Effect of transcranial direct current stimulation on post-stroke fatigue

This was a double-blind, sham-controlled study with a single session of bilateral anodal tDCS chosen as the method of intervention. Patients visited the laboratory on three separate occasions, with tDCS applied only on the first visit. The second visit took place 1 week later with the third visit taking place one month after visit two and will be termed week and month throughout the manuscript (Fig. 1). The primary outcome measure was a change in trait fatigue. Secondary outcome measures included state fatigue, explicit and implicit measures of PE and motor cortex physiology measures of resting motor thresholds (RMT) and slope of recruitment curves (IO_Slope) of the affected and un-affected hemisphere assessed using TMS. The primary outcome measure was recorded at three distinct time points (pre stimulation, week and month). All other outcome measures were recorded at four distinct time points (pre stimulation, immediately post stimulation, week and month).

The study was approved by London Bromley Research Ethics Committee (16/LO/0714). Stroke survivors were recruited via Clinical Research Network at University College NHS Trust Hospital, a departmental Stroke Database and from the community.

Inclusion criteria: Date of stroke > 3 months, first-time stroke, age ≥ 18 years; Fatigue Severity Scale (FSS-7) ≥ 4. A score of ≥ 4 on FSS-7 indicates the presence of clinically significant fatigue [39]. Exclusion criteria: use of centrally-acting medication which may affect the level of fatigue, depression, and anxiety; depression scores ≥ 11 (Hospital Anxiety and Depression Scale—HADS); grip strength and manual dexterity (nine-hole peg test) ≤ 60% of unaffected hand.

The minimal clinically important difference on the FSS-7 is 0.45, with differences greater than 0.45 predicting a significant effect on the quality of life [40, 41]. To detect the minimal clinically important difference in fatigue with 80% power (0.80) and a significance level alpha of 0.05, a sample size of 11 subjects per group is needed. Twice the number of patients were allocated to the real stimulation group than necessary, as previous studies using tDCS in multiple sclerosis fatigue and healthy individuals showed that approximately 50% of patients respond to tDCS [31, 42]. Thirty-three patients were recruited into the study and were randomly allocated to the real (n = 22) or sham (n = 11) stimulation groups (Fig. 2). All patients gave written informed consent in accordance with the Declaration of Helsinki. Patient demographics for both groups are found in Table 1.

Trait and state measures of fatigue were captured during the study. Trait fatigue represents the experience and impact of fatigue on day to day living for a pre-determined time period leading up to the day of testing, whereas state fatigue characterizes fatigue at a given moment in time. Trait fatigue was quantified using the FSS-7, a seven-item questionnaire asking for ratings of fatigue ranging from one to seven (strongly disagree to strongly agree) over the preceding week from the day of administration [43]. An average score of one indicates no fatigue while an average score of seven indicates very severe fatigue. State fatigue was quantified using a visual analogue scale (VAS) ranging from zero to ten (Not at all tired to extremely tired). Patients also completed the HADS, a 14-item questionnaire with a depression and anxiety subscale, prior to the stimulation. A score of 0 to 7 for either subscale could be regarded as being in the normal range, with a score of 11 or higher indicating the probable presence of the mood disorder [44].

tDCS was applied using two battery-driven stimulators (DC-Stimulator Plus, NeuroConn, Germany) while patients were awake and at rest. Four 35 cm² rubber electrodes coated with conductive paste were secured with self-adhesive bandages. The anode of each stimulator was placed over the left and right M1 (C3 and C4 according to the 10–20 EEG system), while the cathode of each stimulator was placed over the ipsilateral left and right shoulders. This tDCS montage has previously been shown to reduce the perception of effort and increase corticospinal excitability in healthy individuals[38]. Real tDCS involved two 20-min sessions of stimulation at 2 mA separated by a 10-min break in between. The current was ramped up for 30 s until reaching 2 mA and ramped down for 30 s at the end of the stimulation period. Stimulation intensity and duration complied with current safety recommendations [45]. For sham stimulation, the current was ramped down immediately after ramping up, providing effective blinding [46]. The patient and researchers were blind to the applied stimulation (real or sham). At the end of stimulation, patients were explicitly asked whether they thought they received real or sham stimulation.

PE was measured in an isometric handgrip task with a hand-held dynamometer (Biometrics Ltd, Newport, UK) performed using the dominant hand [21]. Force data from the dynamometer were acquired at 500 Hz via a data acquisition interface (Power1401, CED) and recorded in MATLAB (2016b, MathWorks). Each trial was 5 s long, in which patients were required to sustain a grip force for 3 s at 20%, 40%, or 60% of their maximum voluntary force. Immediate force feedback was shown on the monitor as a filling of a red bar, which turned green once the minimal required target force, indicated by a cross on the screen, was reached. The grip force–visual feedback relationship was individually adjusted for every patient to eliminate potential influence on PE. Before the experiment, patients practiced each force level with their dominant hand to familiarize themselves with the effort required and performed a line familiarization. In the line familiarization, patients were shown 3 “short’ lines (1, 2, and 3 cm), and 3 “long” lines (10, 11, and 12 cm). After the presentation of the 6 lines, patients were shown each of the learned lines without information about the category it belonged to and were asked to judge the line length. Patients responded using the keyboard: left arrow key for “short” and right arrow key for “long”. They were then asked to rate their confidence in their response using a VAS. If patients’ response was < 100% correct, the procedure was repeated until they were able to distinguish between short and long lines.

During the PE task, each grip was followed by a line length estimation. The line presented could have a length of 3.5–8.5 cm with a total of 24 different line lengths, 12 short and 12 long. Twenty-four lines presented under the 3 force conditions resulted in a total of 72 trials divided into 3 blocks. The order of forces and line lengths was randomized with equal numbers of the 3 different force levels in each block. Participants reported if the presented line was short or long based on the length of lines presented during the familiarization phase. If they determined the presented line to be shorter than half the length of the longest line presented during the familiarization (12 cm), they reported short; otherwise, they reported long. These blocks were used as an implicit measure of PE.

After 3 blocks, participants performed a final block of 9 trials. This block was used as an explicit measure of PE. Each trial consisted of a 5-s grip with visual feedback at the 3 different force levels, 20%, 40%, or 60% of maximum voluntary force, with 3 trials for each force level. This was followed by the question, “How effortful was the squeeze?” Patients had to respond using a VAS ranging from “not at all” to “very hard.”

Electromyogram (EMG) recordings were obtained from the first dorsal interosseous (FDI) muscle using surface electrodes (1041PTS Neonatal Electrode, Kendell) in a belly-tendon montage with the ground positioned over the flexor retinaculum of the hand. The signal was amplified with a gain of 1000 (D360, Digitmer, Welwyn Garden City, UK), bandpass filtered (100–1000 Hz), digitized at 10 kHz (Power1401, CED, Cambridge, UK) and recorded with Signal version 6.04 software (CED, Cambridge, UK).

TMS (figure-of-eight coil with wing diameter, 70 mm; Magstim 2002, Magstim, Whitland, UK) was used to stimulate the hand area of the M1. The coil was held tangentially on the scalp at 45° to the mid-sagittal plane to induce a posterior-anterior current across the central sulcus. The subjects were instructed to stay relaxed with their eyes open and their legs uncrossed. The motor ‘hotspot’ of the FDI muscle was determined as described previously [7]. RMT was defined as the lowest intensity required to evoke a motor evoked potential (MEP) at the hotspot of at least 50 μV in a minimum of 5 of 10 consecutive trials while subjects were at rest. IO curves were acquired at rest at the hotspot using TMS intensities set at 100, 110, 120, 130, 140 and 150% of RMT. Six pulses at each of the 6 intensities were delivered in a randomized order with an inter-trial interval of 4 s, giving thirty-six trials in total. This procedure was repeated for both the affected (RMT-A, IO_Slope-A) and unaffected (RMT-U, IO_Slope-U) hemispheres.

The FSS-7 score was calculated by averaging all items for each of the three-time points.

The total score was taken for the anxiety and depression subscales of HADS, HADS-Anxiety and HADS-Depression respectively, and were considered as independent measures.

The data were analysed using custom-written routines in Matlab (2018a, Mathworks). Peak-to-peak MEP amplitudes for each condition were estimated from the EMG recordings. All trials were visually inspected and approximately 7% of trials with pre-contraction and size ≤ 0.025 mV were excluded across all participants. A linear fit was applied to all the MEP data across the six conditions (100–150% RMT) for each participant at each session. The quality of the linear fit was evaluated by calculating the r-squared value for each participant across each session. The grand mean r-squared value across all individuals and all session and the resulting standard deviation (R² = 0.86 ± 0.12) demonstrate an overall good fit of the MEP data with little variability within conditions. The gradient of the linear fit was subsequently calculated for each participant in each of the four sessions and for each hemisphere (affected and un-affected hemisphere) giving us the slope of the recruitment curve.

To obtain a measure of explicit PE, VAS scores were averaged across all trials in each force level in each individual. As there was no difference across force level, the average VAS score across all force levels was used as an explicit measure of PE for each of the four-time points. To obtain a measure of implicit PE, the sum of the number of lines reported as long for each individual in each force level was calculated. As there was no difference across force level, the average number of lines across all force levels was used as an implicit measure of PE for each of the four-time points.

All statistical analysis was performed using R (RStudio Version 1.2.5033). Assumptions of a normal distribution of the primary and all secondary outcome variables were assessed using the Shapiro–Wilk test. All data were non-normally distributed (p < 0.05). To test for changes over time, a non-parametric Friedman test was performed for the primary outcome variable (trait fatigue) and all secondary outcome variables (state fatigue, RMT-A, RMT-U, IO_Slope-A, IO_Slope-U, PE-implicit and PE-explicit), separately for the sham and real intervention groups as in Saiote et al. [31]. When significant results were found, pairwise comparison between baseline and each post-measurement day were performed using Wilcoxon signed-rank test. To analyse the effect of real stimulation versus sham stimulation across both primary and secondary outcome measures, the changes in scores were calculated by normalising each day to baseline (pre stimulation) and then compared within each day using a Wilcoxon signed-rank test. Adjustment for multiple comparisons was performed using Bonferroni correction.

A spearman correlation was used to examine the association between baseline trait fatigue scores and the change in trait fatigue a week after stimulation in both the sham and real stimulation groups. To identify the potential mechanisms that drive the change in trait fatigue in the real stimulation group a multiple linear regression was used with demographic data and secondary outcome variables that were significantly different between the real and sham stimulation groups used as predictors. Collinearity amongst the predictors used in the multiple regression model was assessed by computing the variance inflation factor (VIF). No VIF value exceeded a score of 5, demonstrating that there was no collinearity amongst the predictors used. Goodness of fit was assessed using the BIC (lower BIC indicates a better fitting model) to identify the combination of variables that best predicted the outcome variable, the change in trait fatigue. Assumptions of normality and homoscedasticity of the residuals for each model were assessed visually using quantile–quantile normal plots and fitted- versus residual-value plots.

The Friedman test showed a significant effect of time on trait fatigue in the real stimulation group (χ²(2) = 15.5; p < 0.001) but not in the sham stimulation group (χ²(2) = 0.154; p = 0.926). Post-hoc analysis with Wilcoxon signed-rank tests showed that FSS-7 scores decreased compared to baseline both at the week (V = 198, Z = 0.777, p = 0.002) and month (V = 142, Z = 0.535, p = 0.047) time point in the real stimulation group. FSS-7 at the week time point was also lower than at the month time point (V = 21, Z = 0.639, p = 0.016). Post-hoc Wilcoxon signed-rank test of the normalised FSS-7 scores revealed a significant difference between the sham and real stimulation group at the week time point (W = 52.5, Z = 0.382, p = 0.0386) but not at the month time point (W = 65.5, Z = 0.277, p = 0.134), seen in Fig. 3a.

The Friedman test showed no significant effect of time on state fatigue in the real (χ²(3) = 0.97; p = 0.809) or sham (χ²(3) = 3; p = 0.392) stimulation group. Post-hoc Wilcoxon singed-rank test for the normalised state fatigue scores revealed no significant difference between the sham and real stimulation group at any time point (Fig. 3b).

The Friedman test showed a significant effect of time on IO_Slope-A in the real stimulation group (χ²(3) = 11.2, p = 0.0106) but not in the sham stimulation group (χ²(3) = 0.75, p = 0.861). Post-hoc analysis with Wilcoxon signed-rank tests showed that IO_Slope-A decreased compared to baseline at the week (V = 96, Z = 0.730, p = 0.024) but not at the immediate post and month time point in the real stimulation group. Post-hoc Wilcoxon signed-rank test of the normalised IO_Slope-A scores revealed a significant difference between the sham and real stimulation group at the week time point (W = 23, Z = 0.480, p = 0.024) but not at the immediate post and month time point. The Friedman test showed no significant effect of time on RMT-A in the real (χ²(3) = 2.62, p = 0.454) or sham (χ²(3) = 2.06, p = 0.560) stimulation groups. Post-hoc Wilcoxon singed-rank test for the normalised RMT-A revealed no significant difference between the sham and real stimulation group at any time point.

The Friedman test showed no significant effect of time on RMT-U or IO_Slope-U in the real (χ²(3)2.71, p = 0.438; χ²(3) = 3.38, p = 0.337) or sham (χ²(3) = 5.62, p = 0.132; χ²(3) = 2.2, p = 0.552) stimulation groups. Post-hoc Wilcoxon singed-rank test for the normalised RMT-U and IO_Slope-U scores revealed no significant difference between the sham and real stimulation group at any time point (Fig. 3c–f).

The Friedman test showed no significant effect of time on PE-implicit or PE-explicit in the real (χ²(3) = 5.60, p = 0.905; χ²(3) = 0.2, p = 0.978) or sham (χ²(3) = 4.24, p = 0.237; χ²(3) = 6.73, p = 0.0809) stimulation groups. Post-hoc Wilcoxon singed-rank test for the normalised PE-explicit scores revealed no significant difference between the sham and real stimulation group at any time point. Post-hoc Wilcoxon singed-rank test for the normalised PE-implicit scores revealed a significant difference between the sham and real stimulation group at the week time point (W = 39.5, Z = 0.391, p = 0.0491) but not at the immediate post and month time point (Fig. 3g–h).

There was no association between baseline FSS-7 scores and the change in FSS-7 score at the week time point in the real (r_s = − 0.2, p = 0.39) and sham (r_s = 0.2, p = 0.57) stimulation group (Fig. 4a). The multiple linear regression model with normalised IO_Slope-A at the week time point and baseline HADS-Anxiety scores as predictors was the best fitting model (ΔFSSweek =  − 2.775 + 0.436ΔIO_Slope-A + 0.285HADS-Anxiety). A significant regression equation was found (F(2,13) = 5.345, p = 0.020), with an adjusted R² of 0.37. Baseline HADS-Anxiety significantly explained the change in FSS-7 scores (t = 2.925, p = 0.013), whereas normalised IO_Slope-A at the week time point (t = 1.760, p = 0.104) was not a significant predictor of the change in FSS-7 scores (Fig. 4b). Beta coefficients of the predictors used in the model together with their associated 95% confidence intervals and p-values are found in Table 2.

Anodal tDCS increases cortical excitability [28] and reduces the perception of effort in healthy participants during an endurance performance task [38]. The effects of tDCS are not specific to the targeted brain regions but spread to distinct cortical and subcortical structures [47]. A single session of anodal tDCS was expected to increase motor cortex excitability temporarily resulting in reduction of PE by modulating excitability and connectivity of regions upstream of M1 [12, 13, 48]. By resetting PE in a physical task, the carry over effects on behaviour would accumulate eventually reducing fatigue levels. This reduction in fatigue levels would be evident by changes in trait fatigue but not necessarily state fatigue. The current results partly support this theory by showing a reduction in trait fatigue a week post-tDCS. Fatigue has been identified as one of the top unmet needs for chronic stroke survivors within the community and interferes most with their activities of daily living [3, 4]. Currently, there are no effective interventions for PSF [49, 50]. The proposed intervention, if confirmed to be effective in a larger and a more heterogeneous cohort of patients, will represent a simple, low-cost and risk-free procedure for reducing fatigue symptoms [24, 30].

RMT is a measure of motor corticospinal excitability and anodal tDCS reduces RMT [28]. The lack of effect of tDCS on RMT in this study could have been driven by our split stimulation paradigm of two 20-min periods invoking homeostatic metaplasticity [51], with the second 20-min stimulation reversing the effects of the first 20 min. The slope of IO curves following real tDCS reduced significantly in the affected hemisphere when compared to sham tDCS. IO curves are thought to measure the excitability of pathways upstream of the motor cortex, therefore a measure of inputs to the motor cortex [52, 53]. The slope of IO curves represents the gain of descending corticospinal tract, possibly driven from higher motor areas [54, 55]. However, a reduction and not an increase in slope is puzzling. Steeper IO represent greater recruitment of higher-order motor areas such as the supplementary motor area and premotor cortex [53]. As there are no normative values for IO curves, it is hard to speculate if tDCS induced shallowing of IO curves is a reflection of homeostatic normalisation of inputs to the motor cortex, but offers one possible explanation of the IO results.

PE has mostly been tested in post-exercise paradigms where measures such as VAS and Borg Scales have been validated against physiological measures of exertion such as heart rate and maximal aerobic capacity [56, 57]. They conclude PE is less subject to bias in some populations [58, 59]. However, in the current study PE is measured in a non-exercise paradigm and in a disease population with a condition that is highly stigmatised and under recognised [60, 61]. Hence, we expected to see a response bias in PE. Therefore, in addition to an explicit measure of PE, we introduce a novel implicit measure of PE based on line length perception. This measure takes advantage of the susceptibility of visual perception to physical effort where high effort unfavourably biases distance estimation [62]. On similar lines, a line length estimation task was developed and shown to be biased by prior exertion, which we use as a measure of implicit PE [21, 63]. We showed that trait fatigue is explained by implicit PE but not explicit PE in a physical task [21]. In the current study, we show a significant difference between real and sham stimulation in implicit PE but not in explicit PE a week after anodal tDCS. It could be that implicit, perceived effort is the first to respond to tDCS and if the effects were maintained, would result in a reduction in explicit PE. A second possible explanation is that reducing implicit PE that is sufficient to alleviate fatigue. This difference appears to be primarily driven by an increase in implicit PE in the sham stimulation group. Perhaps in the real stimulation group this task-specific increase in PE was suppressed. The test–retest reproducibility of the paradigm used to measure implicit PE has not been examined in the absence of tDCS, and could potentially shed light on the current finding. It is important to note that all patients in this study were physically well recovered, evident from their upper limb clinical scores (grip and NHPT, Table 1). Therefore, we did not expect there to be an effect of hemiparetic side on their ability to perform the task successfully.

One of the aims of this work was to identify the potential mechanisms that underlie the reduction in trait fatigue following anodal tDCS. The expectation was that a change immediately post stimulation in M1 neurophysiology and PE will result in a reduction in trait fatigue a week later. However, no change immediately post stimulation, but only a week later in these measures, makes it harder to interpret the neurophysiology and perception results. The results of the multiple linear regression model suggest that measures of M1 neurophysiology, specifically of the affected hemisphere a week after stimulation, appear to be related to the improving fatigue symptoms. Whether the change in fatigue is a consequence of a change in neurophysiology or whether they occur via independent mechanisms cannot be inferred from the current study. Therefore, from a mechanistic point of view, there are still questions to be addressed with regards to how anodal tDCS results in reduced trait fatigue. The improvement in fatigue symptoms does however appear to be modulated by anxiety levels prior to the stimulation itself. PSF and anxiety co-occur more often than any other problems such as pain, depression and sleep [64]. In chronic pain, anxiety exacerbates pain [65]. Similarly, anxiety levels may exacerbate fatigue which manifests as a small or no change in fatigue in those with high baseline anxiety. Anxiolytic medication (e.g. diazepam, lorazepam) targets the neurotransmitter GABA and anodal tDCS enhances cortical excitability by decreasing GABA concentration [66]. This could have prevented the efficacy of tDCS in reducing fatigue, however, patients included in the current study were not on any centrally acting medication. Therefore, the true interaction between anxiety and effect of tDCS remains unclear.

Despite providing a potential intervention to improve fatigue symptoms after stroke, this study is not without limitations. Firstly, the current study is limited to non-depressed, minimally impaired stroke survivors. Given the nature of the symptom being investigated and the heterogeneous cohort that are stroke survivors, the effect of anodal tDCS on post-stroke fatigue should be investigated in a wider range of stroke survivors. Secondly, due to the nature of the symptom being investigated, a small number of trials was used for the recruitment curve data and only a single session of anodal tDCS was performed to ensure that all patients could complete the study. Having multiple sessions of tDCS as in previous studies in multiple sclerosis [31, 33‐37, 67, 68] might result in improvement in fatigue scores lasting longer than a week. Finally, despite differences in neurophysiological and PE measures between the real and sham stimulation groups, the reduction in fatigue scores is not fully explained by these measures, leaving the question of the mechanism of tDCS-induced reduction in fatigue, still open.