A clinically actionable framework for personalizing iCBT to improve depression outcomes

Abstract

Internet-based cognitive behavioral therapy (iCBT) is effective for depression, but its impact is constrained by low engagement and modest response rates. Personalization may address these limitations, yet a gap remains between research evidence and clinically actionable implementation. This narrative review synthesizes evidence on personalization in iCBT for depression using a three-stage framework: pre-implementation optimization, stratifying treatment based on patient characteristics, and dynamically adapting therapy using progress monitoring. Evidence was evaluated using principles of evidence grading, with attention to the volume, consistency, and directness of findings for each stage. The strongest evidence supports pre-implementation optimization of engagement and stratification of initial support based on baseline severity, treatment history, and related clinical characteristics. Evidence for dynamic adaptation is promising but less developed, with support for early identification of nonresponse and adjustment of treatment intensity, but limited iCBT-specific trials testing adaptive treatment strategies in depression. Across stages, engagement and clinical outcomes are related but distinct targets for personalization. Emerging research on responsiveness, progress monitoring, and digital biomarkers offers future opportunities for more precise and scalable personalization. More rigorous depression-specific iCBT studies are needed to determine when, how, and for whom personalized interventions improve engagement and clinical outcomes.

Introduction

Internet-delivered cognitive behavioral therapy (iCBT) for depression is one of the most widely used digital mental health interventions. iCBT adapts the core components of cognitive behavioral therapy (CBT), a structured, time-limited, evidence-based approach targeting maladaptive thoughts and behaviors, to online delivery via computers or mobile devices¹. Over the past two decades, scalable iCBT programs have been developed as alternatives to clinician-delivered CBT² with meta-analytic evidence indicating comparable efficacy to face-to-face CBT^3–5. iCBT is delivered with therapist or coach support, or as a fully self-guided intervention⁶ and is available through multiple access models, including subscription-based platforms, and publicly funded services. iCBT can be used as a stand-alone treatment or integrated into blended care alongside traditional CBT.

While iCBT is effective on average, it has only moderate real-world impact for a substantial portion of individuals with depression who use it. Up to 50% of patients may not achieve sufficient symptom relief (i.e., they continue to report substantial symptom levels following treatment), highlighting that outcomes are moderate and variable across individuals^5,7,8. Indeed, large meta-analyses show small to moderate effect sizes, with many patients not reaching clinical response thresholds despite statistically significant improvements for the sample as a whole⁹. Some iCBT trials report number needed to treat (NNT) around 8 or higher, meaning many participants do not experience a clinically meaningful reduction in symptoms⁹.

Moreover, many patients drop out of iCBT prematurely, even in research settings¹⁰, with reported dropout rates in guided iCBT for depression ranging from 0% to 75% with an average of around 32% in published trials¹¹. Therefore, increasing engagement and improving an individual’s treatment outcomes is crucial in maximizing the potential of iCBT as a scalable and efficacious intervention for depression.

Approaches to understanding iCBT treatment engagement and outcomes

A range of approaches have been proposed to improve iCBT outcomes, including procedures to optimize the iCBT treatment package¹², to personalize iCBT based on patient characteristics (e.g. ref. ¹³), or to respond dynamically to individuals with poor iCBT treatment trajectories¹⁴). Conceptual approaches have identified personalization variables that might be used to design and evaluate digital mental health intervention (DMHI) apps and websites (see refs. ^15–17). However, these conceptual models do not focus on iCBT specifically, and do not directly link to current research evidence.

In short, despite the vast empirical literature on iCBT, clinicians, service developers and providers are lacking actionable insights from the research-evidence. This is particularly concerning because clinicians have been reported to be alienated by the recent technological advancements associated with iCBT. For example, they are unsure of how to make use of advanced DMHIs¹⁸, or how to interpret rapid developments in the field of precision psychiatry, with its specified algorithmic methods for prediction and individualized iCBT treatment (e.g., ref. ¹⁹).

Aims and methods

We aimed to provide a clinically meaningful synthesis of research evidence on personalization of iCBT for depression. Through extensive discussion informed by our domain expertise, research knowledge, and clinical experience we aimed to develop a three-stage personalization framework to organize the evidence: 1) Pre-implementation optimization of the iCBT package and delivery context; 2) Stratification based on patient characteristics at treatment entry; and 3) Responsiveness during treatment based on patient progress. These stages differ in their level of action: pre-implementation optimization concerns design and service-delivery decisions made before patient use; stratification concerns initial patient-treatment matching; and responsiveness concerns adaptations made during treatment, either within the iCBT program itself or in the surrounding care pathway. To achieve this, we reviewed relevant iCBT research evidence and supplemented our review with indirect research evidence that reported on DMHIs or face-to-face CBT more broadly. This inclusion was necessitated by the scant evidence specifically for iCBT, but also by the likely relevance of the findings for iCBT.

We evaluated the strength of research evidence for each personalization strategy using a narrative synthesis informed by principles of evidence grading, such as those outlined in the GRADE framework²⁰. Specifically, we considered the volume, consistency, and directness of the available literature to categorize evidence as follows: strong evidence; weak or mixed evidence; no evidence or negative evidence; and indirect evidence. ‘Strong evidence’ was used when the available literature was substantial (e.g., multiple randomized trials and/or meta-analyses) and findings were consistent in direction. Weak or mixed evidence was used if the findings were from only a few research trials, or research trials identified both positive and negative outcomes for the variable. No evidence was used when either no research trials were available, or the research trials yielded a lack of an association. Direct evidence referred to findings from the iCBT literature. Indirect evidence referred to evidence from CBT (not iCBT), and DMHI (which might include iCBT but also other approaches).

In our review we focused on patient engagement and clinical outcomes. For iCBT to be effective, the patient must be willing to try it and to continue to engage with it²¹, as greater engagement with iCBT (i.e., completing more sessions) has been associated with better outcomes, including lower depressive symptoms and a higher likelihood of treatment response⁹. However, this association is correlational and may reflect both treatment effects and the tendency for participants who improve to remain engaged. Across the reviewed studies, engagement-related constructs were operationalized inconsistently, using terms such as adherence, participation, number of completed sessions or modules, or the time spent, often without clear definitions. Given the limited and non-comparable evidence for each construct, we collapsed these indicators under the umbrella term engagement, rather than attempting distinctions that could not be meaningfully evaluated. Supplementary Note 1 provides definitions and relevance of several of the constructs used in our review.

Pre-implementation optimization

Before iCBT reaches the patient, the intervention package and delivery context can be optimized through design, feature selection, cultural adaptation, co-creation, and decisions about the level and type of support. Table 1 summarizes the research on the key optimization variables for improving engagement and effectiveness, and relevant insights for clinicians. The detailed narrative review of the evidence for Table 1 is available in Supplementary Note 2.

Table 1.

Pre-implementation optimization: variables that determine how the therapy package is delivered

Personalization variables	Effect on engagement (strength of evidence)	Effects on outcome (strength of evidence)	Relevance for clinicians or iCBT designers
Adaptive and interactive treatment features	+ (Strong evidence) Direct & Indirect	0 (Strong evidence) Direct & Indirect	Choose programs that use chatbots, text messages, multiple channels, and gamification to improve engagement, but recognize that personalization of messages, content, message delivery, or personalized interfaces have not been consistently shown to improve clinical outcomes.
Patient co-creation of intervention	+ (Strong evidence) Direct & Indirect	0 (Strong evidence) Direct & Indirect	Choose interventions developed with input from people with lived experience of mental health disorders.
Culturally appropriate interventions	+ & 0 (Mixed evidence) Direct & Indirect	+ & 0 (Mixed evidence) Direct & Indirect	Choose interventions that are culturally appropriate, and acceptable to the target group
Brief human support	+ (Strong evidence) Direct & Indirect	+ & 0 (Mixed evidence) Direct & Indirect	Use brief human support (<10 min) for patients with higher levels of depression.
Offer core iCBT components	+ (Strong evidence) Direct & Indirect	+ (Strong evidence) Direct & Indirect	Include behavioral activation as a core component; if only one component is included, prioritize behavioral activation.
Increase acceptability by providing a treatment rationale	+ & 0 (Mixed evidence) Indirect evidence	+ & 0 (Mixed evidence) Indirect evidence	Explain the reason for the approach taken, and how it aligns with the current evidence-base.

+ = positive relationship; − = negative relationship; 0 = no impact in either direction; + & − = both positive and negative relationships reported; + & 0 = both positive and no impact in either direction reported. Strong evidence = substantial literature, typically including multiple randomized trials and/or meta-analyses, with findings aligned in the same direction; Weak evidence = few research trials; Mixed evidence = mix of positive and negative trial results; No evidence = no trials or no significant correlations; Direct evidence = research on iCBT specifically; Indirect evidence = research on DMHIs or CBT more generally.

Engagement

Across iCBT and related DMHI research, personalization strategies are more consistently associated with engagement than with clinical outcomes. Engagement is facilitated by features that enhance perceived fit, usability, and accountability, including adaptive and interactive design elements, brief human support, co-creation, cultural relevance, and acceptability-enhancing strategies, such as providing a clear treatment rationale.

Within iCBT, flexible module sequencing and module choice are associated with higher engagement²² and behavioral activation appears particularly engaging relative to other CBT components²³. Although direct iCBT evidence is still limited, CBT-based and digital mental health studies suggest that automated engagement supports (e.g., chatbots, gamified elements, progress-based prompts, text reminders, and automated mood monitoring) may increase sustained use and perceived personal fit^12,24,25. This pattern is echoed in DMHI reviews²⁶. These system-delivered interactions can create a sense of responsiveness or support but typically do not involve individualized clinical input.

A related but distinct engagement strategy is brief human support. In iCBT, even minimal guidance from a person (a clinician, coach, or trained supporter) appears to be one of the most robust facilitators of engagement and reduced attrition. Even minimal, standardized guidance improves engagement and reduces attrition, regardless of provider expertise^27–30. Automated reminders are widely used and improve engagement in, for example, transdiagnostic iCBT contexts and iCBT for insomnia, though iCBT for depression-specific evidence remains sparse^31–33. Thus, automated supports may improve usability and continuity, whereas human support may add accountability, relational connection, and clinical reassurance.

Contextual factors such as social support, co-creation, and cultural relevance further enhance acceptability and perceived fit, but do not reliably prevent dropout^26,34,35.

Clinical outcomes

Meta-analytic and randomized evidence indicates that greater choice, tailoring, or interactivity does not reliably improve symptom outcomes once evidence-based iCBT content is delivered.

Symptom-tailored (transdiagnostic) and disorder-specific iCBT protocols demonstrate comparable efficacy³⁶, suggesting that adapting content to users’ symptom profiles does not necessarily improve outcomes relative to standardized protocols. Increasing guidance intensity or personalization also does not consistently enhance outcomes in iCBT or DMHIs³⁷. Similarly, module completion is not a reliable predictor of symptom change, and shorter structured iCBT programs can achieve outcomes comparable to standard-length interventions³⁸, mirroring broader DMHI findings^26,39.

Cultural adaptation and co-creation show promising but mixed outcome effects. Adapted iCBT often outperforms waitlist controls and sometimes unadapted versions^40,41. For depression-specific iCBT, published randomized controlled trials of culturally adapted programs appear to have compared the adapted intervention with waitlist, treatment-as-usual, or active/control conditions, rather than directly comparing culturally adapted iCBT with the original unadapted iCBT. Reviews similarly note that direct comparisons of culturally adapted versus non- or less-adapted DMHIs are lacking⁴².

Human support shows the clearest outcome-related signal, but effects are modest and moderated by severity. Guided iCBT yields small incremental benefits on average^43,44, with greater benefit for patients with higher baseline severity³⁸ and little advantage for mild depression⁴⁵. Behavioral activation emerges as a key driver of efficacy^43,45, though dismantling trials confirm that multiple CBT components contribute meaningfully to outcomes²³.

Personalizing treatment through stratification

Personalization through stratification aims to match patients to treatments based on characteristics that predict differential response, implemented at treatment entry. In other words, personalization of stratification focuses on patient-treatment matching that considers the unique patient’s characteristics early on in treatment. Most evidence supporting stratified care comes from psychotherapy more broadly rather than iCBT specifically (e.g., ref. ¹⁵). Patients may be stratified using demographic, clinical, psychological, or biological markers associated with treatment response⁴⁶, and growing evidence suggests such approaches can improve outcomes in psychotherapy^16,47,48.

Most psychotherapy research relies on post-hoc moderator or mediator analyses to examine differential treatment response and to determine mechanisms and predictors of response (for a systematic review on mechanisms of change in DMHI, see ref. ⁴⁹, while conceptual mechanism-agnostic models propose data-driven and biologically informed models as potential future tools for stratification⁵⁰.

Table 2 provides an overview of the stratification variables in iCBT. The detailed narrative review of the evidence for Table 2 is available in Supplementary Note 3.

Table 2.

Personalization of stratification variables

Personalization variables	Effect on engagement (strength of evidence) Nature of evidence	Effects on outcome (strength of evidence) Nature of evidence	Relevance for clinicians and iCBT service providers
Demographics: Age	+ & − (Mixed evidence) Direct & Indirect	+ & 0 (Mixed evidence) Indirect evidence	iCBT can be offered to those of all ages.
Demographics: Gender	+ (Weak evidence) Direct and Indirect	0 (Strong evidence) Direct and Indirect	Women may be more likely to take up a digital intervention, but men and women have similar treatment outcomes.
Other Demographics: Location, relationship status, educational level, income.	0 (Strong evidence) Direct and Indirect	0 (Strong evidence) Direct and Indirect	iCBT can be offered to patients of all different demographics
Clinical factors: Severity of depression symptoms	− (Weak evidence) Indirect	+ (Strong evidence) Direct & Indirect	Patients with severe symptoms may engage less but if they engage, iCBT is likely to be effective for these patients
Clinical factors: Previous treatment	0 (Weak evidence) Indirect evidence	− (Strong evidence) Direct & Indirect	iCBT is most effective for those who have not had treatment before. If patients have had other treatments, iCBT is less likely to be effective
Clinical factors: Psychiatric comorbidity, psychotropic medication, medical comorbidity	− (Weak evidence) Indirect evidence	0 (Weak evidence) Direct & Indirect	There is no need to assess other clinical factors to determine eligibility of iCBT.
Psychological factors: Personality and temperament factors	+ & − (Weak evidence) Indirect evidence	(No evidence)	These factors may not influence engagement in iCBT, but more evidence is needed
Psychological factors: Attachment	+ (Weak evidence) Indirect evidence	(No evidence)	No need to assess for attachment security when considering iCBT
Psychological factors: Adverse life events and cognitive bias	(No evidence)	(No evidence)	No need to assess for previous adverse life events when considering iCBT
Psychological factors: Social support	+ (Weak evidence) Direct & Indirect	+ (Weak evidence) Direct evidence	Patients with social support may be more likely to engage with and benefit from iCBT
Biological factors (Neurological and genetic)	+ (Weak evidence) Indirect evidence	+ (Weak evidence) Direct & Indirect	Patients with high inhibitory control may be more likely to benefit from iCBT but more evidence is needed.
Digital biomarkers	(No evidence)	(No evidence)	These factors require more investigation
Multimodal digital phenotypes	(No evidence)	(No evidence)	These factors require more investigation

+ = positive relationship; − = negative relationship; 0 = no impact in either direction; + & − = both positive and negative relationships reported; + & 0 = both positive and no impact in either direction reported. Strong evidence = substantial literature, typically including multiple randomized trials and/or meta-analyses, with findings aligned in the same direction; Weak evidence = few research trials; Mixed evidence = mix of positive and negative trial results; No evidence = no trials or no significant correlations; Direct evidence = research on iCBT specifically; Indirect evidence = research on DMHIs or CBT more generally.

Engagement

Evidence linking patient characteristics to engagement with iCBT is limited and largely indirect. No demographic variable (including age, gender, education, employment status, or relationship status) shows a consistent association with engagement in depression-focused iCBT. Most evidence derives from the broader DMHI literature, suggesting that younger adults are more likely to initiate digital interventions, whereas middle-aged and older adults who enrol tend to persist longer or engage with more content^26,51,52. Similarly, an individual participant data meta-analysis of self-guided DMHIs for depression (many CBT-based but not exclusively iCBT) found lower dropout risk with increasing age⁵³. However, few iCBT-for-depression trials have examined engagement using comparable metrics, limiting firm conclusions.

Clinical predictors of engagement are also poorly understood. Baseline symptom severity has not been systematically examined in iCBT, and indirect DMHI evidence suggests a dual role: greater symptom burden may increase interest in digital interventions, while core depressive symptoms such as fatigue and anhedonia may hinder sustained engagement²⁶. Prior mental health treatment history has not been shown to predict engagement in iCBT.

Psychological and contextual factors show stronger but still limited links to engagement. Higher perceived social support is associated with greater engagement in adolescent iCBT³⁵, and DMHI research suggests that personality traits such as higher neuroticism and lower extraversion may influence engagement tendencies²⁶. Evidence for attachment style, cognitive functioning, or biological and digital markers remains indirect or absent⁵⁴.

Clinical outcomes

Several patient characteristics show more consistent, but modest, associations with clinical outcomes in iCBT. Demographic variables, including age, gender, education, employment status, and relationship status, do not meaningfully moderate treatment efficacy in large individual participant data meta-analyses of iCBT for depression^9,38,43,45. Although women are overrepresented in iCBT trials, outcomes appear equivalent for men and women.

Baseline symptom severity is the most extensively studied predictor of outcome. Individual participant data meta-analyses indicate that patients with higher baseline severity show larger absolute symptom reductions following iCBT⁹, although this may partly reflect greater room for improvement⁵⁵. More specific iCBT studies suggest that higher baseline cognitive symptom burden predicts greater gains⁵⁶, and that tailored iCBT may be particularly beneficial for patients with higher initial severity²⁷.

Prior psychological treatment history reliably moderates outcomes in iCBT, with previously treated patients showing smaller symptom improvements than treatment-naïve individuals^43,56, consistent with broader CBT evidence⁵⁷. In contrast, psychiatric comorbidity, psychotropic medication use, and medical comorbidity do not appear to significantly moderate iCBT outcomes^1,58,59.

Evidence for psychological, biological, and digital predictors of outcome remains sparse. Social support has been associated with better iCBT outcomes in adolescents³⁵, but findings are correlational and sample-specific. Biological predictors show early promise, one study linked inhibitory control in emotion-regulation networks to greater symptom reduction following iCBT⁶⁰, yet most evidence remains indirect. Recent multimodal machine-learning approaches integrate clinical, behavioral, and biological data to improve prognostic accuracy^8,61 but incremental gains are small and have not translated into improved outcomes via prescriptive treatment matching. Although multimodal digital phenotyping has shown promise for characterizing depressive symptoms and predicting future symptom trajectories, there is currently no direct evidence that such data predict engagement with iCBT or improve iCBT treatment outcomes, nor that using these signals to personalize treatment confers clinical benefit.

Responsiveness during treatment

Responsiveness, - adapting treatment during delivery based on patient progress -, is conceptually well aligned with iCBT, given the routine collection of symptom and usage data. Responsiveness may occur at two levels: within the iCBT intervention itself, such as adapting modules, sequencing, timing, or feedback; and at the care-pathway level, such as increasing guidance, adding check-ins, or stepping patients to more intensive care. Table 3 provides a summary of the dynamic factors that offer the potential for leverage during treatment. A more detailed table describing how adaption can occur during treatment, and how this is managed manually, automatically or via artificial intelligence as well as the narrative review of the evidence for these dynamic factors during treatment is provided in Supplementary Table 1 and Supplementary Note 4.

Table 3.

Dynamic factors that could be used during treatment

Personalization variables	Effect on engagement (strength of evidence)	Effects on outcome (strength of evidence)	Relevance for clinicians
Adjustments of treatment intensity	(No evidence)	(No evidence)	Stepped care interventions that are driven by data protocols may offer clinical insights
Progress monitoring and adjustments of content	+ (Weak evidence) Indirect evidence	+ (Strong evidence) Indirect evidence	Data driven reminders can be sent directly to the patient to enact change Deviation from predicted trajectory can be used to initiate change
Capturing real-time data during treatment to inform decisions	(No evidence)	(No evidence)	Provides insight into day-to-day activities of the patient that may be used to make decisions and gain insights
Adjusting the timing of interventions	(No evidence)	(No evidence)	Opportunity to offer just in time interventions

+ = positive relationship; − = negative relationship; 0 = no impact in either direction; + & − = both positive and negative relationships reported; + & 0 = both positive and no impact in either direction reported. Strong evidence = substantial literature, typically including multiple randomized trials and/or meta-analyses, with findings aligned in the same direction; Weak evidence = few research trials; Mixed evidence = mix of positive and negative trial results; No evidence = no trials or no significant correlations; Direct evidence = research on iCBT specifically; Indirect evidence = research on DMHIs or CBT more generally.

Engagement

Evidence that responsiveness strategies improve engagement in iCBT is limited and largely indirect. While stepped-care models are widely implemented at the service level, no studies have demonstrated that dynamically adjusting treatment intensity within iCBT based on patient response improves adherence or sustained use⁶². Routine outcome monitoring (ROM) and early symptom tracking are ubiquitous in iCBT trials, and early response reliably predicts subsequent outcomes^38,63, but few studies have tested whether feeding this information back to patients or systems enhances engagement. Indirect evidence from DMHIs and blended-care contexts suggests that progress tracking, feedback, and symptom monitoring tools may support engagement^26,64. However, just-in-time interventions and real-time data–driven adaptations have not been implemented in iCBT for depression⁶⁵ and digital continuous monitoring (e.g., passive sensing) has not yet been shown to increase engagement when embedded in iCBT protocols⁶⁶.

Clinical outcomes

With respect to clinical outcomes, responsiveness strategies show stronger evidence in psychotherapy more broadly than in iCBT specifically. ROM and feedback systems robustly improve outcomes in face-to-face CBT and blended care by identifying patients who are “not on track” and guiding treatment adjustments⁶⁷. In iCBT, repeated symptom measurement and early response are reliable prognostic indicators^38,45, but few studies have tested whether using these data to dynamically adapt iCBT content or intensity improves outcomes^68–70. Evidence for stepped care within iCBT remains indirect, and no trials have demonstrated outcome benefits from within-program stepping based on response⁶². Similarly, although digital continuous monitoring, passive sensing, and adaptive timing are increasingly feasible, evidence remains limited that acting on these data to personalize iCBT delivery improves depression outcomes. Chatbots and other AI-enabled systems may personalize prompts, feedback, or user experience, but few controlled trials have isolated whether such real-time personalization improves depression outcomes beyond comparable non-personalized iCBT. Overall, responsiveness in iCBT is currently supported primarily as a prognostic and monitoring tool, rather than as an empirically validated mechanism for improving outcomes through dynamic adaptation.

Discussion

This narrative review synthesized evidence on personalization strategies to enhance iCBT outcomes and thus provides useful evidence to underpin an actionable framework for delivering iCBT through optimization, stratification and responsiveness strategies. The review is likely to be of benefit to clinicians, who integrate iCBT into blended care, refer patients to iCBT as a stand-alone or suggest iCBT as an interim intervention while patients are on a waitlist, and to iCBT developers and iCBT service providers, who aim to improve their services.

Optimization: strong evidence for engagement, selective evidence for outcomes

The strongest and most consistent evidence supports optimization strategies implemented prior to treatment initiation. Across iCBT trials and individual participant data meta-analyses, brief human support, inclusion of core CBT components (particularly behavioral activation), and clear treatment rationales reliably improve engagement. However, optimization strategies that increase engagement, such as increased choice, interactivity, personalization of delivery features, or cultural adaptation, do not consistently translate into superior symptom outcomes. This dissociation underscores a critical point: Engagement-enhancing features are necessary but not sufficient for improving efficacy and should be viewed as facilitators of treatment exposure rather than mechanisms of change.

Effective optimization of iCBT depends not only on patient factors but also on clinician involvement in iCBT. Therapist involvement reliably improves engagement and outcomes in guided and blended formats^38,45. Although iCBT is often conceptualized as a standardized, technology-driven treatment, its implementation and impact are shaped by external factors such as clinician support, service delivery models, and integration with routine care. Clinicians influence engagement, adherence, and outcomes through guidance, monitoring, encouragement, and clinical decision-making (e.g., stepping care up or down, addressing risk, or coordinating additional treatments). As a result, iCBT outcomes are not determined solely by the intervention itself, but by how it is embedded within a system of care. Recognizing this interaction is critical for understanding variability in outcomes and for designing effective personalization strategies that extend beyond the treatment content to include delivery, support, and clinical integration.

In blended care, iCBT can be used strategically to deliver structured psychoeducation, behavioral activation, and skills practice, allowing face-to-face sessions to focus on formulation, motivational work, and application of skills. When iCBT is offered during a waitlist period, it should be treated as an active intervention rather than a passive placeholder, with explicit engagement goals and agreed-upon monitoring plans.

Stratification: limited prescriptive value beyond severity and history

Evidence for stratification of iCBT based on patient characteristics remains modest. Among the many demographic, clinical, psychological, and biological variables examined, baseline symptom severity and prior treatment history emerge as the most clinically relevant predictors of outcome. Patients with higher severity benefit more from guided or blended iCBT, while those with milder symptoms often do equally well in self-guided formats. By contrast, most demographic variables, including age, gender, education, and employment, do not meaningfully moderate iCBT outcomes, and evidence linking them to engagement is weak or indirect. Emerging multimodal and machine-learning approaches show promise for prognostic modeling, but incremental gains are modest and have not yet translated into actionable clinical decision rules. At present, stratification in iCBT should therefore be conceptualized as risk-informed triage, not precision matching.

Responsiveness: strong rationale, limited iCBT-specific evidence

Responsiveness, adapting treatment during delivery based on progress, is conceptually well aligned with iCBT, given its capacity for repeated symptom and usage monitoring. Within depression-focused iCBT, early symptom changes and early indicators of nonresponse are consistently prognostically meaningful, supporting a “watch early, act early” approach. However, experimental evidence that dynamically adapting iCBT content, sequencing, timing, or support improves engagement or outcomes in depression remains limited. At the same time, adjacent areas such as iCBT for insomnia provide emerging proof-of-concept that early identification of likely nonresponders followed by adapted intervention can improve outcomes, suggesting the potential value of similar approaches in depression iCBT⁷¹. The most evidence-supported responsive strategy in current iCBT practice is adjustment of treatment intensity, such as increasing guidance, adding check-ins, or stepping patients up to more intensive care when early non-response or disengagement is detected. Notably, these strategies primarily operate at the level of care delivery or treatment allocation, rather than through direct personalization of iCBT content or structure. When patients fail to respond early, prioritizing simpler, action-oriented components may be more effective. More granular adaptations, such as automated content changes, just-in-time interventions, or adaptations driven by passive sensing, remain promising but unvalidated. Much of the enthusiasm for these approaches is extrapolated from face-to-face CBT or DMHIs more broadly, highlighting a major gap between conceptual models and iCBT-specific evidence.

Future research directions

This review highlights substantial progress in personalization research alongside critical gaps, particularly for iCBT for depression. Addressing these gaps will require clearer distinctions between evidence derived from iCBT, DMHIs or CBT more broadly, as well as greater methodological rigor.

A primary priority is strengthening the evidence base for stratification within iCBT. Although numerous clinical, psychological, biological, genetic, and digital variables have been identified as predictors of outcome using moderator analyses, routine-care datasets, and machine-learning approaches, most findings remain predictive rather than prescriptive. Few studies have validated stratification variables in independent samples, examined differential effects on treatment components, or experimentally tested whether stratified treatment allocation improves outcomes. Notably, no studies have demonstrated superiority of stratified iCBT over non-stratified iCBT, in contrast to broader psychotherapy trials showing benefits of stratified care (e.g. ref. ⁷²). Future work should therefore prioritize experimental designs to directly test whether stratification improves engagement or outcomes in iCBT.

A second key direction concerns data selection for personalization models. While models can integrate diverse data sources, evidence from psychotherapy and DMHIs suggests that costly biomarkers often add little predictive value beyond well-established self-report measures, except in specific contexts^73,74. For iCBT, future research should prioritize high-resolution yet feasible data (e.g., repeated symptom measures, usage metrics, and electronic health records), which already show promise for predicting clinically meaningful outcomes (e.g. ref. ⁷⁵). Evaluating added value will require decision-curve analyses, implementation studies, and health-economic evaluations, rather than reliance on predictive accuracy alone.

Further work is also needed on treatment matching, including when iCBT is the optimal first-line intervention versus when alternative or blended approaches are preferable. Personalized Advantage Index approaches have demonstrated outcome benefits across psychotherapy and pharmacotherapy but have rarely been applied to iCBT and almost never prospectively. Similarly, algorithmic personalization of CBT modules has shown promise in face-to-face and transdiagnostic contexts but remains largely untested in iCBT for depression.

Finally, advancing personalization in iCBT may require redefining treatment targets. Emerging frameworks emphasize individual-specific baselines and recovery endpoints, rather than population-level thresholds, to capture meaningful change within individuals⁷⁶.

Conclusion

Current evidence supports optimizing engagement before treatment, stratifying initial support based on severity and treatment history, and responding early to non-response by adjusting treatment intensity. Engagement and clinical outcomes are related but distinct, underscoring that strategies improving one do not reliably improve the other, while emerging research on responsiveness and digital biomarkers offers future opportunities for more precise personalization.

Supplementary information

Jan09.Submitted.Supplement. A.B.C.Dec28 ^{(286KB, pdf)}

Author contributions

K.A.V.D.: conceptualization, literature review, manuscript writing, and editing. H.C.: conceptualization, literature review, manuscript writing, and editing.

Data availability

No datasets were generated or analyzed during the current study.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41746-026-02844-7.