CoChat
Sign Up
Cited by 2Cognitive Abilities and TestingCognitive Functions and MemoryNeural and Behavioral Psychology Studies

Meta-analysis of computerised working memory training: behavioural gains, training parameters, transfer mechanisms, and neural correlates

Open access
Li, Geng, Liu, Yang, Chen, Antao·2026·npj Digital Medicine
DOI 10.1038/s41746-026-02478-9
Open this paper in CoChat — free accountOpen PDF

Abstract

Working memory training (WMT) is widely used to enhance cognitive task performance, yet traditional delivery poses challenges for sustained implementation; computerised working memory training (CWMT) provides a scalable digital format with standardised delivery and integrated monitoring and feedback. We conducted the first meta-analysis integrating behavioural and neuroimaging evidence to quantify the behavioural effects and neural correlates of CWMT, drawing on 45 neuroimaging studies. Multivariate meta-analysis indicated a moderate benefit of CWMT relative to controls (Hedges' g = 0.503, 95% CI [0.363-0.642]). Seed-based d mapping (SDM) identified training-related decreases in activation in the left angular gyrus (L-AG), bilateral superior frontal gyrus (SFG), right inferior parietal lobule (R-IPL), left cerebellum, and right middle frontal gyrus (R-MFG), a pattern compatible with reduced recruitment following training. Moderator analyses showed significant effects of task type, training compliance, total training dose, and age, but not cognitive status and sex. Moreover, CWMT-induced brain activation changes were associated with behavioural improvements, and significant co-activation was observed among the brain regions identified in the overall analysis. Together, these findings provide convergent evidence that CWMT is associated with improved cognitive task performance and reproducible modulation of task-related activation, supporting its relevance as a scalable digital approach to cognitive health.

AI snapshot

Written from the paper’s full text.

TL;DR

This first meta-analysis integrating behavioural and neuroimaging evidence from 45 computerised working memory training (CWMT) studies (1,492 participants) found moderate cognitive gains versus controls (Hedges’ g = 0.503, 95% CI [0.363–0.642]) and reproducible task-related activation decreases in fronto-parietal-cerebellar regions. Task type, training dose, compliance, and age moderated outcomes, whereas cognitive status and sex did not.

Key findings

  • CWMT significantly improved cognitive task performance relative to controls: Hedges’ g = 0.503, 95% CI [0.363–0.642], t = 7.132, p < 0.0001. Gains were larger for trained tasks (g = 0.677, 95% CI [0.467–0.886]) than untrained tasks (g = 0.355, 95% CI [0.189–0.520]), with a significant task-type moderator effect (F = 8.358, p = 0.004).
  • SDM meta-analysis identified CWMT-related decreases in activation in the left angular gyrus (L-AG), bilateral superior frontal gyrus (SFG), right inferior parietal lobule (R-IPL), left cerebellum, and right middle frontal gyrus (R-MFG); no regions showed CWMT > controls activation increases.
  • Cognitive status did not significantly moderate behavioural gains (F = 0.167, p = 0.682), with comparable effects in cognitively unimpaired (g = 0.483, 95% CI [0.345–0.620]) and cognitively impaired (g = 0.544, 95% CI [0.026–1.061]) participants.
  • Total training dose was associated with larger SFG activation decreases (β = −0.0002, p = 0.011) and greater behavioural improvement (β = 0.0004, p = 0.021); training compliance was negatively associated with SFG activation changes (β = −0.007, p = 0.001); older age was associated with larger R-MFG decreases (β = −0.003, p = 0.033). Greater behavioural improvement covaried with larger L-AG decreases (β = −0.094, p = 0.036).

Methods

  • Systematic search of PubMed, Web of Science, APA PsycINFO, MEDLINE, and Embase returned 4,659 records; after screening, 45 neuroimaging studies met inclusion criteria, comprising 1,492 participants (mean age 33.71 years, range 11.9–69.6; 47.06% female; 67.27% cognitively healthy). Training averaged 4.67 sessions/week, 38.86 min/session, over 4.65 weeks; 85.45% used adaptive difficulty.
  • Behavioural synthesis used a three-level multivariate meta-analysis in R (metafor and clubSandwich), modelling effect sizes nested within samples and studies, with Hedges’ g as the effect size and within-sample correlation fixed at 0.5.
  • Neural synthesis used seed-based d mapping (SDM version 5.15) on whole-brain peak coordinates, with 20 mm FWHM kernel, 500 randomisations, random-effects inference, and cluster-forming threshold of uncorrected p < 0.0001, peak SDM-Z ≥ 1, and ≥10 contiguous voxels. Moderator and regression analyses tested task type, cognitive status, age, sex, total training dose, and compliance; robustness checks included jackknife sensitivity, alternative within-study correlations, n-back load/stimulus-type subgroups, non-independence exclusions, and Egger’s tests.

Limitations

  • The paper states substantial heterogeneity across training tasks/parameters, sample characteristics, and imaging paradigms; many primary studies were modest in size; Egger’s tests did not indicate small-study effects but have limited power in heterogeneous literatures, so selective reporting cannot be fully excluded.
  • The paper states that SDM analyses used an uncorrected voxel-wise threshold (p < 0.0001), so residual false-positive risk inherent to coordinate-based whole-brain meta-analyses cannot be fully quantified or eliminated; pooled estimates should not be over-interpreted as evidence for preventing dementia onset or slowing disease progression.
  • The paper states behavioural outcomes were largely task-based, and observed improvements may partly reflect practice-related gains and strategy optimisation rather than durable increases in cognitive capacity; the text also makes evident that mechanistic interpretations of activation attenuation (efficiency, strategy change, habituation, reduced control demands) cannot be disentangled from coordinate-based evidence.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41746-026-02478-9

Cites

105 of 105 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

16 of 43 cited sources checked · 4 supported · 1 flagged · accuracy not yet measured

  • Interpretationcitation

    Reduced activation in SFG and R-MFG may indicate diminished control demands with training or shifts in processing mode.

    “Similarly, reduced activation in the SFG and R–MFG—prefrontal nodes involved in sustained attention, updating, and interference control—may indicate diminished control demands with training or shifts in processing mode67,75,76.”
    Quote in the paperNot a substantial claim1 of 3 sources checked
    • cites [67] Schneiders, J. A. et al. The impact of auditory working memory training on the fronto-parietal working memory network. Front. Hum. Neurosci. 6, 173 (2012). source

      The Discussion (fMRI results; page unavailable) links reduced activation in the superior portion of the right middle frontal gyrus to more efficient processing and less demand on attentional control. It does not report a corresponding decrease in the superior frontal gyrus (SFG), and it describes strategic changes in processing after working-memory training as unlikely. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites [75] Kelly, A. M. C. & Garavan, H. Human functional neuroimaging of brain changes associated with practice. Cereb. Cortex N. Y. N. 15, 1089–1102 (2005). source
    • cites [76] Wager, T. D. & Smith, E. E. Neuroimaging studies of working memory: a meta-analysis. Cogn. Affect. Behav. Neurosci. 3, 255–274 (2003). source
  • Interpretationcitation

    Behavioural improvements may partly reflect practice-related gains and strategy optimisation rather than durable increases in cognitive capacity.

    “Finally, behavioural outcomes were largely task-based, and observed improvements may partly reflect practice-related gains and strategy optimisation rather than durable increases in cognitive capacity79.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [79] Malinovitch, T., Jakoby, H. & Ahissar, M. Training-induced improvement in working memory tasks results from switching to efficient strategies. Psychon. Bull. Rev. 28, 526–536 (2021). source
  • Interpretationcitation

    Decreased activation in L-AG and R-IPL may reflect changes in task representations and reduced reliance on parietal engagement.

    “Decreased activation in the L–AG and R–IPL—parietal regions implicated in information maintenance and manipulation—may reflect changes in task representations and reduced reliance on parietal engagement56,67, consistent with their correlated changes and prior evidence for coupling between default-mode and frontoparietal control systems.”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [56] Nęcka, E. et al. The effects of working memory training on brain activity. Brain Sci. 11, 155 (2021). source

      The Discussion (page unavailable) reports decreased parietal activation in participants with lower task accuracy, but identifies the regions as the left superior parietal lobule and right posterior supramarginal gyrus, not the L–AG and R–IPL. It does not interpret the decrease as a change in task representations or reduced reliance on parietal engagement; instead, it reports greater activation with better accuracy and says it found no support for a general neural-efficiency explanation. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
    • cites [67] Schneiders, J. A. et al. The impact of auditory working memory training on the fronto-parietal working memory network. Front. Hum. Neurosci. 6, 173 (2012). source

      The Results (Brain imaging results; page unavailable) report decreased activation in the right inferior parietal lobule after auditory working-memory training. The Discussion (fMRI results; page unavailable) interprets this as reduced demand on attentional control, supporting the reduced-parietal-engagement interpretation. The paper does not establish the same decrease in the left angular gyrus or attribute these decreases to changed task representations. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
  • Interpretationcitation

    The pattern of decreased activation is compatible with a training-related rebalancing of recruitment within a fronto-parietal-cerebellar circuit.

    “Together, these patterns are compatible with a training-related rebalancing of recruitment within a fronto–parietal–cerebellar circuit, although attenuation could reflect strategy change, habituation, reduced post-training control demands, or increased neural efficiency79.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [79] Malinovitch, T., Jakoby, H. & Ahissar, M. Training-induced improvement in working memory tasks results from switching to efficient strategies. Psychon. Bull. Rev. 28, 526–536 (2021). source

      The Discussion and conclusions reports that improvement on the n-back task can be explained by a switch to an efficient strategy, supporting strategy change as a possible explanation. It does not address decreased brain activation or rebalancing of recruitment in a fronto–parietal–cerebellar circuit; the citation therefore does not support that neural interpretation. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
  • Findingown data

    Behavioural improvements were not significantly moderated by training compliance.

    “behavioural improvements were not significantly moderated by training compliance, sex, or age.”
    Quote in the paperNot a substantial claim
  • Classified as probably not a substantial claim. 25
  • May combine two claims. 32

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation

    To strengthen methodological rigour, the reporting quality of included neuroimaging studies was evaluated using a 17-item checklist.

    “To strengthen methodological rigour, we evaluated the reporting quality of included neuroimaging studies using a 17-item checklist98.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [98] Lavallé, L., Brunelin, J., Jardri, R., Haesebaert, F. & Mondino, M. The neural signature of reality-monitoring: a meta-analysis of functional neuroimaging studies. Hum. Brain Mapp. 44, 4372–4389 (2023). source

      The Methods section (‘Quality assessment’) says the source paper used a checklist to assess study quality, but explicitly states that its final checklist had 18 items, not 17. The cited paper therefore does not support the attributed checklist length. (LLM, conf=0.91)

      Checked by a reasoning model — may be wrong.

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    Working memory is a core component of human cognition, enabling short-term maintenance and online manipulation of information while supporting higher-order processes including attentional regulation, executive control, and reasoning.

    “Working memory is a core component of human cognition, enabling the short-term maintenance and online manipulation of information while supporting higher-order processes including attentional regulation, executive control, and reasoning7–9.”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites [7] Zhang, Y., Fu, J. & Zhao, X. Neural correlates of working memory training: An fMRI meta-analysis. NeuroImage 301, 120885 (2024). source
    • cites [8] Vartanian, O. et al. What Is targeted when we train working memory? Evidence from a meta-analysis of the neural correlates of working memory training using activation likelihood estimation. Front. Psychol. 13, 868001 (2022). source

      The Introduction defines working memory as active maintenance during ongoing processing and describes information being maintained and manipulated. The Discussion treats working-memory capacity as a resource that can enable cognitive processes, but the cited passages do not specifically establish its role in attentional regulation, executive control, and reasoning. (LLM, conf=0.82)

      Experimental — accuracy not yet measured
    • cites [9] Grynszpan, O. et al. Efficacy and specificity of computer-assisted cognitive remediation in schizophrenia: a meta-analytical study. Psychol. Med. 41, 163–173 (2011). source

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Working memory impairment is prevalent across neuropsychiatric and neurological conditions and is among the most sensitive early indicators of cognitive function decline, with links to everyday functional limitations and reduced quality of life.

    “Working memory impairment is prevalent across neuropsychiatric and neurological conditions and is among the most sensitive early indicators of cognitive function decline, with clear links to everyday functional limitations and reduced quality of life10–13.”
    Quote in the paperNot a substantial claim3 of 4 sources checked
    • cites [10] Constantinidis, C. & Klingberg, T. The neuroscience of working memory capacity and training. Nat. Rev. Neurosci. 17, 438–449 (2016). source
    • cites [11] Jiang, Y., Guo, Z., Zhou, X., Jiang, N. & He, J. Exploration of working memory retrieval stage for mild cognitive impairment: time-varying causality analysis of electroencephalogram based on dynamic brain networks. J. Neuroeng. Rehabil. 22, 58 (2025). source

      Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured
    • cites [12] Cloutier, S., Chertkow, H., Kergoat, M.-J., Gauthier, S. & Belleville, S. Patterns of cognitive decline prior to dementia in persons with mild cognitive impairment. J. Alzheimers Dis. JAD 47, 901–913 (2015). source

      Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured
    • cites [13] Kirova, A.-M., Bays, R. B. & Lagalwar, S. Working memory and executive function decline across normal aging, mild cognitive impairment, and Alzheimer’s disease. BioMed. Res. Int. 2015, 748212 (2015). source

      Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    Existing behavioural and neuroimaging syntheses are constrained by inadequate modelling of within-study dependent effect sizes and absence of an effect-size framework that can jointly integrate activation increases and decreases.

    “Nevertheless, existing behavioural and neuroimaging syntheses remain constrained by key methodological limitations, including inadequate modelling of within-study dependent effect sizes and the absence of an effect-size framework that can jointly integrate activation increases and decreases7,8,20.”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites [7] Zhang, Y., Fu, J. & Zhao, X. Neural correlates of working memory training: An fMRI meta-analysis. NeuroImage 301, 120885 (2024). source
    • cites [8] Vartanian, O. et al. What Is targeted when we train working memory? Evidence from a meta-analysis of the neural correlates of working memory training using activation likelihood estimation. Front. Psychol. 13, 868001 (2022). source

      The Introduction’s ‘Aims of Present Meta-Analysis’ section says prior neuroimaging meta-analyses found both activation increases and decreases, and this paper examines the directions separately. The Method’s ‘Activation Likelihood Estimation’ section describes pooling activation coordinates, but neither passage identifies inadequate modelling of within-study dependent effect sizes or a missing effect-size framework for jointly integrating both directions. Searches for those methodological limitations found no supporting passage; page numbers were unavailable in our copy. (LLM, conf=0.72)

      Experimental — accuracy not yet measured
    • cites [20] Melby-Lervåg, M. & Hulme, C. Is working memory training effective? A meta-analytic review. Dev. Psychol. 49, 270–291 (2013). source

  • May combine two claims.

    Premisecitation

    The cerebellum contributes to cognitive sequencing and temporal organisation via cerebello-frontal circuits, which may support information integration and cross-task transfer.

    “The cerebellum, beyond motor coordination, contributes to cognitive sequencing and temporal organisation via cerebello–frontal circuits, which may support information integration and cross-task transfer78,83.”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [78] Stoodley, C. J. & Schmahmann, J. D. Functional topography in the human cerebellum: a meta-analysis of neuroimaging studies. NeuroImage 44, 489–501 (2009). source
    • cites [83] Marvel, C. L. & Desmond, J. E. Functional topography of the cerebellum in verbal working memory. Neuropsychol. Rev. 20, 271–279 (2010). source

      The Functional Dissociation Between the Dorsal and Ventral Dentate section (page unavailable) describes cerebellar connections with prefrontal regions involved in working memory and executive control. Cerebellar Support of Covert Speech and Working Memory (page unavailable) discusses timing of motor sequences and the organisation of task-relevant information through inner speech. The paper does not establish the broader claim about cognitive temporal organisation or cross-task transfer. (LLM, conf=0.84)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationcitation

    If older or higher-risk individuals rely more on compensatory control recruitment at baseline, post-training attenuation may reflect reduced compensatory demand from a scaffolding perspective.

    “From a scaffolding perspective82, if older or higher-risk individuals rely more on compensatory control recruitment at baseline, post-training attenuation may reflect reduced compensatory demand.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [82] Park, D. C. & Reuter-Lorenz, P. The adaptive brain: aging and neurocognitive scaffolding. Annu Rev. Psychol. 60, 173–196 (2009). source

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationcitation

    Attenuation of activation could reflect strategy change, habituation, reduced post-training control demands, or increased neural efficiency.

    “Together, these patterns are compatible with a training-related rebalancing of recruitment within a fronto–parietal–cerebellar circuit, although attenuation could reflect strategy change, habituation, reduced post-training control demands, or increased neural efficiency79.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [79] Malinovitch, T., Jakoby, H. & Ahissar, M. Training-induced improvement in working memory tasks results from switching to efficient strategies. Psychon. Bull. Rev. 28, 526–536 (2021). source

      The Discussion and conclusions (page unavailable) links training-related improvement to switching to strategies that reduce working-memory manipulations, supporting strategy change as a possible explanation. Our copy does not connect those strategies to attenuated neural activation or discuss habituation, reduced post-training control demands, or increased neural efficiency as explanations for attenuation. (LLM, conf=0.87)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Interpretationown data

    The left cerebellum showed decreased activation and its covariation with R-IPL suggests cerebello-parietal pathways may contribute to training-related functional modulation.

    “The left cerebellum, a region linked to sequencing and temporal organisation77,78, also showed decreased activation; its covariation with the R–IPL suggests that cerebello–parietal pathways may contribute to training-related functional modulation.”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [77] Stoodley, C. J., Valera, E. M. & Schmahmann, J. D. Functional topography of the cerebellum for motor and cognitive tasks: an fMRI study. NeuroImage 59, 1560–1570 (2012). source
    • cites [78] Stoodley, C. J. & Schmahmann, J. D. Functional topography in the human cerebellum: a meta-analysis of neuroimaging studies. NeuroImage 44, 489–501 (2009). source

  • Classified as probably not a substantial claim.

    Study design and conductcitation

    The study systematically synthesised evidence on task-related activation changes following CWMT in line with established methodological standards for neuroimaging meta-analyses.

    “In line with established methodological standards for neuroimaging meta-analyses85, we systematically synthesised evidence on task-related activation changes following CWMT.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [85] Müller, V. I. et al. Ten simple rules for neuroimaging meta-analysis. Neurosci. Biobehav. Rev. 84, 151–161 (2018). source

  • May combine two claims.

    Findingown data

    CWMT-related changes were detectable in both cognitively unimpaired and impaired samples without significant between-group differences.

    “CWMT-related changes were detectable in both cognitively unimpaired and impaired samples without significant between-group differences”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    This meta-analysis provides convergent support for CWMT-induced cognitive improvements alongside reproducible changes in task-related brain function.

    “In conclusion, by integrating behavioural and neuroimaging evidence, this meta-analysis provides convergent support for CWMT-induced cognitive improvements alongside reproducible changes in task-related brain function.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Findingnone given

    The current evidence is derived from task-related behavioural performance and corresponding brain activation.

    “It is important to note that the current evidence is derived from task-related behavioural performance and corresponding brain activation.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Task type, training compliance, total training dose, and age significantly influenced outcomes.

    “Moderator analyses further revealed that task type, training compliance, total training dose, and age significantly influenced outcomes”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Cognitive status and sex had limited impact on outcomes.

    “whereas cognitive status and sex had limited impact.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Computerised working memory training (CWMT) standardises delivery, captures process-level metrics to quantify realised dose, and adapts difficulty in real time.

    “By contrast, computerised working memory training (CWMT) standardises delivery, captures process-level metrics to quantify realised dose, and adapts difficulty in real time19.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [19] Fang, P. et al. Effects of computerized working memory training on neuroplasticity in healthy individuals: a combined neuroimaging and neurotransmitter study. NeuroImage 298, 120785 (2024). source

  • May combine two claims.

    Premisecitation

    WMT improves performance on trained tasks and yields measurable behavioural gains on untrained tasks.

    “Accumulating evidence indicates that WMT improves performance on trained tasks and yields measurable behavioural gains on untrained tasks, supporting its promise as a non-pharmacological approach to cognitive health14,15.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [14] López Zunini, R. A. et al. Event-related potentials elicited during working memory are altered in mild cognitive impairment. Int. J. Psychophysiol. J. Int. Organ. Psychophysiol. 109, 1–8 (2016). source
    • cites [15] Peng, P. et al. A meta-analysis on the relation between reading and working memory. Psychol. Bull. 144, 48–76 (2018). source

  • May combine two claims.

    Interpretationreasoning

    The observed attenuation is compatible with both efficiency-related and compensatory-demand accounts, and available evidence does not allow a definitive mechanistic attribution.

    “Overall, the observed attenuation is compatible with both efficiency-related and compensatory-demand accounts, and the available evidence does not allow a definitive mechanistic attribution.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    The R-MFG is a key node of the prefrontal control network and supports task-set maintenance and cognitive flexibility.

    “Similarly, the R–MFG, a key node of the prefrontal control network, supports task-set maintenance and cognitive flexibility80,81.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [80] Miller, E. K. & Cohen, J. D. An integrative theory of prefrontal cortex function. Annu. Rev. Neurosci. 24, 167–202 (2001). source
    • cites [81] Dosenbach, N. U. F., Fair, D. A., Cohen, A. L., Schlaggar, B. L. & Petersen, S. E. A dual-networks architecture of top-down control. Trends Cogn. Sci. 12, 99–105 (2008). source

  • May combine two claims.

    Study design and conductreasoning

    Egger's tests did not indicate small-study effects, but such tests have limited power in heterogeneous literatures and selective reporting cannot be fully excluded.

    “although Egger’s tests did not indicate small-study effects (publication bias), such tests have limited power in heterogeneous literatures and selective reporting cannot be fully excluded.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning

    CWMT may serve as a low-threshold preventive option in the general population and as an adjunct module for at-risk individuals.

    “consistent with a stratified use case in which CWMT may serve as a low-threshold preventive option in the general population and as an adjunct module for at-risk individuals.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductown data

    Sensitivity analysis excluding potentially non-independent studies retained only one study per first-author set; the overall pattern of findings was unchanged in the SDM meta-analysis.

    “To address potential non-independence related to shared first authorship, we conducted a conservative sensitivity analysis excluding potentially non-independent studies31,48,54,60,67, retaining only one study per first-author set. The overall pattern of findings was unchanged in the SDM meta-analysis (Supplementary Table S5)”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationreasoning

    Higher-load conditions more strongly engage working-memory circuitry and operate at higher baseline activation, leaving greater dynamic range for training-related attenuation of task-related responses.

    “A plausible explanation is that higher-load conditions more strongly engage working-memory circuitry and operate at higher baseline activation, leaving greater dynamic range for training-related attenuation of task-related responses.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Pooled estimates should not be over-interpreted as definitive evidence for preventing dementia onset or slowing disease progression.

    “pooled estimates should not be over-interpreted as definitive evidence for preventing dementia onset or slowing disease progression.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    CWMT allows dose and compliance to be operationalised and monitored using process indicators, enabling more objective quantification of realised exposure and supporting adjustment of training progression.

    “Importantly, CWMT allows dose and compliance to be operationalised and monitored using process indicators, enabling more objective quantification of realised exposure and supporting adjustment of training progression.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [19] Fang, P. et al. Effects of computerized working memory training on neuroplasticity in healthy individuals: a combined neuroimaging and neurotransmitter study. NeuroImage 298, 120785 (2024). source

  • May combine two claims.

    Interpretationreasoning

    Results suggest partially generalisable components of training-related modulation, even as the magnitude of change remains larger for trained tasks.

    “Together, these results suggest partially generalisable components of training-related modulation, even as the magnitude of change remains larger for trained tasks.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Across 45 eligible studies, CWMT yields moderate improvements in cognitive task performance alongside reproducible modulation of task-related activation.

    “Across 45 eligible studies, this meta-analysis indicates that CWMT yields moderate improvements in cognitive task performance alongside reproducible modulation of task-related activation.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Dose and compliance are actionable determinants of effect magnitude that can be quantified objectively via completion and performance trajectories.

    “By integrating behavioural and neuroimaging evidence, we identify dose and compliance as actionable determinants of effect magnitude that can be quantified objectively via completion and performance trajectories, supporting personalised progression and implementation monitoring.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductown data

    Jackknife sensitivity analyses supported the stability of the SDM findings, with the main activation results retained across all leave-one-out iterations.

    “As shown in Supplementary Table S1, jackknife sensitivity analyses supported the stability of the SDM findings, with the main activation results retained across all leave-one-out iterations.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductown data

    Multivariate meta-analysis results were stable to alternative assumptions about within-study correlations among effect sizes (r = 0.6-1.0), with CWMT-related improvements in cognitive performance remaining consistent across all tested values.

    “In the multivariate meta-analysis, results were stable to alternative assumptions about within-study correlations among effect sizes (r = 0.6–1.0), with CWMT-related improvements in cognitive performance remaining consistent across all tested values (Supplementary Table S2).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationcitation

    The pattern of greater attenuation for trained tasks is compatible with closer alignment of trained probes with training content, which may preferentially tune task-specific representations, whereas untrained tasks likely place broader demands on control and strategy implementation.

    “This pattern is compatible with the closer alignment of trained probes with the training content, which may preferentially tune task-specific representations, whereas untrained tasks likely place broader demands on control and strategy implementation16,17.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [16] von Bastian, C. C. & Oberauer, K. Effects and mechanisms of working memory training: a review. Psychol. Res. 78, 803–820 (2014). source
    • cites [17] Klingberg, T. Training and plasticity of working memory. Trends Cogn. Sci. 14, 317–324 (2010). source

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    The present findings have clear translational relevance for CWMT as a scalable, digitally delivered training format.

    “Despite these limitations, the present findings have clear translational relevance for CWMT as a scalable, digitally delivered training format.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Candidate mechanisms for activation attenuation are not mutually exclusive and cannot be disentangled given current coordinate-based evidence.

    “These candidate mechanisms are not mutually exclusive and cannot be disentangled given the current coordinate-based evidence.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductown data

    Robustness was assessed broadly through complementary checks, including sensitivity analyses, alternative model specifications, evaluations of subgroup stability, a non-independence check, and assessment of small-study effects.

    “Robustness was assessed broadly through complementary checks, including sensitivity analyses, alternative model specifications, evaluations of subgroup stability, a non-independence check, and assessment of small-study effects (potential publication bias).”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    When studies reported multiple tasks, distinct populations, or more than one training arm, each condition was extracted and coded separately to avoid conflating effects.

    “When studies reported multiple tasks, distinct populations, or more than one training arm, we extracted and coded each condition separately to avoid conflating effects”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductreasoning

    Residual false-positive risk inherent to coordinate-based whole-brain meta-analyses cannot be fully excluded even under conservative thresholds; therefore, robustness analyses were conducted to examine reproducibility.

    “Nevertheless, residual false-positive risk inherent to coordinate-based whole-brain meta-analyses cannot be fully excluded even under conservative thresholds; therefore, we conducted robustness analyses (see the Robustness Analyses section) to examine the reproducibility of the main findings.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    The stability of the SDM results was examined using jackknife sensitivity analyses, repeating the meta-analysis while sequentially excluding each study.

    “The stability of the SDM results was examined using jackknife sensitivity analyses, repeating the meta-analysis while sequentially excluding each study.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductcitation

    Effect sizes were computed as Hedges' g standardised mean differences to correct for small-sample bias and improve comparability across tasks.

    “Effect sizes were computed as Hedges’ g standardised mean differences (SMDs) to correct for small-sample bias and improve comparability across tasks88.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [88] Hedges, L. V. Distribution theory for glass’s estimator of effect size and related estimators. J. Educ. Stat. 6, 107–128 (1981). source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    To address potential non-independence, the synthesis was re-run after excluding potentially overlapping studies with shared first authors, retaining only one study per first-author set.

    “To address potential non-independence, we re-ran the synthesis after excluding potentially overlapping studies with shared first authors31,48,54,60,67, retaining only one study per first-author set.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [31] Ando, T., Momose, K., Tanaka, K. & Saito, K. Effects of Task Difficulty and Training of Visuospatial Working Memory Task on Brain Activity. in 13th International Conference on Biomedical Engineering (eds Lim, C. T. & Goh, J. C. H.) 657–660 (Springer, 2009). source
    • cites [48] Jolles, D. D., Grol, M. J., Van Buchem, M. A., Rombouts, S. A. R. B. & Crone, E. A. Practice effects in the brain: changes in cerebral activation after working memory practice depend on task demands. NeuroImage 52, 658–668 (2010). source
    • cites [54] Li, X. et al. The remediation effects of working memory training in schizophrenia patients with prominent negative symptoms. Cogn. Neuropsychiatry 24, 434–453 (2019). source

  • Classified as probably not a substantial claim.

    Study design and conductcitation

    To reduce false positives, only clusters that were identified in the overall analysis and additionally exhibited the expected significant between-group difference were reported.

    “To reduce false positives, we report only clusters that were identified in the overall analysis and additionally exhibited the expected significant between-group difference94.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [94] Li, G., Xia, H., Teng, G. & Chen, A. The neural correlates of physical exercise-induced general cognitive gains: a systematic review and meta-analysis of functional magnetic resonance imaging studies. Neurosci. Biobehav. Rev. 169, 106008 (2025). source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    Given heterogeneity in n-back paradigms across included studies, the robustness of subgroup findings was assessed across n-back load and stimulus type.

    “Given heterogeneity in n-back paradigms across included studies, we assessed the robustness of subgroup findings across (i) n-back load (high-load vs. low-load) and (ii) stimulus type (symbolic-only vs. non-symbolic/mixed; see Supplementary Table S9 for operational definitions).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The review protocol was preregistered with PROSPERO (CRD42024599205).

    “The review protocol was preregistered with PROSPERO (CRD42024599205).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Small-study effects (potential publication bias) were assessed using Egger's test for both behavioural and neuroimaging outcomes.

    “Finally, small-study effects (potential publication bias) were assessed using Egger’s test for both behavioural and neuroimaging outcomes.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductown data

    Two reviewers independently conducted all screening steps, resolving discrepancies through discussion or consultation with a senior author.

    “To ensure reliability, two reviewers (GL and YL) independently conducted all screening steps, resolving discrepancies through discussion or, if necessary, consultation with a senior author (AC).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    Two authors independently screened records for inclusion and extracted data using a standardised form.

    “Two authors (GL and YL) independently screened records for inclusion and extracted data using a standardised form.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    Study identification and selection were carried out in accordance with the PRISMA guidelines.

    “Study identification and selection were carried out in accordance with the PRISMA guidelines, with comprehensive details reported in Supplementary Table S7.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductreasoning

    To avoid unit-of-analysis errors, outcomes were not pre-aggregated within studies; multiple outcomes assessed at the same time point were retained as distinct endpoints within the same sample, and their statistical dependence was accommodated through sampling variance-covariance matrix and the three-level multivariate structure.

    “To avoid unit-of-analysis errors, outcomes were not pre-aggregated within studies: multiple outcomes assessed at the same time point were retained as distinct endpoints within the same sample (Level 1), and their statistical dependence was accommodated through sampling variance–covariance matrix and the three-level multivariate structure.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    For the multivariate meta-analysis, sensitivity to modelling assumptions was evaluated by re-estimating the models under alternative within-study correlations (r = 0.6, 0.7, 0.8, 0.9 and 1.0).

    “For the multivariate meta-analysis, we additionally evaluated sensitivity to modelling assumptions by re-estimating the models under alternative within-study correlations among effect sizes (r = 0.6, 0.7, 0.8, 0.9 and 1.0).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Multivariate meta-analysis enables the inclusion of multiple effect sizes reported within the same study and yields more robust pooled estimates by explicitly accounting for within-study dependence.

    “Compared with univariate models, multivariate meta-analysis enables the inclusion of multiple effect sizes reported within the same study and yields more robust pooled estimates by explicitly accounting for within-study dependence (e.g., multiple cognitive outcomes derived from the same participants)21.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [21] Nam, I. S., Mengersen, K. & Garthwaite, P. Multivariate meta-analysis. Stat. Med. 22, 2309–2333 (2003). source
48 of 48

  • The paper relies on this without support.

    Prior reviews have not consistently isolated CWMT as a distinct digital training modality for systematic evaluation and meta-analytic quantification.

    “Critically, prior reviews have not consistently isolated CWMT as a distinct digital training modality for systematic evaluation and meta-analytic quantification.”

  • The paper relies on this without support.

    There is an urgent need for scalable, low-risk, and broadly applicable non-pharmacological interventions to prevent or delay cognitive function decline.

    “Given the limited effectiveness of pharmacological treatments, there is an urgent need for scalable, low-risk, and broadly applicable non-pharmacological interventions to prevent or delay cognitive function decline.”

  • The paper relies on this without support.

    Non-digital WMT formats rely on intermittent assessment and self-report, which may be less sensitive to day-to-day variation in engagement.

    “By contrast, non-digital formats often rely on intermittent assessment and self-report, which may be less sensitive to day-to-day variation in engagement.”
3 of 3

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Trained-task t statistic differs from Table 2

    Minor

    The comparison-analysis text gives t = 7.132 for trained tasks, while Table 2 gives t = 6.431 for that subgroup. The value 7.132 appears in Table 2's overall-analysis row.

  • Reported maximum age excludes a study in Table 1

    Minor

    The reported age range ends at 69.6 years, but Table 1 lists an age of 77.78 years for Salmi.

  • Reported minimum age excludes a study in Table 1

    Minor

    The reported age range begins at 11.9 years, but Table 1 lists an age of 7.8 years for Kelly.

3 of 3

Cites

Flagged first, then by citations

Cited by

Sorted by citations

Authors
Li, G., Liu, Y., & Chen, A.
Affiliations
  • Shanghai Jiao Tong University (CN)
  • Jishou University (CN)
Corresponding author
Chen, Antao
ORCID
Venue
npj Digital Medicine
Publisher
Springer Science and Business Media LLC
Published
2026-03-12
Volume
9
Issue
1
ISSN
2398-6352 (ISSN-L)
DOI
10.1038/s41746-026-02478-9
OpenAlex
W7135038488
Cited by
2
References
95
Citations by year
2026: 2
FWCI
5.04
Type
journal article
Language
en
Open access
Licence
Locations
  • npj Digital Medicine (publishedVersion, cc-by-nc-nd)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • PubMed Central (submittedVersion, cc-by-nc-nd)
Funders
the National Natural Science Foundation of China (32171040)
Topics
  • Social Sciences › Psychology › Experimental and Cognitive Psychology › Cognitive Abilities and Testing
  • Social Sciences › Psychology › Experimental and Cognitive Psychology › Cognitive Functions and Memory
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Neural and Behavioral Psychology Studies
Keywords
working memory, neural correlates, transfer of training, training compliance, training dose, angular gyrus, superior frontal gyrus, inferior parietal lobule

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-08.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-08.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review