Transcranial direct-current stimulation in subacute aphasia: A randomized controlled trial

Abstract

Background:

Transcranial direct current stimulation (tDCS) is a promising adjunct to therapy for chronic aphasia.

Methods:

This single-center, randomized, double-blind, sham-controlled efficacy trial tested the hypothesis that anodal tDCS augments language therapy in subacute aphasia. Secondarily, we compared the effect of tDCS on discourse measures and quality of life and compared the effects on naming to previous findings in chronic stroke. Right-handed English speakers with aphasia <3 months after left hemisphere ischemic stroke were included, unless they had prior neurological or psychiatric disease or injury or were taking certain medications (34 excluded; final sample= 58). Participants were randomized 1:1, controlling for age, aphasia type, and severity, to receive 20 minutes of tDCS (1 mA) or sham-tDCS in addition to 15 45-minute sessions of naming treatment (plus standard care). The primary outcome variable was change in naming accuracy of untrained pictures pre-treatment to one-week post-treatment.

Results:

Baseline characteristics were similar between the tDCS (N=30) and sham (N=28) groups: patients were 65 years old, 53% male, and two months from stroke onset on average. In intent-to-treat analysis, the adjusted mean change from baseline to 1-week post treatment in picture naming was 22.3 (95%CI 13.5, 31.2) for tDCS and 18.5 (9.6, 27.4) for sham and was not significantly different. Content and efficiency of picture description improved more with tDCS than sham. Groups did not differ in QoL improvement. No patients were withdrawn due to adverse events.

Conclusions:

tDCS did not improve recovery of picture naming but did improve recovery of discourse. Discourse skills are critical to participation. Future research should examine tDCS in a larger sample with richer functional outcomes.

Clinical Trial Registration Information:

https://www.clinicaltrials.gov/ct2/show/NCT02674490.

Graphical Abstract

Introduction

Difficulty naming is a common symptom of aphasia after left hemisphere (LH) stroke. Speech and language therapy is the mainstay treatment.^1,2 Therapy is beneficial for language recovery in stroke.^3,4 However, about 100 hours of speech and language therapy are needed to significantly improve functional communication.^5–7 Transcranial direct current stimulation (tDCS) is a promising adjunct to traditional language therapy.^8–11 tDCS is a safe, non-invasive, non-painful electrical stimulation of the brain, which modulates cortical excitability by applying weak electrical currents in the form of direct current brain polarization.¹² Anodal tDCS (A-tDCS) most often has been applied to LH language areas to increase cortical excitability (reduce activation threshold) in patients with chronic aphasia. However, neuroplasticity is greatest early after stroke, so it is plausible that tDCS is most effective in the acute-subacute period.

Three prior studies have evaluated tDCS plus language therapy in acute-subacute aphasia ^13–15 and only two^14,15 were sham-controlled. Jung et al.¹³ observed that aphasia improved following cathodal tDCS over Brodmann area 45 during language therapy. You et al.¹⁴ examined the effects of language therapy with stimulation to Brodmann area 22 with language therapy in 21 patients. Cathodal tDCS was associated with significantly more improvement in auditory comprehension but not overall aphasia severity relative to anodal tDCS or sham. Spielmann et al.¹⁵ found no differences in changes in naming by individuals with mild-moderate subacute aphasia when substituting traditional therapy for naming therapy combined with anodal tDCS over the left inferior frontal gyrus (N=26) or sham (N=32) for 2 weeks. No previous studies have directly compared effectiveness of tDCS using the same treatment at different periods of stroke recovery. Timing may be important for both the duration and mechanism of effects.⁸

The present trial, Stimulating Language in Subacute StrokE (SLISSE), was conducted to determine if tDCS augments computer-delivered language therapy for improving picture naming and discourse. A secondary aim was to compare the effect of tDCS plus language therapy in subacute post-stroke aphasia to the effect of the same language treatments for the same number of sessions (15) on the same outcome (improvement in picture naming) in chronic post-stroke aphasia, which was established in a previous study.¹⁶ We also aimed to determine whether A-tDCS with language therapy improved naming performance, discourse, or quality of life (QoL) more efficiently and for a longer duration than language therapy alone.

Materials and methods

Study design

SLISSE was a randomized, single-center, double-blind, sham-controlled, efficacy study of A-tDCS combined with language therapy for stroke-induced aphasia patients completed at Johns Hopkins Hospital and Johns Hopkins Bayview Medical Center in Baltimore, Maryland. All procedures were approved by the Johns Hopkins Medicine Institutional Review Board (IRB00089018). The study was registered with ClinicalTrials.gov (NCT02674490). Deidentified participant data will be made available upon request to the authors upon publication, subject to review by the Johns Hopkins University School of Medicine Institutional Review Board resulting in a formal data sharing agreement. Study protocol and related documents are available on via ClinicalTrials.gov and from the corresponding author. Research reporting followed CONSORT guidelines.

Patients

All consecutive patients on the stroke service who were right-handed, premorbidly competent adult speakers of English, <3 months of acute ischemic LH stroke were screened. They were screened for aphasia using the Western Aphasia Battery-Revised (WAB-R).¹⁷ Individuals with previous neurological or psychiatric disease, seizures, brain surgery, metal in the head, uncorrected visual or hearing loss, scalp sensitivity, medications that lower the seizure threshold, or N-methyl-D-aspartate (NMDA) antagonists were excluded. Individuals with asymptomatic lacunar infarcts were not excluded, although patients with previous symptomatic stroke were excluded. Details regarding screen failures are included in Table S1.

Randomization and masking

Randomization was completed by the Web Data Coordination Unit (WebDCU™), developed at Medical University of South Carolina, and patients were randomized 1:1 (A-tDCS: sham). The minimal sufficient balancing method was used to prevent serious imbalances from occurring in baseline age, aphasia type, and severity.¹⁸ Aphasia severity was defined using WAB-R Aphasia Quotient (AQ).¹⁷

The subjects, site investigators, and clinicians involved in this study did not know the treatment assignment. Select members of the statistical team at WebDCU were partially blinded, i.e., they knew the treatment group assignment as A or B, but not whether the patient received active tDCS or sham. A relay box was programmed and verified for A-tDCS or sham by a study team member who did not participate in treatment. To mask treatment type (A-tDCS vs. sham) for patients and speech-language pathologists administering the treatment session, the sham consists of 30 seconds of 1 mA stimulation which is faded to 0, mimicking the sensation of continuous (e.g., 20 minutes) 1 mA stimulation, which is generally perceived for only the initial 20 to 30 seconds. The study statisticians (JE, CC, AT) were unblinded once all data were entered, and the database was locked.

Procedures

After providing informed consent, patients completed a baseline evaluation, then received naming treatment. Patients were re-evaluated at one-, five-, and twenty-weeks after treatment.

Baseline evaluation

Baseline NIH Stroke Scale¹⁹ (NIHSS, a measure of overall stroke severity) and speech-language evaluation were performed. Descriptions of the Cookie Theft picture on the NIHSS were analyzed for Content Units (CU) and Syllables/CU.^20,21 CUs are concepts (words or phrases or their exact synonyms) mentioned by healthy controls when describing the picture.²² Each CU is counted once, and the sum provides a measure of narrative speech content in aphasia.²¹ Efficiency of discourse is measured by the total number of syllables (including nonwords) in the description/CU, such that lower syllable/CU indicates greater efficiency. Additional language tests included the WAB-R and color Philadelphia Naming Test²³ (PNT; untrained items) and a set of 80 trained items (Naming 80). We also administered the Stroke Impact Scale²⁴ (SIS), a disease-specific QoL questionnaire.

Patients were screened on the naming treatment task to verify that they were able to understand and complete the activity. The same Naming 80 items were used for all participants, regardless of baseline performance. Patients completed the Naming 80 and the PNT twice on consecutive days for each timepoint. The task was recorded and scored remotely by individuals blinded to the patient’s clinical profile. Scores on each administration were averaged to reduce variability. All participants with no contraindication were asked to complete fMRI. If they consented to fMRI, they were randomized to (1) have the site of stimulation based on greatest area of activation on fMRI in the temporal lobe during a naming task (as in Fridriksson¹⁶) or (2) have the site of stimulation based on the area of lesion (from clinical imaging or structural MRI). If randomized to lesion-based localization (or declined research MRI), a site of undamaged cortex was identified hierarchically by the study neurologist (AEH) based on the Hopkins JHU-MNI Brain Parcellation Map (cmrm.med.jhmi.edu), in which the IFG and PFG do not overlap: the left inferior frontal gyrus (L IFG) if no lesion, left superior temporal gyrus (L STG) if no lesion, or left prefrontal cortex (L PFC).

Treatment

Patients completed 15 45-minute sessions of computer-delivered naming treatment, identical to the treatment used in Fridriksson et al.¹⁶ with a speech-language pathologist over three to five weeks. A-tDCS (1 mA) or sham-tDCS was administered for the first 20-minutes of the computer-delivered treatment. The anode was placed at the stimulation site under a tight-fitting wrap. The reference cathode electrode was placed on the participant’s right orbito-frontal scalp.

During naming treatment, patients were shown an on-screen picture of a target (e.g., mop) for two seconds then a video of a female speaker saying a word that was either the name of the target (e.g., “mop”) or a semantic foil (“broom”), phonological foil (“top”), or unrelated foil (“bridge”). Patients were asked whether the picture and the audio/video presentation of the word matched, using large red “no match” and green “match” response buttons. Half of the pairs matched. The computer provided immediate visual feedback following a response in the form of a “smiley face” for correct answers and a “frowny face” for incorrect answers, and patients received a final accuracy score for the session.

Blood pressure and heart rate were measured before and after each session. Patients rated their discomfort using the Wong-Baker FACES Pain Rating Scale²⁵ after each session and received a brief neurological exam after every fifth session. Patients and clinicians providing therapy were asked to guess the patient’s treatment condition. Patients were not blocked from receiving any clinically-indicated additional therapy for aphasia or other deficits; they were asked to report the number and duration of all language therapy sessions.

Follow-up testing

One, five, and twenty weeks after the final treatment session, patients were re-evaluated on the PNT, Naming 80, picture description, and SIS.

Outcomes

The primary outcome was the change in number of correctly named items on the PNT pre-treatment versus one-week post-treatment. The null hypothesis was H₀: μ₁= μ₂, where μ₁ is the mean change in accuracy of naming untrained items between baseline and one-week post-treatment in the A-tDCS group and μ₂ is the mean change in accuracy of naming untrained items between baseline and one-week post-treatment in the sham group.

The following secondary outcome measures were pre-specified:

1. Change in accuracy of naming untrained pictures (PNT) pre-treatment to 5-weeks and 20-weeks post-treatment

2. Change in CU of picture description pre- to 1-, 5-, and 20-weeks post-treatment.

3. Change in efficiency (Syllables/CU) of picture description pre- to 1-, 5-, and 20-weeks post-treatment

• 6.Change in SIS pre-treatment to 1-, 5-, and 20-weeks post-treatment

Statistical analyses

The primary outcome was change in the number of correctly named items on the PNT (absolute change from pre-treatment and 1-week post-treatment). The primary analysis was conducted for the modified intent-to-treat sample defined as all randomized participants for whom study treatment was initiated (i.e. attended at least one treatment session), regardless of the treatment actually received. For participants who did not complete the 1-week post-treatment assessment, the post-treatment value was imputed using a multiple imputation approach assuming missing at random (MAR) and using a fully conditional specification (FCS) method adjusted for baseline PNT, WAB-AQ, aphasia type, and age.

The mean change from baseline in PNT correct was compared between treatment groups in a linear model adjusted for baseline WAB-AQ score, aphasia type (grouped as Anomia, Broca’s or other), and age. These model was adjusted for these factors because they were balanced in the randomization algorithm. As a sensitivity analysis, the primary analysis was repeated using the completers sample defined as the subset of randomized participants for whom study treatment was initiated and the PNT was collected at 1-week post-treatment.

The primary analysis and sensitivity analyses were tested at two-sided alpha of 0.05. All adverse experiences were summarized in terms of frequency, severity and relatedness to the study treatment using the MedDRA code. All subjects who received tDCS were included in the safety analysis. Cumulative incidences of adverse events were compared between the two treatment groups using Fisher’s exact test at the two-sided alpha level of 0.05.

Sample size was calculated based on the one published sham-controlled study of tDCS combined with speech and language therapy in subacute stroke,¹⁴ for which the outcome variable was change in WAB-R AQ, a composite measure of language after stroke. Based on this study, a total sample size of 40 would yield 89% power to detect a difference in means of 23 (the difference between a A-tDCS mean change in accuracy, μ₁, of 33 and a sham mean change in accuracy, μ₂, of 10) assuming that the standard deviation of change for both groups is 22.2 using a two-group t-test with a two-sided alpha of 0.05.

Results

A total of 92 patients were screened, and 58 patients (63%) were randomized between September 16, 2016 and October 4, 2021 (final follow-up evaluation March 29, 2022). Thirty patients were randomized to receive active tDCS, and 28 were randomized to receive sham tDCS. Twenty-six and 25 patients in active and sham tDCS, respectively, received at least one treatment session. Groups attended similar numbers of sessions (tDCS & sham: Median[IQR] =15[0]) of similar lengths (tDCS: Mean(SD)=60.0(7.5); sham: 61.1(8.8), p=0.63). Overall, 46 subjects (79%) completed 15 treatment sessions, four (7%) were lost to follow-up, two (3%) withdrew consent, and one (1.7%) subject died. See Figure 1 for a full accounting of subjects screened, enrolled, treated, and included in the analysis. Recruitment was stopped after the estimated target number of full datasets was reached. Baseline characteristics were similar between the two treatment groups (Table 1). The blind was maintained throughout the treatment period. Fifteen patients (29.4%) who had at least one treatment session correctly guessed whether they were in the tDCS or sham group. Clinicians correctly guessed the treatment group for 42.2% of patients.

Fig. 1.

CONSORT Subject Flow diagram

The primary analysis completers sample included of all of those for whom the primary outcome measure was collected (PNT one week after treatment). The modified Intent to Treat sample included all of those who participated in at least one treatment session.

† Patient 1 was withdrawn because aphasia was too severe. Patient 2 was withdrawn prior to starting therapy because patient recovered.

‡ Patient was withdrawn to be admitted to subacute rehab facility.

Early terminations in the sham tDCS group were as follows: 1) patient had a right middle cerebral artery stroke after 11 treatment sessions; 2) patient was not able to understand the treatment task, which was discontinued after a single session; 3) patient declined remote follow-up sessions offered at the start of the Covid-19 pandemic.

Table 1.

Baseline characteristics by treatment group

Baseline Characteristics, Mean (SD); Median (IQR)	Active-tDCS (N= 26)	Sham-tDCS (N=25)	p
Mean age (±SD)	65.1±12.2	63.8±15.0	0.73
Male sex, frequency (%)	16(61.5)	11(44.0)	0.27
Race, frequency (%)
White	19(73.1)	13(52.0)	0.34
Black	6(23.1)	10(40.0)
Other/unknown	1(3.8)	2(8.0)
Hispanic, frequency (%)	2(7.7)	0	0.49
Median years of education (IQR)	16(6.0)	14(5.0)	0.40
Mean stroke onset to enrollment (days; ±SD)	54.6 ± 30.7	65.4 ±33.7	0.24
Aphasia Type, frequency (%)
Broca’s	3(11.5)	6(24.0)	0.50
Anomic	9(34.6)	9(36.0)
Other	14(53.8)	10(40.0)
Median WAB-R Aphasia Quotient (IQR)	68.9(39.1)	59.9(48.8)	0.48
Current use (at randomization), frequency (%)
Selective Serotonin Reuptake Inhibitors (SSRIs) Antidepressants	6(23.1)	7(28.0)	0.76
Mixed /Atypical SSRIs	0	0
Non-SSRIs Antidepressants	0	0
Anxiolytics/Antipsychotics	1(3.8)	0	1.00
Anticholinergic Agents	0	0
Dementia medications	0	0
Past medical history, frequency (%)
Chronic Pain	7(26.9)	4(16.0)	0.50
Median average # headaches/month (IQR)	0.0(2.0)	0.0(1.0)	0.57
Seizures	1(3.8)	0	1.00
Surgery	17(65.4)	18(72.0)	0.76
Allergies (yes/no)	5(19.2)	14(56.0)	0.01
Depression	3(11.5)	1(4.0)	0.61
Diabetes	5(19.2)	6(24.0)	0.74
Median Philadelphia Naming Test correct (IQR)	97.8(86.5)	56.5(116.5)	0.36
Median Naming 80 (IQR)	23.5(38.0)	12.0(41.5)	0.36
Median Patient Health Questionnaire-9 (IQR)	5.5(7.0)b	6.0(8.0)d	0.72
Median Stroke Impact Scale-16 (IQR)	78.5(6.0)b	70.5(21.0)c	0.04
Median Cookie Theft Picture Content Units (CU; IQR)	8.0(9.0)a	6.0(9.0)d	0.09
Median Cookie Theft Picture Syllables (IQR)	105.0(75.0)a	62.0(46.0)d	0.04
Median Cookie Theft Picture Syllable/CU (IQR)	19.4(17.2)a	10.3(10.6)d	0.14
Current Occupation, frequency (%)
Professional	7(13.73)	5(9.80)
Self-employed	2(3.92)	1(1.96)
Clerical	0	1(1.96)
Salesperson	1(1.96)	0
Agriculture, forestry, and fishery	2(3.92)	0
Skilled worker	1(1.96)	2(3.92)
Service worker	1(1.96)	1(1.96)
Retired	8(15.69)	10(19.61)
Not employed/ on disability	2(3.92)	2(3.92)
Other	2(3.92)	3(5.88)

^aMissing data for 1 subject.

^bMissing data for 2 subjects.

^cMissing data for 3 subjects.

^dMissing data for 10 subjects (could not calculate syllables/CU when CU=0).

Primary analysis: Does tDCS improve picture naming in response to language therapy?

The primary analysis was conducted for the modified intent-to-treat sample. The adjusted mean change from baseline to one-week post treatment in PNT correct was 22.3 (95%CI 13.5, 31.2) for the active and 18.5 (95%CI 9.6, 27.4) for the sham group and was not statistically significantly different (p=0.54, two-tailed). The observed mean difference was 3.8 points (95%CI −8.42,16.05; see Figure 2). See Table S2 for individual performance data by condition and stimulation site.

Fig. 2.

Comparison of results from subacute vs chronic stroke

Two outliers were removed from the figure: 1 chronic participant who received A-tDCS experienced profoundly decreased performance of 129.5/175 and 1 participant in the present study who received sham tDCS experienced profoundly increased performance of 108.5 items.

Secondary Analyses

How does the application of tDCS in the subacute phase compare to that in the chronic phase?

The prior study of tDCS to augment naming treatment in chronic aphasia examined both the naming of untrained (PNT) and trained items (Naming 80). In a parallel analysis of results of the present study, the adjusted mean change from baseline to 1-week post treatment in untrained PNT+80 trained items correct was 37.7 (95%CI 25.3, 50.0) for the active and 31.6 (95%CI 19.0, 44.1) for the sham group. This difference was not significant (difference of 6.08; 95%CI −10.95, 23.12; p=0.48, two-tailed). Indeed, the magnitude of difference between tDCS and sham groups was nearly identical to the mean difference observed in individuals with chronic aphasia in which significant differences were observed of 5.7 (95%CI −0.9, 12.3),¹⁶ using the same treatments and assessments (Fig. 2). This discrepancy is attributable to the fact that there was greater variability in change in this trial of subacute aphasia. As a sensitivity analysis, the primary analysis was repeated using the completers sample, and the results were highly consistent. See Table 2.

Table 2.

Primary Analysis: Change in Naming at Immediate Post Testing

	Active-tDCS				Sham-tDCS
	N	Adjusted Mean	Lower 95%CI	Upper 95%CI	N	Adjusted Mean	Lower 95%CI	Upper 95%CI	p-value
Modified intent-to treat sample
PNT+Naming 80	26	37.7	25.3	50.0	25	31.6	19.0	44.1	0.48
PNT	26	22.3	13.5	31.2	25	18.5	9.6	27.4	0.54
Naming 80	26	15.3	10.6	20.1	25	13.1	8.2	18.0	0.50
Completers sample
PNT+Naming 80	25	37.9	24.5	51.2	21	30.3	16.6	44.1	0.42
PNT	25	22.3	12.8	31.7	21	18.0	8.3	27.8	0.52
Naming 80	25	15.6	10.5	20.7	21	12.3	7.1	17.6	0.36

The primary outcome is absolute change from pre-treatment to 1-week post-testing on the PNT.

The modified Intent-to-treat sample is defined as all randomized participants who began treatment. Missing data were imputed using multiple imputation. The primary analysis was adjusted for baseline aphasia type, severity, and age. The Completers sample is the subset of randomized participants for whom the PNT was collected at 1-week post-treatment. P-values are two-tailed.

Are there improvements to other aspects of language and QoL & are these preserved over time?

Table 3 shows the adjusted mean change from baseline in the pre-specified secondary outcomes at each follow-up. There were statistically significant differences in the following secondary outcomes: change in content of picture description pre-treatment to 5 weeks post-treatment (CUs); and change in efficiency of picture description pre-treatment to 5 weeks and 20 weeks post-treatment (syllables/CUs).

Table 3.

Change in Secondary Outcomes

	Active-tDCS				Sham-tDCS
	N	Adjusted Mean	Lower 95%CI	Upper 95%CI	N	Adjusted Mean	Lower 95%CI	Upper 95%CI	p-value
Change from baseline to 1-week post-therapy
Cookie Theft picture description
CU	25	4.10	1.93	6.27	17	1.66	−0.89	4.21	0.15
Syllables/CU	25	−12.23	−19.99	−4.46	12	−1.29	−11.42	8.83	0.08
SIS	22	2.56	−0.72	5.83	20	3.85	0.50	7.20	0.58
Change from baseline to 5-weeks post-therapy
Cookie Theft picture description
CU	24	5.42	3.22	7.62	17	1.42	−1.14	3.97	0.02
Syllables/CU	24	−12.64	−20.46	−4.82	13	2.47	−7.48	12/43	0.02
SIS	23	2.08	−1.17	5.33	18	4.18	0.78	7.58	0.37
PNT	25	29.95	18.84	41.07	20	20.82	9.36	32.27	0.25
Change from baseline to 20-weeks post-therapy
Cookie Theft picture description
CU	24	5.15	2.94	7.35	17	3.28	0.72	5.85	0.27
Syllables/CU	24	−14.75	−22.58	−6.91	12	6.53	−3.63	16.69	0.001
SIS	23	2.60	−0.65	5.86	18	4.28	0.88	7.69	0.47
PNT	24	34.04	22.90	45.18	21	25.55	14.13	36.97	0.28

P-values are two-tailed. CU=Content Units (CU). Syllables/CU captures spoken efficiency, with fewer syllables leading to more densely informative communication. PNT=Philadelphia Naming Test. SIS=Stroke Impact Scale.

The differences between tDCS and sham groups in improvement in content (CU) of picture description at 5 week follow-up [5.42 (3.22, 7.62) vs. 1.42 (−1.14, 3.97); p=0.0196] indicated a medium effect size (Cohen’s d=0.74).

There were significantly greater gains in efficiency of discourse in the tDCS compared to the sham group at both long-term follow-up points. Fewer syllables/CU reflects fewer perseverations, circumlocutions, and interjections (e.g., “um”) and greater communicative efficiency. The group differences at 5-week follow-up [−12.64 (−20.46, −4.82) vs. 2.47 (−7.48, 12.43); p=0.0164], and at 20-week follow-up [−14.75 (−22.58, −6.91) vs. 6.53 (−3.63, 16.69); p=0.0011) indicated large effect sizes (d=0.80 and 1.13, respectively).

There were no significant differences between groups in the following secondary outcome measures: change in accuracy of naming untrained pictures (PNT) pre-treatment to 5-weeks or 20-weeks post-treatment; or change in SIS pre-treatment to any follow-up period.

Safety

There were four serious adverse events (SAEs); two in active and two in sham. All SAEs were thought to be unlikely related or unrelated, as recurrent strokes constitute approximately 1 in 5 of all strokes.²⁶ The most common AE was mild irritation at the application site in three subjects (12%) (Table S3).

Discussion

This study examined whether A-tDCS coupled with computer-delivered language therapy improved naming performance of participants with acute-subacute phase of post-stroke aphasia more efficiently and for greater duration than language therapy alone. There was no significant difference between baseline and one-week post-treatment naming performance. While the magnitude of difference between tDCS and sham groups in naming improvement (6.08 more pictures named correctly) was almost identical to the magnitude of difference previously observed in chronic aphasia (5.7 more pictures named correctly) on the same assessment, the variability of improvement was greater in subacute aphasia (95%CI −10.95, 23.12) than in chronic aphasia (95%CI −0.9, 12.3). The greater variability accounts for the discrepancy between our findings and those in Fridriksson et al. 2018. Future trials of tDCS in subacute stroke may need to enroll more participants than trials in chronic aphasia to account for the larger variability in improvement early after stroke.

In secondary analyses, discourse content and efficiency improved to a significantly greater degree in individuals who received tDCS when measured at 5-weeks post-treatment. Differences in efficiency between the two groups were retained when measured 20-weeks post-treatment. These differences showed a medium or large effect size. This is of particular interest because, while the treatment targeted lexical semantics, only individuals who received tDCS had improvements that generalized to discourse. There were no differences between groups in change in QoL at any follow-up point. One double-blind sham-controlled trial published since the start of this one also failed to find a significant impact of A-tDCS over the L IFG on naming in subacute stroke;¹⁵ however, that study did not measure change in discourse. The only other prior sham-controlled study of A-tDCS in subacute stroke was notably underpowered. In 4/7 patients in the A-tDCS group, the stimulation site was over necrotic tissue. One important strength of the present design was that it deliberately decreased the probability of stimulating areas of lesion.

A-tDCS in subacute stroke remains controversial. While plasticity-dependent recovery may be greatest in this phase,^27–29 there are concerns that overstimulation could be detrimental to recovery long-term.³⁰ Prior reviews have noted minimal safety concerns within the windows of investigation.^9,31,32 This trial found no evidence of detrimental effect of tDCS over a relatively extended window (20 weeks). While there is insufficient evidence regarding long-term outcome of tDCS in the subacute versus chronic phases,³⁰ this trial was the first to compare outcomes of tDCS in subacute versus chronic aphasia directly using the same treatment and primary outcome measures.

These findings are consistent with those from recent systematic reviews demonstrating that tDCS is beneficial only in certain instances for improving language when combined with language therapy.^9,11,33 Recently, the NORTHSTAR trial³⁴ observed that repetitive transcranial magnetic stimulation of the right pars triangularis improved naming in a similarly-sized sampling of patients who received individualized language therapy (i.e., therapy that was not held constant) to treat subacute aphasia, while cathodal tDCS of the same region and sham did not. This effect was driven by individuals with infarcts sparing Broca’s area; those with infarcts to Broca’s area tended to improve more with sham. The magnitude of improvement observed in the subacute phase was larger than that observed in the chronic phase.³² Our study examined a single therapy approach. There are many well-established aphasia interventions.^35–37 Stimulation location, therapy approach, and dosage vary across trials, and most trials are very low to moderate quality. Thus, one contribution of this study is to add to the accumulating number of well-designed studies of this adjuvant. Another contribution is showing that tDCS may have a moderate to large effect size on improving discourse, but a smaller effect on picture naming, in subacute aphasia. Discourse skills are critical to life participation.

Limitations include small sample size and limitations of this design in answering questions related to the underlying mechanisms. Based on PNT performance, 436 patients in each group have 80% power to detect a difference in means of 3.8 assuming that the common standard deviation is 20 using a two-group t-test with a 5% two-sided significance level. Design limitations include that stimulation of left prefrontal or inferior frontal cortex may have resulted in increased cross-hemispheric inhibition of the right homologous cortex, and this effect was not measured. Therapy targeted the same set of nouns across all participants. It is possible that a design that varied the targeted nouns responsively may have better measured improvement among those who performed well on the targeted words at baseline. However, groups were matched in aphasia severity, and it is unlikely this design differentially impacted one treatment group over another. Also, the reason for the differential effects on naming versus discourse is unclear. Due to losses to follow-up, only 17 of the 25 patients in the sham condition and 24 of the 26 patients in the active condition completed follow-up testing including the Cookie Theft, leaving the groups more unbalanced than anticipated when considering these assessments and limiting their interpretation. Finally, although fMRI to guide place of stimulation was offered to participants, fewer than half agreed to additional imaging, resulting in inadequate power to evaluate the effect of using fMRI to guide stimulation location. While these are important limitations, they predominantly represent appropriate extensions to the current work.

Future research should examine subacute application of anodal-tDCS in a larger sample to account for greater variability in language gains early after stroke, and the effects on functional communication. Additional work is also needed to identify which aphasic individuals are likely to benefit from tDCS to augment aphasia recovery.

Supplementary Material

Supplemental Publication Material

Table S1. Reasons Patients were Ineligible.

Table S2. Dataset.

Table S3. Frequency of all Adverse Events by Treatment Group.

Abbreviations:

A-tDCS

anodal transcranial direct-current stimulation

AQ

Aphasia Quotient

CU

Content Unit

LH

left-hemisphere

L IFG

left inferior frontal gyrus

L PFC

left prefrontal cortex

L STG

left superior temporal gyrus

NIHSS

National Institutes of Health Stroke Scale

NMDA

N-methyl-D-aspartate

PNT

Philadelphia Naing Test

QoL

Quality of Life

SIS

Stroke Impact Scale

SLISSE

Stimulating Language in Subacute StrokE

WAB-R

Western Aphasia Battery-Revised