ABSTRACT

Introduction

Major depressive disorder (MDD) and bipolar disorder (BD) are often associated with persistent cognitive deficits that impair psychosocial functioning. While pro‐cognitive interventions show promise, trial findings are inconsistent, potentially due to baseline factors influencing treatment response. This systematic review summarizes evidence on pre‐treatment characteristics associated with cognitive improvement and offers methodological recommendations.

Methods

A systematic search was conducted in PubMed/MEDLINE, EMBASE, PsycINFO, and Cochrane Library from inception to February 28, 2025. Eligible studies included primary or secondary analyses of randomized controlled trials (RCTs) investigating predictors of cognitive response to pro‐cognitive interventions in MDD and/or BD. Two researchers independently conducted study selection and risk of bias assessments. Findings were synthesized qualitatively.

Results

Forty studies (N = 3864) were identified, covering pharmacological treatments (k = 20; N = 2299), psychological therapies (k = 16; N = 1165), brain stimulation (k = 2; N = 168), and physical activity (k = 2; N = 232). Poorer baseline cognitive performance was the most consistent predictor of greater cognitive improvement, though the direction of the effect was not entirely uniform across all studies. Baseline depression severity showed no significant association with cognitive outcomes. Age, education, sex, IQ, diagnosis, and medication status were similarly non‐predictive. Risk of bias was high in 77% of studies, mainly due to deviations from specified outcomes, poor randomization processes, and inconsistent handling of missing data. Considerable heterogeneity in interventions, outcome measures, and sample characteristics limited replicability and precluded meta‐analysis.

Conclusion

Poorer baseline cognition emerged as the most reliable predictor of greater cognitive improvement across interventions. More rigorous, well‐powered studies are needed to replicate these findings and identify robust predictors to guide personalized pro‐cognitive treatment approaches in mood disorders.

Summary Significant outcomes○Poorer baseline cognitive performance was the strongest predictor of post‐treatment cognitive improvement, although the direction of the effect was not uniform.○Baseline depressive symptom severity was not significantly associated with post‐treatment cognitive change across studies.○In contrast, age, education, sex, IQ, diagnosis, and medication status showed no consistent association with treatment‐related cognitive outcomes. Limitations○Risk of bias evaluation indicated some or high risk of bias for the majority (77%) of the studies.○Key concerns were outcome deviations, poor randomization, and inconsistent handling of missing data.○Substantial heterogeneity across studies contributed to limited replicability and did not allow for a meta‐analytic approach.

1 Introduction

A significant proportion of individuals with major depressive disorder (MDD) and bipolar disorder (BD) display cognitive impairment across various domains, with moderate severity indicated by group means [1, 2, 3, 4] and more pronounced impairments in BD [5, 6, 7]. Recent clustering studies reveal considerable heterogeneity within both disorders, identifying at least three cognitive subgroups ranging from intact to severely impaired profiles [8, 9].

Cognitive deficits often persist beyond acute mood episodes [10, 11], impeding functional recovery and occupational functioning [12, 13, 14, 15, 16] and exacerbating the socio‐economic burden of mood disorders [17]. Deficits are also linked to worse illness trajectories, including increased risk of recurrence [18, 19], hospitalization [20], poorer treatment outcomes [21, 22], and lower quality of life [23, 24].

Recognizing cognition as a core feature has driven the search for pro‐cognitive treatments, leading to numerous clinical trials. The ISBD Targeting Cognition Task Force has reviewed randomized controlled trials (RCTs) assessing pharmacological, psychological, and neuromodulation interventions across 57 RCTs [25, 26, 27]. Cognitive remediation (CR) therapies showed the most promising, though preliminary, effects. Among pharmacological agents, modafinil and erythropoietin (EPO) benefited both MDD and BD [25, 26, 27]; mifepristone and lurasidone improved cognition in BD, while vortioxetine showed benefits in MDD. For neuromodulation, repetitive transcranial magnetic stimulation (rTMS) showed no cognitive effects, while transcranial direct current stimulation (tDCS) indicated some improvements in MDD [26]. Despite some promising findings, evidence remains preliminary, often limited by high or unclear risk of bias [25, 26, 27], as also noted by other reviews [28, 29].

Broad methodological challenges limit the generalizability of findings. These issues, along with recommendations for improving future trials, have been extensively reviewed by the ISBD Targeting Cognition Task Force [27, 30]. A key barrier to progress is the limited understanding of pre‐treatment characteristics influencing response. Although factors such as age, symptom severity, and baseline cognition have been examined, findings are scarce and inconsistent. Identifying predictors of treatment response (and non‐response) could reduce outcome variability, optimize intervention delivery, and guide adaptations for specific patient subgroups.

This systematic review by the ISBD Targeting Cognition Task Force summarizes, for the first time, evidence from cognition trials examining differential treatment response according to pre‐treatment factors in mood disorders. Our aims are to: (a) identify baseline characteristics predicting response, (b) review methodologies used to study predictors, and (c) provide methodological guidance for future investigations into differential treatment response.

2 Materials and Methods

This systematic review was conducted and reported in line with the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses guidelines (PRISMA 2020 statement; [31]). A protocol was registered on PROSPERO (CRD420250648711) prior to the initiation of the review.

2.1 Eligibility Criteria

Only studies investigating the relationship between one or more pre‐treatment/baseline factors and the response to interventions targeting cognition, as examined by one or more primary or secondary/exploratory objectively measured cognitive outcomes (i.e., measured with performance‐based neuropsychological tests), were considered for inclusion. Eligible studies: (a) were peer‐reviewed articles in a scientific journal, published in English, (b) reported a primary or secondary/post hoc analysis of an RCT, and (c) had a sample primarily consisting of adult participants with a mood disorder (MDD/BD) diagnosis (i.e., ≥ 50%) according to the Diagnostic and Statistical Manual of Mental Disorders [32] or the International Classification of Diseases [33]. Where necessary, authors were contacted to request either additional data or data specific to the mood disorder subsample. We excluded: (a) non‐RCT study designs (e.g., case series), (b) conference/meeting abstracts, reviews, and meta‐analyses, and (c) studies without objective cognitive outcomes or not examining pre‐treatment/baseline predictors of response. We also did not consider trials examining baseline factors only as covariates/confounders in their efficacy analysis or assessing post‐randomization factors (e.g., therapy duration) and early treatment changes as predictors of response.

2.2 Search Strategy

A comprehensive systematic search was conducted on PubMed/MEDLINE, EMBASE, PsycINFO (using the Ovid interface), and Cochrane Library databases from inception to the 28th of February 2025. In line with the PICO framework (Population, Intervention, Comparison, Outcome), our search strategy included the following main components: mood disorder, treatment, RCT, cognition, predictor. Multiple keywords for each component were combined using ‘AND’ and ‘OR’ clauses. The search syntax is presented in detail in Supporting Information. To identify any additional eligible papers, the reference lists of included studies, previous reviews, and relevant publications in the field were hand‐searched.

2.3 Study Selection

The study selection process was performed independently by two authors (DT and MK) blinded to each other's selections, using the Rayyan open‐source review management software [34]. This first included screening titles and abstracts of all articles, then screening the full‐text papers of those retrieved as potentially eligible. Percent agreement between the two authors was 85% for the initial screening (corresponding to 0.7 Cohen's kappa) and > 90% for the full‐text screening (corresponding to > 0.82 Cohen's kappa). Disagreements were discussed until reaching consensus and, when needed, were resolved by the senior author (KWM).

2.4 Data Extraction

Data were extracted by DT and cross‐checked by MK. The following information was extracted from included studies: authors and year of publication, country, study design, sample size details, baseline mood state, basic demographic and clinical characteristics, treatment and control condition main features, cognitive outcomes and corresponding assessment time points, predictors assessed, type of analysis, and main findings for the effect of the predictor on the outcomes. For studies reporting multiple time points, we adopted the time point specified as primary or otherwise the latest one.

Data synthesis was organized according to the type of predictors examined. A quantitative meta‐analysis was not undertaken due to substantial heterogeneity in intervention modalities and treatment outcomes between studies.

2.5 Risk of Bias Assessment

Risk of bias (RoB) within and across the included studies was assessed independently by two authors (DT and JMS) using the Cochrane Collaboration's Tool for Assessing Risk of Bias 2.0 (RoB 2 tool; [35]). The RoB assessments focused on the analyses of cognitive outcomes in the included studies. Where necessary, we conducted additional searches in trial registries and published trial protocols to extract relevant details on study design and outcomes of interest. If these details were inadequately reported in primary or secondary publications, this was reflected in the RoB assessment. Scoring discrepancies were discussed until a consensus was reached and, if needed, were consulted with the senior author (KWM).

3 Results

Our systematic literature search, combined with additional hand searches, identified 2504 articles after removing duplicates. After initial screening, 224 were retrieved for full‐text evaluation. Of these, 184 reports were excluded with reasons provided, and 40 were included in the qualitative synthesis. A detailed PRISMA flowchart is presented in Figure 1. In total, studies reported on 3864 participants from 35 original RCTs assessing pharmacological or behavioral interventions targeting cognition in patients with mood disorders. A summary of the main study characteristics is displayed in Table 1.

3.1 Study Characteristics

Most studies were conducted in Europe (k = 20), 12 in the USA, two in Australia, one in Canada, one in New Zealand, and four across multiple countries. Most of the included articles reported a primary (k = 14) or secondary analysis of an RCT (k = 17) with cognition as the main outcome, while fewer were either primary (k = 5) or secondary (k = 4) RCT analyses with cognition as an exploratory outcome. The mean sample size was 97 participants (SD = 110; range: 18–598), with a treatment group mean of 53 (SD = 73; range: 11–402) and a control group mean of 44 (SD = 41; range: 0–196). On average, pharmacological studies had a larger sample size (N = 115) than non‐pharmacological ones (N = 78).

3.2 Sample Characteristics

Participants had a mean age of 44.8 years (SD = 9.5; range: 31.4–72.3; k = 38), were mostly women (mean = 58.6%; SD = 10.4; range: 36.4%–82%; k = 38), with 14.4 years of education (SD = 3.1; range: 11.8–16.4; k = 26), and an average illness duration of 15.4 years (SD = 10.7; range: 10.1–24.4; k = 16). Across studies, the percentage of participants with a mood disorder diagnosis was 96% (SD = 13; range: 50–100%). Sixteen studies included participants with MDD, 14 included participants with BD, and 10 included mixed samples with both diagnoses. Mood state was balanced between studies recruiting remitted and/or partially remitted participants (k = 23) and those recruiting partially and/or not remitted participants (k = 17), with non‐pharmacological studies being over‐represented in the former and pharmacological studies more prevalent in the latter category.

3.3 Treatment Characteristics

Twenty studies investigated predictors of response to pharmacological treatments (N = 2299), 16 studies investigated predictors of response to psychological therapies (N = 1165), two studies investigated predictors of response to brain stimulation techniques (N = 168), and two investigated physical activity interventions (N = 232).

3.4 Control Characteristics

Of the 20 pharmacological studies, 18 used an inactive substance (e.g., placebo, saline) as the comparator, and two tested active compounds in both arms. For non‐pharmacological studies, 12 had an active control condition (e.g., computer games, psychoeducation), six had a non‐active control condition (e.g., standard treatment, treatment as usual), and two involved no control group participants in their secondary analysis.

3.5 Risk of Bias Assessment

Figure 2 presents the RoB evaluations. Low RoB was detected in 22.5% of the studies, while some concerns and high RoB were found in 52.5% and 25% of the studies, respectively. The primary methodological concerns were lack of adherence to pre‐specified outcomes and analyses (15% high risk; 35% some concerns), limited details on or lack of randomization (7.5% high risk; 37.5% some concerns), and suboptimal handling of missing outcome data (5% high risk; 40% some concerns). In general, fewer pharmacological studies were of high RoB (15%) compared to non‐pharmacological studies (35%). Overall, studies often had methodological challenges regarding statistical analyses, which were most pronounced in non‐pharmacological studies.

3.6 Predictors of Treatment Response

Twenty‐three studies (58%) examined one variable as a predictor of response to a cognitive intervention, 10 studies (25%) assessed two predictors, and seven studies (17%) evaluated multiple predictors. Following a previously published approach [76], predictors of response were grouped into five categories: demographic, cognitive/functional, biological, clinical/illness‐related, and psychological. To facilitate results aggregation and interpretation, closely related predictors were grouped together under the same type. Key findings for each predictor type are summarized per category in Table 2.

3.6.1 Demographic Factors

Six demographic predictors were examined across eight studies (N = 534): age, education, sex, premorbid IQ, relationship status, and housing situation. Of the 17 instances when a demographic response predictor was tested, an association with cognitive improvement was only reported twice (12%), once for age and once for education, without any replication in different samples (Table 2).

Specifically, five of six studies found no association between age and cognitive improvement [60, 63, 69, 73, 75], while one reported that younger MDD and BD partially remitted outpatients benefited more than older ones in category fluency following compensatory cognitive training [68].

Three of the four studies found no influence of education [50, 60, 69, 74], while one study showed that higher education predicted greater verbal memory improvement after both bupropion XL and escitalopram treatment in depressed individuals with MDD [50].

No significant associations were found for sex [60, 69, 73], premorbid IQ [60, 69], relationship status, and housing situation [60].

3.6.2 Cognitive and Functional Factors

Collectively, five types of cognitive/functional predictors were considered, including baseline objectively measured cognition, subjectively reported cognitive complaints, psychosocial functioning, employment status, and employment type. In 20 analyses across 12 studies (N = 1425), a significant association was found ten times (50%).

Objectively measured cognition at baseline was the most consistent predictor of post‐treatment cognitive changes (six of 10 studies), although the direction of this association was not uniform: five studies (83%) found that poorer baseline cognition predicted better outcomes [45, 48, 56, 59, 69], while one study found that better executive function predicted greater improvement in this domain [73] (Table 2). Specifically, two secondary RCT analyses reported that cognitively impaired MDD (depressed) and BD (partially remitted) outpatients were more likely to achieve a clinically relevant improvement in memory [45] and global cognition following treatment with EPO [48]. Importantly, the effect of baseline cognition was specific to participants receiving active treatment, indicating it was not merely due to regression to the mean. A similar association—where lower baseline cognition predicted greater treatment‐related improvement—was observed in two RCTs of cognitive remediation (CR) paradigms in remitted BD [69] and partially remitted patients with MDD or BD [59]. In a secondary analysis of an RCT of action‐based CR, a non‐significant trend (p = 0.08) also suggested that poorer baseline performance predicted greater executive function improvement in the active versus control groups [61]. A comparable effect was reported in a secondary analysis of a functional remediation (FR) trial [56]: while the original RCT found no significant cognitive effects, this analysis—focused on cognitively impaired participants—revealed greater post‐treatment verbal memory improvement following FR compared to treatment as usual.

In contrast to five studies linking poorer baseline cognition to greater intervention benefits, one rTMS study in depressed MDD outpatients found that better baseline executive function predicted greater improvement across groups [73]. This small study (N = 48) had the potential for selective outcome reporting (see Figure 2). Finally, one study in remitted MDD outpatients assessed the correlation between baseline and post‐intervention performance across the full sample, including both modafinil and placebo groups [40]. While a non‐specific positive association was observed—indicating that individuals with lower baseline scores also tended to score lower post‐intervention—the authors did not examine the relationship between baseline performance and cognitive change scores, nor whether this relationship was moderated by modafinil. As such, it remains unclear whether baseline performance was associated with cognitive change in response to the intervention.

Of the five studies investigating subjective cognitive complaints, only one found that greater self‐reported difficulties were weakly related to memory improvement in the active versus control groups [45], while four reported no significant effects [60, 61, 69].

Regarding psychosocial functioning, two studies found that this was not predictive of post‐treatment cognitive changes [42, 60]. Employment status and type were examined as response predictors in a large RCT comparing vortioxetine to placebo in depressed MDD participants [43]. Further, being employed and having a ‘professional’ based employment type predicted greater improvement in multiple cognitive outcomes compared to the entire sample and working subgroup, respectively. However, the effect of employment status was not replicated in a smaller CR study [60], rendering the evidence inconsistent.

3.6.3 Neuroimaging and Biological Factors

Eight types of neuroimaging and biological markers were considered by 14 studies (N = 991). A significant association with cognitive changes following treatment was detected in 12 of 25 analyses (48%), primarily for structural neuroimaging markers (Table 2).

Cortical volume or thickness was identified as predictors of response by two studies: larger left precentral gyrus surface area moderated the effect of EPO versus placebo on verbal memory in a mixed sample of symptomatic and remitted mood disorder outpatients [44], while lower dPFC thickness and smaller superior temporal volume were associated with greater treatment‐specific executive function improvement in fully or partially remitted BD participants receiving action‐based CR [62]. Total hippocampal volume did not affect outcomes in two studies [39, 44], whereas more right‐ward hippocampal volume asymmetry (a marker of brain reserve) predicted greater verbal memory benefits for EPO‐treated patients [44]. Finally, more white matter hyperintensities were associated with greater cognitive decline for donepezil‐ versus placebo‐treated older MDD individuals [39].

Regarding functional neuroimaging markers, dorsal prefrontal cortex (dPFC) hypoactivity during a working memory task at baseline was associated with greater efficacy of CR on executive function [61]. Further, higher baseline [18F] FDDNP binding (a measure of brain amyloid and neurofibrillary tangle load) in the frontal lobe predicted more executive function benefits across participants treated with memantine or placebo adjunctive to escitalopram [41].

A study comparing high‐ and low‐dose tDCS in depressed outpatients found that Brain‐Derived Neurotrophic Factor (BDNF) Val66Met and Catechol‐O‐methyltransferase (COMT) Val158Met polymorphisms interacted with tDCS dose, resulting in greater improvements in verbal memory and fluency in the high‐dose group [72]. No other examinations of genetic factors showed significant findings [39, 46].

Other baseline biomarkers associated with post‐intervention cognitive changes included inflammatory cytokine levels [51], cortisol awakening response [55], and olfactory capacity [49]. However, for all three, there were also examinations showing no significant associations [36, 39, 53, 67], rendering these findings inconsistent.

3.6.4 Clinical and Illness‐Related Factors

In this broad category, the relationship between 16 candidate predictors and treatment response on cognition was examined in 44 instances by 20 studies (N = 1881), but a significant association was only reported five times (11%). Notably, baseline depression severity did not moderate any post‐treatment cognitive improvement despite being the most intensively investigated clinical variable (Table 2). Likewise, treatment response was unrelated to baseline medication use, mood disorder diagnosis type, BD subtype, psychosis history, number of mood episodes or hospitalizations, baseline anxiety or (hypo)mania, childhood trauma, or previous/current psychotherapies (investigated in a few studies; Table 2).

Two of four studies examining the role of baseline remission status showed significant associations with cognitive improvements but in opposite directions (Table 2). In one trial testing pramipexole versus placebo in fully or partially remitted BD outpatients, the fully remitted subgroup experienced greater post‐treatment improvements compared to the full‐sample analysis [37]. In contrast, another trial found that depressive symptoms (vs. being in remission) at baseline predicted greater memory improvements for MDD participants receiving online CR [65].

A large secondary analysis found that elderly depressed outpatients with illness onset after age 50 showed greater post‐treatment gains in global cognition across vortioxetine, duloxetine, and placebo groups [54]. Finally, individual trials found better responses to psychological therapies in those with shorter illness duration [60] and mood disorders versus schizophrenia‐spectrum diagnoses [70], though these effects were not specific to the treatment groups.

3.6.5 Psychological Factors

Four studies (N = 204) examined psychological predictors of cognitive outcomes in partially or fully remitted MDD outpatients, but none found significant associations. Motivation did not differ between attention improvers and non‐improvers after CR [60]. Further, therapy‐related hopes, expectations, and skepticism were unrelated to cognitive improvements [65], and baseline rumination did not moderate working memory training effects [71].

4 Discussion

This systematic review aimed to identify baseline factors associated with treatment response in primary or secondary analyses of RCTs on pro‐cognitive interventions for mood disorders. We identified 20 baseline factors predicting cognitive outcomes of pro‐cognitive interventions, but with limited replication and evidence convergence between studies. Of the factors considered, poorer cognitive function at baseline was the most consistent predictor of greater post‐treatment improvements. In contrast, depression severity and medication use consistently showed no significant associations with changes in post‐treatment cognitive outcomes. Other clinical and demographic variables, albeit less intensively investigated, also showed no consistent association with treatment response. Structural and functional differences in the prefrontal cortex, temporal lobe, and hippocampus were the most identified neuroimaging biomarkers of cognitive improvement, though examined in only a few studies. More than 75% of the included studies presented with some methodological concerns or a high risk of bias.

4.1 Implications and Recommendations

The association between poorer baseline cognition and greater treatment‐related cognitive improvement was the most consistently replicated finding in this review, reported in five of ten studies [45, 48, 56, 59, 69]. In contrast, four studies found no significant association, and one small rTMS study reported a non‐specific opposite effect, where better baseline executive function predicted improvement across both active and sham groups [73]. This may reflect practice effects, as individuals with executive impairments typically show reduced gains from repeated testing, whereas those with better cognition may improve simply due to retesting [77]. These observations highlight the need to examine baseline cognition specifically in relation to treatment effects, rather than interpreting overall cognitive change, which may be confounded by practice effects or regression to the mean. Despite this, only three of six studies tested for an interaction between baseline cognition and treatment arm (Table 3), all of which found a significant influence [25, 48, 56]. Analytical approaches and definitions of cognitive impairment varied (Table S1), reflecting heterogeneity in how baseline cognition was operationalised and assessed. Importantly, associations were found in studies both with and without pre‐screening for cognitive impairment.

This association between poorer baseline cognition and better treatment response extends findings from a prior review [22]. Further support comes from non‐randomised studies showing that individuals with greater cognitive deficits are more likely to benefit from pro‐cognitive interventions [79]. These results align with the hypothesis that lower baseline cognitive levels provide greater ‘room for improvement,’ while intact or mildly impaired individuals experience ceiling effects [30].

Taken together, these findings underscore the importance of pre‐screening for objective cognitive impairment, aligning with ISBD Taskforce recommendations [27, 30] (Table 3). Brief screening tools, such as the Screen for Cognitive Impairment in Psychiatry (SCIP; [80]), offering both global and domain scores, are recommended. Screening for domain‐specific impairment should involve more than one test per domain, and trials should avoid defining cognitive impairment by a single test as the primary outcome [27]. Based on prior recommendations, feasible thresholds include 0.5 SD below normative means for composite scores and 1 SD below the mean on two or more tests for domain scores [30]. Given that many individuals with mood disorders show no or only mild cognitive impairment [81], strict pre‐screening based on norms could impede recruitment [82]. An alternative is estimating illness‐related cognitive decline by comparing current performance to premorbid IQ, using discrepancy scores [83]. Tools like the National Adult Reading Test (NART; [84]) or Test of Premorbid Functioning (TOPF; [85]) can assist. Applying a predefined discrepancy threshold enables identifying individuals with relative impairments—those likely to benefit—while avoiding recruitment barriers [86]. In trials where cognition is a secondary outcome, post hoc analyses adjusting for premorbid IQ are recommended.

Unlike objective measures, baseline subjectively reported cognitive difficulties showed little association with objective cognitive outcomes, consistent with poor correspondence between subjective and objective cognition in mood disorders [87, 88]. This highlights that relying solely on self‐reports for trial inclusion is insufficient. Similarly, baseline medication use showed no consistent association with cognitive change when participants had stable medication regimens, supporting the continued inclusion of stably medicated participants in cognition trials [27].

Notably, baseline depressive symptom severity showed no impact on treatment‐related cognitive gains, despite varying study designs, sample sizes, interventions, and outcomes. Cognitive improvements thus appear largely independent of baseline mood, suggesting strict pre‐screening for remission may not be essential when cognition is the primary outcome. Although depressive symptoms differed across studies, mean baseline scores indicated mild symptoms overall (HDRS: M = 14, k = 6; BDI‐II: M = 21, k = 2; MADRS: M = 14, k = 1). This finding suggests an updating of earlier recommendations to restrict inclusion to fully or partially remitted patients [27], allowing inclusion of patients with slightly higher depressive symptom levels. However, when the aim is to target trait‐like cognitive deficits persisting during remission—and to separate cognitive from mood improvements—enrolling partially or fully remitted participants remains advisable. In trials where this distinction is less critical, mood fluctuations should be controlled for by including mood changes as covariates in analyses. Severe acute mood symptoms may impair engagement, especially with psychological interventions, reducing treatment effects [58]. Thus, a minimum degree of mood stability should be ensured prior to enrollment to support participation and interpretation of results.

Finally, identifying biomarkers of treatment response can clarify neurobiological mechanisms underlying cognitive improvements and highlight future intervention targets [30]. Several promising biomarkers were found, particularly structural and functional neuroimaging measures involving the prefrontal cortex, temporal cortex, and hippocampus. Although findings varied, this may reflect the distributed nature of cognitive processes. Most studies were grounded in strong theoretical frameworks about neural intervention effects. However, small samples testing multiple biomarkers likely increased type I error risk [89]. To improve insights into neural target engagement, neuroimaging should be hypothesis‐driven and focused on brain abnormalities linked to targeted cognitive impairments [27]. Nevertheless, the clinical utility of neuroimaging biomarkers remains limited due to accessibility and cost barriers.

4.2 Methodological Limitations and Considerations

A major limitation across the included studies was poor methodological quality, with 77% rated as having a high risk of bias. This was often due to cognition being examined as a secondary or exploratory outcome. Other key issues included a lack of prespecified primary outcomes, deviations from specified outcomes, limited reporting of randomisation procedures, and inadequate or missing handling of missing data—particularly in non‐pharmacological trials. Furthermore, studies frequently assessed multiple predictors and cognitive outcomes without correcting for multiple comparisons, increasing the risk of false positives. Conversely, small to moderate sample sizes may have led to false negatives due to insufficient statistical power. Although methodological limitations may thus introduce false positive or negative associations, this risk is reduced when predictors show consistent results across multiple studies. For example, baseline cognition demonstrated associations in most of 10 studies, whereas baseline depression severity showed none across nine, suggesting reliable patterns rather than random error.

Heterogeneity across studies—in intervention modalities, outcome measures, predictor assessments, analytical approaches, and sample characteristics (e.g., mood state at entry)—further limited replication and hindered robust conclusions. This diversity also precluded meta‐analysis. Nevertheless, qualitative synthesis offered preliminary insights into factors predicting response to pro‐cognitive interventions and informed recommendations for future research. Although publication bias was not formally assessed, the frequent reporting of non‐significant associations suggests a relatively low risk.

An important methodological consideration for future cognition trials is the approach to analyzing baseline predictors of response. Across the studies reviewed, we identified a range of methods, including correlation analyses, linear and logistic regression, subsample analyses, and subgroup comparisons (Table 2). While all are relevant, it is essential to determine whether observed associations are specific to the active treatment or reflect natural regression to the mean [90]. Moderation analysis—including an interaction term between treatment and baseline predictor—is recommended [91, 92], as it allows identification of treatment‐specific effects.

Another limitation was the potential confounding effects of concomitant pharmacotherapy, as most studies employed naturalistic medication management [93].

Finally, the limited and inconsistent efficacy of cognitive interventions themselves remains a critical challenge. Without a consistently strong treatment signal, the identification of robust predictors is inherently constrained. Addressing this will be essential for advancing precision approaches in cognition trials.

4.3 Future Directions

Future cognition trials should systematically examine and report baseline predictors of treatment response to expand the evidence base. While positive results are preferentially published [94], reporting of null findings is equally crucial to avoid bias and enhance understanding. The limited replication observed in this review likely reflects small sample sizes and exploratory study designs. Advancing the field requires large, adequately powered, hypothesis‐driven studies [76], with data sharing or pooling efforts needed to reliably estimate the effects of candidate predictors, such as baseline cognition.

To minimise variability, future studies should adopt standardised cognitive outcomes [95]. In line with previous recommendations, the primary outcome should ideally be a composite score covering core cognitive domains, unless a specific domain is targeted—where the use of multiple tests for that domain is advised [27]. Calculation of a g‐factor via Principal Component Analysis across the cognitive battery is also recommended to enhance sensitivity and comparability.

By strengthening methodological rigor and standardization, future trials can more reliably identify predictors of cognitive improvement and move the field closer to precision‐guided interventions for individuals with mood disorders.

4.4 Conclusions

This systematic review provides initial evidence on baseline factors influencing treatment response and outcome variability in cognition trials for mood disorders. We identified five broad predictor categories, most of which showed non‐significant findings, conflicting associations, or lacked replication. Substantial sources of bias were also evident across studies. Baseline objective cognition emerged as the most consistent predictor of response to pro‐cognitive interventions, although not uniformly across all studies. While preliminary, these findings support several methodological recommendations previously issued by the ISBD Targeting Cognition Taskforce (summarized in Table 3) and offer guidance for harmonizing future research. Improving the methodological quality of cognition trials remains a key priority. To reliably estimate the effects of baseline cognition and other candidate predictors, adequately powered, hypothesis‐driven analyses in large cohorts or merged datasets are essential. A clearer understanding of pre‐treatment factors may ultimately enable more efficient and effective application of pro‐cognitive strategies for individuals with mood disorders.

Ethics Statement

The authors have nothing to report.

Conflicts of Interest

Christopher R. Bowie has been a consultant for Boehringer Ingelheim. He receives book royalties from Oxford University Press. He has received in‐kind research user accounts from Scientific Brain Training Pro. Kate Burdick received honorarium and grant funding for her leadership role in the Breakthrough Discoveries for thriving with Bipolar Disorder (BD2) and she serves on scientific advisory boards for Suven Life Sciences and Alto Neuroscience. Katie Douglas reports using software provided free of charge by Scientific Brain Training Pro for Cognitive Remediation trials. Gregor Hasler reports speaker/consultant fees from Janssen, Lundbeck, OM Pharma, Otsuka, Sanofi, Schwabe, Servier, Sunovion and Takeda. Beny Lafer has received grants from Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES), Conselho Nacional de Desenvolvimento Científico e Tecnólógico (CNPq) and the Baszucki Brain Research Fund. Michail Kalfas has received honoraria from Neurocentrx Pharma Ltd. and is supported by the National Institute for Health Research (NIHR) Biomedical Research Centre. Hanne L. Kjærstad has received honoraria from Lundbeck. Anabel Martinez‐Aran receives support by d'Economia i Coneixement (2021 SGR 01128), the Centro de Investigación Biomédica en Red de Salud Mental‐CIBERSAM and the Centres de Recerca de Catalunya‐CERCA Programmme. Roger S. McIntyre has received research grant support from CIHR/GACD/National Natural Science Foundation of China (NSFC) and the Milken Institute; speaker/consultation fees from Lundbeck, Janssen, Alkermes, Neumora Therapeutics, Boehringer Ingelheim, Sage, Biogen, Mitsubishi Tanabe, Purdue, Pfizer, Otsuka, Takeda, Neurocrine, Neurawell, Sunovion, Bausch Health, Axsome, Novo Nordisk, Kris, Sanofi, Eisai, Intra‐Cellular, NewBridge Pharmaceuticals, Viatris, Abbvie, Atai Life Sciences. Dr. Roger McIntyre is a CEO of Braxia Scientific Corp. Kamilla W. Miskowiak is supported by a European Research Council (ERC) Consolidator Grant. She has received honoraria from Allergan, Gedeon Richter, Angelini and Lundbeck in the past 3 years. Richard J. Porter reports use of computer software for research –provided at no cost by SBT‐pro and funding for travel to educational meetings by Lundbeck and Servier. Ivan J. Torres has received consulting fees from Boehringer Ingelheim (Canada). Tamsyn E. Van Rheenen is supported by an Al and Val Rosenstrauss Fellowship from the Rebecca L Cooper Medical research Foundation and a Dame Kate Campbell Fellowship from the University of Melbourne. Paul R.A. Stokes reports grant funding from the Medical Research Council UK, National Institute for Health and Care Research (NIHR), H. Lundbeck A/S and King's Health Partners, funding from NIHR, non‐financial support from Janssen Research and Development LLC for an MRC funded study led by PRAS, editorial honoraria and non‐financial support from Frontiers in Psychiatry outside the submitted work. He is a member of the UK Advisory Council on the Misuse of Drugs (ACMD), and Speciality Chief Editor, Mood Disorders section, Frontiers in Psychiatry. Eduard Vieta has received grants and served as consultant, advisor or CME speaker for the following entities (unrelated to the present work): AB‐Biotics, Abbott, Abbvie, Aimentia, Angelini, Biogen, Biohaven, Boehringer Ingelheim, Casen‐Recordati, Celon, Compass, Dainippon Sumitomo Pharma, Ethypharm, Ferrer, Gedeon Richter, GH Research, Glaxo Smith‐Kline, Idorsia, Janssen, Lundbeck, Novartis, Organon, Otsuka, Rovi, Sage, Sanofi‐Aventis, Sunovion, Takeda, and Viatris. Lakshmi N. Yatham reports speaker/consultant fees from Abbvie, Alkermes, DSP, Gedeon Richter, Intracellular Therapies, Merck, Otsuka, Sanofi, Sunovion, and grant funding from Allergan (now AbbVie), CIHR, and Dainippon Sumitomo outside the submitted work. Allan H. Young receives funding from the National Institute for Health and Care Research (NIHR) Maudsley Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King's College London. Lars Vedel Kessing reports speaker fees from Lundbeck and Teva. Ayal Schaffer reports speaker/consultant fees from Abbvie, Lundback and Otsuka. Jeff Zarp has received honoraria from Lundbeck. The remaining authors report no conflicts of interest. The views expressed are those of the author(s) and not necessarily those of the NIHR or the Department of Health and Social Care.