array of interventions, including choices tailored to prior response and the severity of the presentation. The guideline provides a clinical framework while stopping short of endorsing every novel personalized algorithm as an established treatment. NICE (2022) Lutz et al. (2022)
Anxiety and Transdiagnostic Emotional Disorders
Anxiety presentations offer a second major application because avoidance, threat interpretation, attentional processes, and regulation of emotion may appear across several diagnostic groups. The Unified Protocol organizes treatment around emotional processes rather than only a single diagnostic category. In the Barlow and colleagues randomized trial, the transdiagnostic program was evaluated against diagnosis-specific protocols in adults with anxiety disorders, establishing a direct comparative test of one integrated, process-oriented package. The findings support the possibility that a shared treatment architecture can provide outcomes comparable with several established alternatives within the studied context. The design does not show that every person with anxiety symptoms requires transdiagnostic therapy or that arbitrary combinations of exposure and emotion-focused methods are equivalent to the tested protocol. Barlow et al. (2017)
Matching in anxiety care can concern whether a transdiagnostic approach, a diagnosis-specific intervention, a particular delivery format, or additional support will best address an individual's presentation. Co-occurring depression and emotion dysregulation may influence feasibility, but they should not be treated as proven moderators without evidence of treatment-by-characteristic interactions. Early therapeutic engagement and changes in avoidance can help clinicians assess progress while remaining separate from the stronger causal question of whether a different treatment would have produced a better result at that time. Trial protocols provide reference points for competence and fidelity; adaptation should preserve the procedures required for evidence-supported exposure, learning, and treatment safety. Barlow et al. (2017) Kazdin (2007)
At the service level, the advantage of a transdiagnostic program can be logistical as well as clinical. A smaller set of training protocols may make care accessible to clinicians serving patients with overlapping emotional difficulties. Such efficiencies are valuable outcomes but should be measured rather than assumed: additional modules, longer assessments, or complex supervision requirements can offset the efficiencies of a unified manual. A balanced evaluation reports treatment outcomes and the organizational costs of adopting the program, including retention and access for patients who would otherwise wait for specialty care. Farchione et al. (2024) Lutz et al. (2022)
Youth Presentations and Multidomain Modular Treatment
Children and adolescents may present overlapping anxiety, depressive symptoms, behavioral conflict, school difficulties, and family stressors. These interdependencies challenge fixed one-problem manuals and make a structured modular approach attractive. The randomized effectiveness trial by Weisz and colleagues tested a modular treatment derived from established youth therapy protocols against standard manualized therapy and usual care. Its design directly evaluated the organization and sequencing of evidence-based elements. The research contribution lies in asking whether an adaptable system can address multiple domains in routine clinical circumstances; it does not establish a universal best sequence of modules for every family. Weisz et al. (2012)
Youth treatment matching must also account for the person's developmental stage, caregiver involvement, school and community context, capacity to attend, and the young person's preferences. A treatment plan may involve caregiver sessions or environmental changes when those are supported by the underlying treatment model. These are substantive modifications with implications for consent and therapeutic goals, rather than interchangeable administrative details. Evidence generated in one age range, problem mix, or health system must be treated as context-dependent. Measuring effects on both symptoms and family or school functioning helps prevent a modular program from appearing successful solely because one questionnaire improves while a different impairment remains unaddressed. Weisz et al. (2012)
The methodological opportunity in modular youth care is to make adaptive decisions observable. If a clinician changes the sequence after new information from family or school emerges, the decision point and grounds for revision can be recorded and potentially studied. A more ambitious trial could randomize competing module strategies at those decision points, estimate effects on engagement and functioning, and examine whether greater flexibility benefits some groups at the expense of others. This agenda connects responsive clinical practice with the logic of factorial experiments and sequential assignment designs. Collins et al. (2007) Weisz et al. (2012)
Complexity, Comorbidity, and Prior Treatment Experience
Many people enter psychotherapy with more than one clinically relevant problem or with a history of incomplete response. A diagnostic label captures part of the presentation but often fails to describe functional impairment, previous interventions, interpersonal context, and practical obstacles. Integrative treatment matching must therefore accommodate competing priorities. One problem may require an immediate evidence-supported intervention, while another calls for a change in delivery format, coordination with medical care, or a later therapeutic module. The scientific challenge is to distinguish a coherent staged plan from indefinite treatment drift. Prior treatment response is clinically informative but can be confounded by adherence, therapist competence, care accessibility, and spontaneous change. NICE (2022) Lutz et al. (2022)
Comorbidity is particularly vulnerable to interpretive error. An individual with anxiety and depression might appear to benefit from a transdiagnostic approach because it addresses overlapping processes, but an apparent interaction may reflect baseline severity, access to a qualified therapist, or different outcome measurement. Without an appropriate comparative study, these are hypotheses rather than established indications. Treatment selection should therefore specify whether it seeks to improve symptoms across all domains, prioritize the principal impairment, reduce risk, or enhance the person's functioning. The evidence for transdiagnostic approaches is encouraging in defined settings, and its transport to broader clinical complexity remains an empirical question. Barlow et al. (2017) Farchione et al. (2024)
Prior failed treatment can motivate a new plan but does not, by itself, reveal why the previous intervention was ineffective. Inadequate dose, poor treatment fit, therapist limitations, inconsistent attendance, unforeseen external events, and a mismatch between the measured outcome and the patient's goal can produce similar apparent nonresponse. A rigorous reassessment reconstructs what was actually delivered, what changed, what remained difficult, and which constraints still apply. Repeating an ineffective course without modification may waste resources, while rapid switching based on normal week-to-week variation may prevent an effective intervention from having sufficient opportunity to work. This tradeoff underlines the importance of clinically interpretable feedback thresholds and longitudinal outcome assessment. Lutz et al. (2019) Lutz et al. (2022)
Implementation in Mental Health Systems
A treatment-matching tool becomes a real intervention only when its recommendations affect clinical decisions within a service. Implementation includes screening, clinician training, information systems, interpretation of risk and benefit estimates, patient consultation, scheduling, and procedures for revising a recommendation. The trial of stratified care versus stepped care in depression examined such a policy at the service level, while randomized therapist matching tested assignment based on historical therapist-specific outcome patterns. Both lines of work show that treatment selection can be evaluated as an organizational action rather than remaining a conceptual promise about individualized therapy. Neither result automatically validates every different triage or provider-allocation method. Delgadillo et al. (2022) Constantino et al. (2021)
Service capacity changes the outcome of matching. Recommending a particular therapist has little practical effect when appointments are unavailable for months, and recommending an intensive program may be counterproductive if the service cannot deliver it competently. Implementation evaluation should therefore track wait times, uptake, deviations from the recommended pathway, treatment continuity, staff burden, cost, and patient experience. Selective missing data are particularly consequential: clients who decline treatment or leave early may be absent from a model trained only on completed episodes. An apparently successful allocation rule can thereby reflect the characteristics of people retained in care rather than improved outcomes for everyone referred. Lutz et al. (2022) Windle et al. (2020)
Continuous quality improvement offers a defensible method of refinement once core clinical safety and patient rights are protected. A service can audit whether predicted advantages correspond to observed outcomes in later cohorts and whether recommendation quality is stable across staff and demographic subgroups. Changes to the model should be documented with versioning, recalibration tests, and comparison to the previous policy. Outcomes from a model's development period are insufficient proof of continuing validity when therapies, referral criteria, or the population change. Rigorous governance treats a selection system as a clinical intervention whose effects on behavior and access need ongoing evaluation. Schwartz et al. (2021) Lutz et al. (2022, prospective evaluation)
Artificial Decision Support in the Artificial Era
The Artificial Era expands psychotherapy research by making nonbiological computation part of how clinical information is organized, compared, and used. Statistical models and increasingly capable artificial systems can summarize patient-reported outcome trajectories, estimate differential treatment response, classify engagement risks, and present competing care pathways for review. Their legitimate scientific role is defined by observable performance within an accountable clinical process. A system able to generate plausible therapeutic language is not thereby validated as a therapist or as a reliable allocator of medical resources. A model that recommends an intervention must demonstrate performance on the defined treatment contrast, patient population, outcome, and deployment setting. Lutz et al. (2019) Schwartz et al. (2021)
Clinical matching algorithms are particularly exposed to distribution shift and automation bias. Prediction can degrade when new services use different measures, people self-select into care differently, clinicians deliver altered protocols, or a model encounters populations absent from training data. Clinicians may give undue weight to a confident numerical recommendation even when the underlying contrast is uncertain. Conversely, opaque recommendations may be ignored and provide no benefit even when they contain a modest statistical signal. System evaluations should therefore record actual usage, interpretation, override decisions, and downstream patient outcomes, rather than treating a technically successful prediction as clinical effectiveness. Lutz et al. (2022, prospective evaluation) Nye et al. (2023)
Human–artificial relations introduce further empirical questions about trust, consent, disclosure, continuity of care, and the distribution of responsibility when technology influences treatment choice. Patients should retain access to understandable explanations and qualified human clinical assessment, particularly when safety-sensitive decisions are involved. Research on artificial cognition and artificial agency can expand the conceptual vocabulary used to describe computational decision systems, while clinical claims still require direct evidence from patient outcomes. The core scientific contribution of artificial systems in treatment matching lies in testable decision assistance and transparent information integration; stronger philosophical interpretations belong to their own research programs. Lutz et al. (2022) Cuijpers et al. (2026)
Critical Comparison of Integration and Matching Strategies
The research literature supports a structured comparison of strategies rather than a single hierarchy of integration. Theoretical integration offers conceptual breadth but needs operational definitions and specific predictions before its mechanisms can be evaluated. Technical eclecticism offers responsiveness and pragmatic choices but may weaken reproducibility when procedures are borrowed without a coherent selection rule. Transdiagnostic treatment replaces some diagnosis-specific boundaries with process-oriented organization; its evidence derives from concrete protocols and trials rather than from the general attractiveness of a shared mechanism. Modularity makes conditional sequencing explicit but adds implementation complexity. Baseline statistical matching estimates comparative treatment advantage, whereas feedback-informed care revises plans using emerging trajectories. Each strategy answers a different decision problem, and none should inherit the favorable results of another merely by sharing the adjective 'integrative.' Norcross & Goldfried (2005) Barlow et al. (2017) Weisz et al. (2012)
The quality of evidence differs by strategy and inferential target. A randomized trial of modular versus usual care can evaluate the modular system's effectiveness while leaving the individual importance of specific modules unidentified. A predictor derived from randomized treatment arms can estimate potential comparative advantage without establishing that acting on its recommendation improves outcomes in practice. A randomized therapist-allocation trial tests a service policy more directly but may rely on therapist performance histories unavailable elsewhere. A meta-analysis showing a small benefit for personalized care aggregates multiple forms of personalization. These distinctions explain why the most defensible synthesis contains several conditional conclusions rather than a sweeping claim that combining methods is inherently superior. Weisz et al. (2012) DeRubeis et al. (2014) Constantino et al. (2021) Nye et al. (2023)
Across strategies, an accountable clinical decision can be described through the same elements: an explicitly defined set of options, a patient and context description, an intended outcome, an evidence-supported selection or adaptation rule, and a method for evaluating what happens after the choice. That structure connects integrative concepts to experimental testing. It also identifies the limitations of informal reasoning: without recording the comparator and the decision rule, clinical improvement following a customized plan cannot establish that customization caused the improvement. Integrative psychotherapy develops scientifically when its flexibility becomes a specified, replicable, and falsifiable feature of treatment organization. Kazdin (2007) Collins et al. (2007) Cohen & DeRubeis (2018)
What the Empirical Literature Establishes
Several findings have comparatively direct support. A randomized trial has tested a modular psychotherapy system for multiple youth problem domains, demonstrating that a structured flexible arrangement can be subjected to rigorous comparative evaluation. A randomized adult anxiety trial has found the Unified Protocol and diagnosis-specific protocols equivalent on the designated outcome within the studied population, supporting transdiagnostic delivery as an evidence-informed alternative. A randomized clinical trial has shown benefit from therapist-to-patient matching based on historical domain-specific effectiveness in the evaluated outpatient service. A cluster randomized depression trial has tested stratified allocation against stepped care. These are demonstrations of the tested protocols and policies under particular conditions, not general authorization for all flexible or algorithmic systems. Weisz et al. (2012) Barlow et al. (2017) Constantino et al. (2021) Delgadillo et al. (2022)
Other findings are promising while more uncertain. The Personalized Advantage Index and related prediction studies show how patient-level differential treatment response can be modeled, but their transport to novel services and prospective policy impact require additional evidence. A routine-care holdout analysis illustrates limited overall generalization despite signals among people with larger predicted differences. Patient preference accommodation has comparatively consistent links with dropout and alliance but more variable evidence for direct symptom advantage. The 2023 meta-analysis of personalized interventions found a modest pooled improvement with substantial risk-of-bias concerns across the underlying studies. Collectively, these findings establish an active field of scientific investigation with variable maturity across intervention classes. DeRubeis et al. (2014) Schwartz et al. (2021) Windle et al. (2020) Nye et al. (2023)
The combined literature does not support a universal rule that the most complex or technologically advanced matching system is best. A simple, shared decision informed by strong average-effect evidence may be superior to a poorly calibrated algorithm for a population unlike its derivation sample. A protocol-based intervention can be preferable to an elaborate eclectic plan when fidelity and clinician expertise are decisive. Conversely, a fixed protocol can underperform when service organization fails to respond to clear signs of nonengagement or nonresponse. These are empirically testable contingencies, and the choice among strategies should remain open to prospective comparison rather than being settled by conceptual allegiance. Cohen & DeRubeis (2018) Lutz et al. (2022) NICE (2022)
Unresolved Questions and Research Priorities
An immediate priority is prospective validation of matching policies across independent settings. Studies should recruit populations reflecting the full range of people seeking care, specify treatment alternatives before analysis, and measure clinically meaningful benefit with attention to deterioration and dropout. Where prospective trials are feasible, randomization to a selection rule versus a defensible comparator can establish whether decision support actually improves treatment outcomes. Researchers should assess the impact of overriding the recommendation for personal or clinical reasons and evaluate whether its benefits persist when therapist availability changes. Transparent reporting of unsuccessful rules is necessary to prevent selective enthusiasm around favorable retrospective subgroup findings. Nye et al. (2023) Schwartz et al. (2021)
A second priority is better evidence about adaptive sequences and component interactions. Clinicians frequently revise plans after early symptoms change, but the optimal timing of switching, intensifying, or adding modules remains poorly specified for many conditions. Sequential randomization, factorial optimization, and well-measured longitudinal designs can identify which decisions matter and when. The mechanisms linking alliance, expectancy, skill acquisition, and targeted behavioral change to later outcomes also require experimental work beyond correlations. These programs should compare costs and burden, because more elaborate measurement and intervention systems may consume time that could otherwise expand access to competent care. Collins et al. (2007) Kazdin (2007) Lutz et al. (2022, prospective evaluation)
A third priority concerns fairness and interpretability. Researchers should examine whether matching rules yield comparable calibration and clinical benefit across sites, age groups, language communities, levels of disability, and different pathways into care. Patient-defined goals and preferences require appropriate measurement without allowing historical access inequalities to become proxies for treatment suitability. Decision systems also need rigorous privacy governance, safeguards against inappropriate automation, and comprehensible explanations of uncertainty. As artificial systems acquire greater roles in clinical information processing, the evidentiary standard for their influence on care should remain prospective benefit, transparent limitations, and accountable clinical action. Windle et al. (2020) Cuijpers et al. (2026) Lutz et al. (2022)
Conclusion
Integrative psychotherapy and treatment matching form a coherent research domain when treatment integration is understood as a specified organization of therapeutic knowledge and treatment matching as a causal decision problem among credible care options. The scientific challenge begins with accurate definitions of psychological processes and alternative treatments, extends through randomized comparisons, moderation research, and repeated measurement, and culminates in determining whether an adaptive selection policy improves outcomes for patients in real services. The field's historical movements—from theoretical integration and technical eclecticism to transdiagnostic manuals, modular interventions, personalized prediction, and clinician feedback—contribute complementary tools for this task. Norcross & Goldfried (2005) Cohen & DeRubeis (2018) Lutz et al. (2022)
The strongest evidence supports carefully bounded claims. Several structured integrative and adaptive approaches have performed well in specific randomized evaluations; patient preference matters for engagement; and algorithmic approaches can inform selection while facing demanding standards for external validation. Models based on baseline predictors, therapist performance, or ongoing trajectories answer different questions and require distinct trials and outcome measures. Prediction of a difficult clinical course is not equivalent to evidence about the best alternative treatment, and successful retrospective prediction is not identical to demonstrated benefit from deploying a recommendation. The validity of clinical matching rests on a credible comparator and the observed consequences of a decision. Barlow et al. (2017) Constantino et al. (2021) Nye et al. (2023)
Future integrative psychotherapy should strengthen its relationship between theory, evidence, personal meaning, service capacity, and accountable decisions. More precise experiments and prospective evaluations can reveal when a stable manual is enough, when treatment components should be reorganized, and when measurement or personalized allocation changes what care achieves. Artificial decision support adds computational capacity to this research architecture while increasing the importance of transparency, uncertainty, equity, and human clinical responsibility. A mature science of treatment matching will be judged by validated improvement in patient outcomes, safety, participation, and durability of benefit across the contexts in which care is actually delivered. Schwartz et al. (2021) Lutz et al. (2022, prospective evaluation) NICE (2022)
References
- Barlow, D. H., Farchione, T. J., Bullis, J. R., Gallagher, M. W., Murray-Latin, H., Sauer-Zavala, S., Bentley, K. H., Thompson-Hollands, J., Conklin, L. R., Boswell, J. F., Ametaj, A., Carl, J. R., Boettcher, H. T., & Cassiello-Robbins, C. (2017). The Unified Protocol for Transdiagnostic Treatment of Emotional Disorders compared with diagnosis-specific protocols for anxiety disorders: A randomized clinical trial. JAMA Psychiatry, 74(9), 875–884. https://doi.org/10.1001/jamapsychiatry.2017.2164
- Cohen, Z. D., & DeRubeis, R. J. (2018). Treatment selection in depression. Annual Review of Clinical Psychology, 14, 209–236. https://doi.org/10.1146/annurev-clinpsy-050817-084746
- Collins, L. M., Murphy, S. A., & Strecher, V. (2007). The multiphase optimization strategy (MOST) and the sequential multiple assignment randomized trial (SMART): New methods for more potent eHealth interventions. American Journal of Preventive Medicine, 32(5 Suppl.), S112–S118. https://doi.org/10.1016/j.amepre.2007.01.022
- Constantino, M. J., Boswell, J. F., Coyne, A. E., Swales, T. P., & Kraus, D. R. (2021). Effect of matching therapists to patients vs assignment as usual on adult psychotherapy outcomes: A randomized clinical trial. JAMA Psychiatry, 78(9), 960–969. https://doi.org/10.1001/jamapsychiatry.2021.1221
- Cuijpers, P., Harrer, M., & Furukawa, T. A. (2026). Innovations to improve outcomes and uptake of psychotherapies for mental disorders: A state-of-the-art review. World Psychiatry, 25(1), 4–33. https://doi.org/10.1002/wps.70002
- Cuijpers, P., Miguel, C., Harrer, M., Plessen, C. Y., Ciharova, M., Ebert, D., & Karyotaki, E. (2023). Cognitive behavior therapy vs. control conditions, other psychotherapies, pharmacotherapies and combined treatment for depression: A comprehensive meta-analysis including 409 trials with 52,702 patients. World Psychiatry, 22(1), 105–115. https://doi.org/10.1002/wps.21069
- Delgadillo, J., Ali, S., Fleck, K., Agnew, C., Southgate, A., Parkhouse, L., Cohen, Z. D., DeRubeis, R. J., & Barkham, M. (2022). Stratified care vs stepped care for depression: A cluster randomized clinical trial. JAMA Psychiatry, 79(2), 101–108. https://doi.org/10.1001/jamapsychiatry.2021.3539
- DeRubeis, R. J., Cohen, Z. D., Forand, N. R., Fournier, J. C., Gelfand, L. A., & Lorenzo-Luaces, L. (2014). The Personalized Advantage Index: Translating research on prediction into individualized treatment recommendations. A demonstration. PLOS ONE, 9(1), e83875. https://doi.org/10.1371/journal.pone.0083875
- Farchione, T. J., Long, L. J., Spencer-Laitt, D., Moreno, J., & Barlow, D. H. (2024). State of the science: The Unified Protocol for Transdiagnostic Treatment of Emotional Disorders. Behavior Therapy, 55(6), 1189–1204. https://doi.org/10.1016/j.beth.2024.04.007
- Flückiger, C., Del Re, A. C., Wampold, B. E., & Horvath, A. O. (2018). The alliance in adult psychotherapy: A meta-analytic synthesis. Psychotherapy, 55(4), 316–340. https://doi.org/10.1037/pst0000172
- Flückiger, C., Del Re, A. C., Wlodasch, D., Horvath, A. O., Solomonov, N., & Wampold, B. E. (2020). Assessing the alliance-outcome association adjusted for patient characteristics and treatment processes: A meta-analytic summary of direct comparisons. Journal of Counseling Psychology, 67(6), 706–711. https://doi.org/10.1037/cou0000424
- Kazdin, A. E. (2007). Mediators and mechanisms of change in psychotherapy research. Annual Review of Clinical Psychology, 3, 1–27. https://doi.org/10.1146/annurev.clinpsy.3.022806.091432
- Lutz, W., Deisenhofer, A.-K., Rubel, J. A., Bennemann, B., Giesemann, J., Poster, K., & Schwartz, B. (2022). Prospective evaluation of a clinical decision support system in psychological therapy. Journal of Consulting and Clinical Psychology, 90(1), 90–106. https://doi.org/10.1037/ccp0000642
- Lutz, W., Rubel, J. A., Schwartz, B., Schilling, V., & Deisenhofer, A.-K. (2019). Towards integrating personalized feedback research into clinical practice: Development of the Trier Treatment Navigator (TTN). Behaviour Research and Therapy, 120, Article 103438. https://doi.org/10.1016/j.brat.2019.103438
- Lutz, W., Schwartz, B., & Delgadillo, J. (2022). Measurement-based and data-informed psychological therapy. Annual Review of Clinical Psychology, 18, 71–98. https://doi.org/10.1146/annurev-clinpsy-071720-014821
- Lutz, W., Schwartz, B., Deisenhofer, A.-K., Schaffrath, J., Eberhardt, S. T., Bommer, J., Vehlen, A., Moggia, D., Poster, K., Weinmann-Lutz, B., Rubel, J. A., & Hehlmann, M. I. (2024). From theory to practice: A transtheoretical treatment and training model (4TM). Clinical Psychology in Europe, 6(Special Issue), Article e12421. https://doi.org/10.32872/cpe.12421
- National Institute for Health and Care Excellence. (2022). Depression in adults: Treatment and management (NICE guideline NG222). https://www.nice.org.uk/guidance/ng222/chapter/Recommendations
- Norcross, J. C., & Goldfried, M. R. (Eds.). (2005). Handbook of psychotherapy integration (2nd ed.). Oxford University Press. https://doi.org/10.1093/med:psych/9780195165791.001.0001
- Norcross, J. C., Karpiak, C. P., & Lister, K. M. (2005). What's an integrationist? A study of self-identified integrative and (occasionally) eclectic psychologists. Journal of Clinical Psychology, 61(12), 1587–1594. https://doi.org/10.1002/jclp.20203
- Nye, A., Delgadillo, J., & Barkham, M. (2023). Efficacy of personalized psychological interventions: A systematic review and meta-analysis. Journal of Consulting and Clinical Psychology, 91(7), 389–397. https://doi.org/10.1037/ccp0000820
- Project MATCH Research Group. (1993). Project MATCH (Matching Alcoholism Treatment to Client Heterogeneity): Rationale and methods for a multisite clinical trial matching patients to alcoholism treatment. Alcoholism: Clinical and Experimental Research, 17(6), 1130–1145. https://doi.org/10.1111/j.1530-0277.1993.tb05219.x
- Project MATCH Research Group. (1998). Matching alcoholism treatments to client heterogeneity: Project MATCH three-year drinking outcomes. Alcoholism: Clinical and Experimental Research, 22(6), 1300–1311. https://doi.org/10.1111/j.1530-0277.1998.tb03912.x
- Schwartz, B., Cohen, Z. D., Rubel, J. A., Zimmermann, D., Wittmann, W. W., & Lutz, W. (2021). Personalized treatment selection in routine care: Integrating machine learning and statistical algorithms to recommend cognitive behavioral or psychodynamic therapy. Psychotherapy Research, 31(1), 33–51. https://doi.org/10.1080/10503307.2020.1769219
- Swift, J. K., & Callahan, J. L. (2009). The impact of client treatment preferences on outcome: A meta-analysis. Journal of Clinical Psychology, 65(4), 368–381. https://doi.org/10.1002/jclp.20553
- Swift, J. K., Callahan, J. L., Cooper, M., & Parkin, S. R. (2018). The impact of accommodating client preference in psychotherapy: A meta-analysis. Journal of Clinical Psychology, 74(11), 1924–1937. https://doi.org/10.1002/jclp.22680
- van Bronswijk, S. C., Lemmens, L. H. J. M., Huibers, M. J. H., & Peeters, F. P. M. L. (2021). Selecting the optimal treatment for a depressed individual: Clinical judgment or statistical prediction? Journal of Affective Disorders, 279, 149–157. https://doi.org/10.1016/j.jad.2020.09.135
- Weisz, J. R., Chorpita, B. F., Palinkas, L. A., Schoenwald, S. K., Miranda, J., Bearman, S. K., Daleiden, E. L., Ugueto, A. M., Ho, A., Martin, J., Gray, J., Alleyne, A., Langer, D. A., Southam-Gerow, M. A., Gibbons, R. D., & Research Network on Youth Mental Health. (2012). Testing standard and modular designs for psychotherapy treating depression, anxiety, and conduct problems in youth: A randomized effectiveness trial. Archives of General Psychiatry, 69(3), 274–282. https://doi.org/10.1001/archgenpsychiatry.2011.147
- Windle, E., Tee, H., Sabitova, A., Jovanovic, N., Priebe, S., & Carr, C. (2020). Association of patient treatment preference with dropout and clinical outcomes in adult psychosocial mental health interventions: A systematic review and meta-analysis. JAMA Psychiatry, 77(3), 294–302. https://doi.org/10.1001/jamapsychiatry.2019.3750
- Zilcha-Mano, S., & Fisher, H. (2022). Distinct roles of state-like and trait-like patient–therapist alliance in psychotherapy. Nature Reviews Psychology, 1, 194–210. https://doi.org/10.1038/s44159-022-00029-z