Dual Transcranial Direct Current Stimulation Modulates Hierarchical Functional Network Organization in Post‐Stroke Cognitive Impairment: A Randomized Controlled Trial

Clinical Question (PICO)

In right-handed adults aged 50–80 with first-ever stroke and cognitive impairment persisting for more than 3 months, does bilateral dual transcranial direct-current stimulation (tDCS) over the dorsolateral prefrontal cortex, added to conventional rehabilitation, improve cognition and activities of daily living more than sham stimulation plus the same usual treatment after 4 weeks? The primary outcome was change in Montreal Cognitive Assessment (MoCA) score; no clinical subgroup treatment effects were reported.

Bottom Line

Among 74 participants who completed clinical assessments, MoCA improved by 5.74 points (SD 2.76) with active dual-tDCS versus 2.69 (SD 2.69) with sham, with a significant group-by-time interaction. This is a promising short-term signal from a small, single-center trial—not evidence of durable benefit or a basis for routine use; the fMRI analyses involved only 36 participants, and there was no post-treatment follow-up.

Design

Trial type: Prospective, single-center, randomized, double-blind, sham-controlled trial; registered in the Chinese Clinical Trial Registry as ChiCTR-2200055500 on January 11, 2022.
N: 76 randomized (38 active, 38 sham); 74 completed the intervention and clinical assessments (38 active, 36 sham). The methods section says the investigators planned to recruit 70 (35 per arm), whereas Figure 1 and the Results report 76 randomized.
- Active dual-tDCS: 38 randomized; 38 in the clinical analysis.
- Sham: 38 randomized; 36 in the clinical analysis; Figure 1 attributes the two missing clinical follow-ups to discharge midway.
Randomization: 1:1 computer-generated sequence managed by an independent statistician; allocation concealed with sealed, opaque envelopes. Patients, neuropsychological assessors, and imaging analysts were blinded; a post-intervention questionnaire assessed blinding.
Setting: Department of Rehabilitation Medicine, The Affiliated Brain Hospital of Nanjing Medical University, China; one center.
Enrollment: April 2022–December 2024.
Mean follow-up: None beyond the end-of-treatment assessment. Outcomes were measured within 3 days before and after the 4-week intervention.
Analysis: Repeated-measures ANOVA for group-by-time interactions, followed by post-hoc t-tests. The paper reports completer denominators for clinical outcomes; it does not describe the analysis as intention-to-treat.
Primary outcome: Global cognition measured by MoCA, assessed before and immediately after the 4-week intervention.

Population

Inclusion Criteria

Age 50–80 years; first-ever stroke confirmed by CT or MRI.
Cognitive impairment persisting more than 3 months after stroke.
Right-handed.
Education-specific MMSE threshold: below 17 for people with no formal literacy, below 20 for primary education, or below 24 for junior-high education or higher.

Exclusion Criteria

Pre-existing cognitive impairment or dementia, another neurodegenerative disease, or a family history of dementia.
Hachinski Ischemic Score below 4 or Fazekas score of 2 or higher.
Major systemic illness, metallic implants, severe sensory or language deficits, epilepsy, or prior tDCS; pregnancy or lactation.

Baseline Characteristics (clinical analysis)

Mean age was 61.42 years (SD 13.45) in the active group and 61.08 (SD 11.66) in the sham group; 26/38 and 28/36, respectively, were men.
Stroke was ischemic in 23/38 active and 22/36 sham participants; baseline median NIHSS was 3 in both arms.
Baseline mean MoCA was 13.24 (SD 4.47) with active allocation and 14.47 (SD 4.64) with sham.

Interventions

Active dual-tDCS: 2.0 mA for 20 minutes per session, 5 days per week for 4 weeks (20 sessions), using saline-soaked sponge electrodes; anode at left F3 and cathode at right F4 over bilateral DLPFC. Both groups also received conventional therapist-led rehabilitation, 30 minutes per session, 5 days per week for 4 weeks, and stable standard medication.
Sham: Same electrode placement and device appearance; current was ramped down after the first 30 seconds. The sham group received the same rehabilitation and medication regimen.

Outcomes

Primary Outcome — MoCA change after 4 weeks:

The group-by-time interaction was significant (F = 4.677, p = 0.034, partial η² = 0.061). Mean change was +5.74 points (SD 2.76) with active tDCS versus +2.69 (SD 2.69) with sham (post-hoc t = 4.799, p < 0.001). The paper does not report 95% confidence intervals for these mean changes. No prespecified clinical subgroup effect estimates were reported.

Secondary Outcomes:

MMSE: Mean change +5.24 (SD 3.11) versus +2.83 (SD 3.05); t = 3.353, p = 0.001; group-by-time F = 11.244, p = 0.001.
Stroop Part A completion time: Change −15.39 seconds (SD 7.13) versus −8.94 (SD 7.90); t = −3.692, p < 0.001.
Stroop Part B completion time: Change −18.37 seconds (SD 9.61) versus −12.67 (SD 8.41); t = −2.710, p = 0.008.
Wechsler Memory Scale: Mean change +26.84 (SD 13.50) versus +15.28 (SD 15.74); t = 3.399, p = 0.001.
Barthel Index: Mean change +22.11 (SD 13.44) versus +13.06 (SD 9.58); t = 3.319, p = 0.001.
No significant group-by-time interactions were reported for Stroop Part C or Trail Making Test (p > 0.05).
The imaging analyses were exploratory mechanistic outcomes, not clinical subgroup analyses. Post-intervention fMRI was available for 41 participants; after motion exclusions, 36 (18 per arm) remained for regional homogeneity and connectivity analyses. ReHo increased in the right parahippocampal/middle temporal regions and left opercular inferior frontal gyrus, and decreased in the left postcentral and orbital inferior frontal regions; FC decreased in the right MTG–left superior frontal and left IFGorb–cerebellar lobule VI connections. Graph-theory analyses included 30 participants (15 per arm), after six additional exclusions for failure to meet the small-worldness criterion; group-by-time interactions were reported for small-worldness (F = 14.367, p = 0.001), normalized clustering (F = 10.434, p = 0.003), path length (F = 7.552, p = 0.010), and global efficiency (F = 6.356, p = 0.018). Changes in MoCA correlated with changes in ReHo, FC, small-worldness, global efficiency, and path length after the paper's stated adjustments/FDR correction. These associations do not establish mediation or a validated monitoring biomarker.

Adverse Events / Safety:

Three participants reported mild tingling, erythema, or itching during initial sessions; symptoms resolved immediately without intervention. No serious adverse events were reported. Figure 1 shows two sham participants lost to clinical follow-up after discharge midway, although the Safety section says there were no withdrawals.

Figures

Figure 1. Participant enrollment and analysis flow. — *Figure 1.* The enrollment process for this study. FC: functional connectivity; fMRI: functional magnetic resonance imaging; ReHo: regional homogeneity; tDCS: transcranial direct-current stimulation.

Figure 2. Clinical scale changes after dual-tDCS. — *Figure 2.* Group × time interaction effects of rating scales following dual-tDCS. (A) MoCA and MMSE scores; (B) completion time for Stroop Parts A, B, and C; (C) Trail Making Test completion time; (D) Wechsler Memory Scale and Barthel Index. *p < 0.05.

Figure 3. Regional homogeneity and functional-connectivity changes. — *Figure 3.* Significant group × time interaction effects on ReHo and FC following dual-tDCS. ReHo interactions were located in the right parahippocampal gyrus and middle temporal gyrus, and the left inferior frontal gyrus, postcentral gyrus, and orbital inferior frontal gyrus. FC interactions involved right MTG–left SFG and left IFGorb–cerebellar lobule VI connections. Imaging significance used voxel-level p < 0.001 and cluster-level p < 0.05, family-wise-error corrected.

Figure 4. Distributions of global network metrics across sparsity levels. — *Figure 4.* Distribution of global-property metrics for each group under different sparsity levels, including clustering, path length, efficiency, and small-worldness.

Figure 5. Group-by-time interactions in graph-theory measures. — *Figure 5.* Group × time interaction effects of graph-theory analysis following dual-tDCS. The reported interactions include global efficiency, normalized clustering coefficient, characteristic path length, and small-worldness. *p < 0.05; **p < 0.001.

Figure 6. Correlations between clinical changes and fMRI measures. — *Figure 6.* Heatmap of partial correlations between changes in clinical/cognitive scores and fMRI measurements. Analyses adjusted for age, sex, education, stroke type, disease duration, lesion volume, and baseline NIHSS/Fazekas scores. The color bar indicates correlation direction and strength; r values are annotated in the cells. *p < 0.05; **p < 0.001 after false-discovery-rate correction.

The JATS-derived figure metadata identifies these six main-text figures and supplies the PMC CDN paths. The article is licensed under CC BY 4.0; attribution is retained with each figure.

Criticisms

Small, single-center completer sample: Clinical results are based on 74 participants from one Chinese center. The methods say 70 were to be recruited, but the flow diagram reports 76 randomized; two sham participants were not in the clinical analysis. The paper does not state that the reported clinical analysis was intention-to-treat.
Limited duration and generalizability: Assessment ended immediately after four weeks. Participants were right-handed, aged 50–80, had a first-ever stroke and impairment beyond 3 months, and patients with several comorbidities or more severe imaging/communication features were excluded. The result cannot be assumed to apply to acute stroke or all PSCI.
Outcome uncertainty: The primary signal is a short-term score change; the paper gives no 95% CI for the mean-change comparison and does not establish a clinically important threshold for the observed MoCA difference. Multiple secondary outcomes were also tested.
Imaging selection and interpretation: Only 36 participants contributed to the main post-treatment imaging analysis and 30 to graph theory. Correlations between imaging changes and clinical gains are associations, not proof that network changes caused recovery or a validated monitoring biomarker.

Funding

The study was funded by the Nanjing Municipal Special Fund Key Project for Health Science and Technology Development (ZKX22042) and the Nanjing Municipal Special Fund General Project for Health Science and Technology Development (YKK22137). The article declares no conflicts of interest. It does not report a funder role in the study design, analysis, or reporting.

The paper

Authors. Wang Y, Liu W, Yang W, Hu J, Cheng X, Yu H, Shi Y, Liu L, Rao J, Xu G.
Title. Dual Transcranial Direct Current Stimulation Modulates Hierarchical Functional Network Organization in Post‐Stroke Cognitive Impairment: A Randomized Controlled Trial.
Journal. CNS Neuroscience & Therapeutics.
Year. 2026.
DOI. 10.1002/cns.71111
PMCID. PMC13538461

Deep Dive — click to expand

What this is

This is a small, prospective, double-blind comparison of bilateral DLPFC dual-tDCS versus sham, each added to the same conventional rehabilitation and medication, in people with persistent cognitive impairment at least 3 months after a first stroke. The headline is a larger four-week MoCA gain with active stimulation; the study also reports short-term gains on several secondary cognitive and daily-function measures and exploratory fMRI correlates. It did not test acute stroke treatment, tDCS without rehabilitation, or long-term durability.

1. Shadow Audit

The paper's strongest bedside-relevant signal is not a network image: it is the between-arm difference in short-term MoCA change, +5.74 versus +2.69 points, alongside a Barthel Index change of +22.11 versus +13.06. Both groups improved, so the result is the incremental advantage of the active assignment over sham on top of ongoing therapy—not the total effect of tDCS alone. The reported endpoint was measured immediately after treatment, with no later follow-up. The imaging narrative is more fragile: only 36 participants had usable post-treatment fMRI and 30 entered graph theory. Correlations between those imaging changes and MoCA are compatible with a mechanism, but do not show that the imaging changes mediate the clinical improvement or can yet guide individual treatment.

2. Inversion Engine

The reported mean-change gap on MoCA is 3.05 points (5.74 minus 2.69). The practical interpretation would invert if a larger, well-conducted replication found little or no active-versus-sham difference, or if the apparent gap failed to translate into durable or patient-important function. This study does not supply a 95% CI or establish a minimal clinically important difference for its MoCA contrast, so the threshold for “clinically meaningful” cannot be settled from this paper alone. The sham arm's 2.69-point gain also shows why the within-arm active improvement cannot be attributed wholly to stimulation: time, conventional rehabilitation, medication, retesting, and other shared care remain plausible contributors.

3. Second-Order Catalyst

The first reasonable mover is a neurorehabilitation service already equipped to deliver and monitor noninvasive stimulation—not an acute-stroke thrombolysis or thrombectomy pathway. If another center wants to act now, the defensible first step is a registered replication or supervised pilot in patients resembling this cohort: first-ever stroke, persistent PSCI beyond 3 months, and the study's eligibility profile. A clinical protocol should preserve the same 20-session schedule and standard rehabilitation in both arms, measure baseline and post-course MoCA/ADL, check adverse effects, and include longer follow-up. The article is not sufficient to make dual-tDCS routine care tomorrow.

4. Asymmetric Leverage

The intervention is a bounded course—20 sessions over four weeks—added to rehabilitation that both groups already received. If the approximately 3-point MoCA mean-change advantage and the larger Barthel change are reproducible and durable, the potential leverage is meaningful because a cognitive rehabilitation adjunct might affect daily independence as well as test performance. But the denominator is only 74 clinical completers at one center, and the paper does not establish a cost, staffing burden, responder rate, or number needed to treat. The leverage is therefore a strong reason to replicate, not a proven service-wide payoff.

5. Paradigm Destroyer

This trial challenges the reflex that persistent post-stroke cognitive impairment beyond the early recovery period is necessarily beyond the reach of rehabilitation: the enrolled cohort had impairment more than 3 months after stroke, and both arms improved during a structured four-week program. It does not prove dual-tDCS is superior to unilateral tDCS, cognitive training, or other active treatments, because those comparisons were not made. The two-sentence protocol change for tomorrow is modest: continue evidence-based rehabilitation and screen/measure cognition rather than assume it cannot improve. Offer dual-tDCS only through a supervised research or carefully governed local pilot until larger multicenter trials confirm benefit and durability.

MVP — Minimum Viable Proof

The minimum result worth replicating is the prespecified primary outcome: among this selected PSCI population, active dual-tDCS plus conventional therapy produced a +5.74-point mean MoCA change versus +2.69 with sham plus the same therapy over four weeks (group-by-time p = 0.034; post-hoc p < 0.001). Replication should retain concealed allocation, blinded assessment, report confidence intervals and intention-to-treat results, and show whether the cognitive difference persists and maps to patient-important function.

Best Combination

The authors describe prior unilateral left-DLPFC tDCS findings in PSCI as inconsistent and position bilateral dual-tDCS as a way to test simultaneous hemispheric modulation. This trial adds a sham-controlled clinical signal to that literature and pairs cognitive measures with local synchrony, inter-regional connectivity, and global-network analyses. The clean synthesis is: treat the clinical result as promising, the fMRI findings as mechanistic hypotheses, and the next study as a direct multicenter replication with adequate follow-up—not as proof that dual placement is the best tDCS configuration.

Overvalue Warning

Do not turn immediate post-treatment improvements across several scales into a claim of lasting cognitive recovery; follow-up stopped at four weeks, and Stroop Part C and TMT did not show significant interactions.
Do not treat the small imaging-subset correlations as a validated biomarker or causal mechanism. Imaging availability and analysis samples were much smaller than the clinical cohort, and correlation is not mediation.

Educational use notice. The Time Is Brain Review is dedicated to medical education. Figures from published articles are embedded for teaching purposes, with attribution to the original publication. We do not claim copyright over any embedded figure; all rights remain with the original publisher and authors under the article's published license. If you are a rights holder and would like a figure removed, please contact the editorial team. See our editorial notice for the full policy.