Standard protocol approvals, registrations, and patient consents
The study was reviewed and approved by the Institutional Review Board at Beth Israel Deaconess Medical Center. We confirm that all research procedures was performed in accordance with relevant regulations and in accordance with the Declaration of Helsinki. A written informed consent was obtained for all trial participants or their legal guardians. A Data Safety Monitoring Board with assistance from a Medical Safety Monitor reviewed the progress of the trial. This clinical trial was registered on Clinicaltrials.gov (URL: http://www.clinicaltrials.gov. Unique identifier: NCT01919112) on 8/08/ 2013.
Trial Design and Participants
FEASt is an early phase-2 double-blind, single-center RCT conducted from September 2013 -September 2019. The rationale, design and protocol of the trial have been published previously . All subjects were recruited at Beth Israel Deaconess Medical Center (BIDMC) in Boston, MA. We enrolled patients with an acute-subacute unilateral hemispheric infarction, day 2-day 6 since the qualifying stroke, between 21–90 years of age, with moderate to severe dysphagia, defined as a Penetration And Aspiration Scale  (PAS) score ≥ 4 on a standardized videofluoroscopic study (VFSS). The main exclusion criteria were pre-stroke swallowing difficulties, severe stroke (NIHSS score ≥ 25), or presence of any other contraindication for tDCS .
Randomization and Masking
Subjects were randomized to 1 of 3 intervention arms–a Low-Dose tDCS, a High-Dose tDCS, or a Sham group, using computer based randomization stratified according to the baseline PAS scores (4–6 versus 7–8). Randomization codes were generated electronically by an independent study Data Coordination Center (DCC). Three individuals, not involved in any other study procedure received confidential electronic notification about the randomization codes and programmed the tDCS device accordingly. They also interrogated the device after each session to verify its fidelity to the stimulation allocation. All de-identified VFSS were sent electronically to the Boston University Medical Center laboratory for detailed analysis and results were fed directly to the DCC. These processes ensured complete concealment of intervention allocation and blinding of all investigators and subjects.
Swallowing Procedures and Outcomes Assessments
All consecutive acute ischemic stroke (AIS) patients admitted to BIDMC were screened for trial participation. All patients received a standardized dysphagia screening using a 3-ounce water swallow test  and underwent a swallow evaluation by a Speech-Language pathologist (SLP) if suspected of having dysphagia. Those with moderate to severe dysphagia (PAS ≥ 4) on a standardized VFSS evaluation were enrolled if they fulfilled all study criteria. The VFSS protocol employed 3 consistencies (nectar, pudding and thin liquids) and 5 swallows at trial enrollment (baseline) and after the 10th or last stimulation session (post-stimulation). The arithmetic mean of the PAS score was computed for each subject (composite PAS score) at these time points. A Functional Oral Intake Scale (FOIS) score  was computed by the study SLP at baseline, after 5th and final session; in addition, at 30-days after the index stroke, an investigator, blinded to the intervention allocation, collected a FOIS score over the phone or in-person using a standardized questionnaire with the subject or a legally authorized representative familiar with subject’s current dietary status. An investigator certified in the performance of the NIH Stroke Scale (NIHSS)  assigned scores for each participant at enrollment, after every second stimulation sessions and after the last session. All subjects underwent a brain MRI with diffusion weighted imaging (DWI) prior to enrollment, which were used to create a binary map of the acute stroke lesion and construct a novel radiological variable, the corticobulbar tract-lesion load (CBT-LL) (below).
The primary efficacy outcome was a change in PAS scores based on VFSS  from baseline to post-stimulation. The primary safety outcomes were incidence of seizures, deterioration in global neurological functions (≥ 4-point increase in the total NIHSS score), motor functions (≥ 2 points increase in the motor sub-item of the NIHSS score on the same limb), swallowing functions (increase in PAS score by ≥ 2 points compared to baseline), and stroke specific mortality during the period of active stimulation. The main secondary efficacy outcomes were improvement in diet as assessed by a change in FOIS scores  at 30-days post-stimulation and changes in swallowing physiology parameters on VFSS from baseline to post-stimulation [pharyngeal delay time (PDT), pharyngeal constriction ratio (PCR), and hyoid, laryngeal and pharyngeal excursion (HLPE)]  [26–28].
The CBT-LL variable was derived by creating a canonical tract of the corticobulbar tract in spatially standardized space using the swallowing related cortical fMRI activation and the posterior pons as seed regions. The fMRI experiment consisted of repeated swallowing trials contrasted with whole hand opening and closing tasks, done at the same frequency as the swallowing tasks. The fMRI sequences were acquired with a gradient-echo T2*-weighted MR pulse sequence using our own modification of a sparse temporal sampling method with clustered volume acquisition to overcome imaging artifacts caused by swallowing motion. The voxel clusters of significant cortical activation were used as a seed region (to identify the cortical origin of the CBT) with the second seed region in the posterior pons. The DTI scans were high resolution studies obtained in 12 healthy elderly controls. Image acquisition, analysis, construction of corticobulbar tracts and computation of CBT-lesion load were performed as previously described by Zhu et al . Overlaying the manually drawn lesion maps derived from DWI sequences of trial participants onto the canonical CBT allowed us to calculate a CBT-lesion-load variable. For clinical interpretability, the CBT-LL variable was dichotomized for each subject based on their volume, into 2 groups (< group median, ≥ group median) to assess for effect modification of this variable on the experimental intervention.
We performed a post-hoc analysis using a FOIS score threshold ≥ 5 as clinically meaningful improvement in dietary intake based on a recent noteworthy publication that utilized this threshold to develop an instrument for prognosticating dysphagia recovery . This cut-off corresponds to a dietary intake of a single consistency or worse and demonstrated to correlate with significant reduction in protein, energy and fluid intake, and suggested as an indication for gastrostomy tube placement after stroke [31–33].
The stimulation session were performed for 20 minutes twice daily over 5 consecutive days. Anodal tDCS (2 mA) or sham was delivered via a battery-driven, constant current stimulator (NeuroConn-DC Stimulator Plus) using saline soaked electrodes (anode 3x5 cm; reference electrode 5x7 cm). We targeted the healthy swallowing motor cortex using the 10–20 EEG electrode placement system, placing the anode mid-distance between C3/T3 [left] or C4/T4 [right] over the unaffected hemisphere and the reference electrode over the contralateral supraorbital region. We had previously verified the location of the stimulating electrode using a combination of functional brain MRI (fMRI) and anatomical brain MRI scans . The electrode positioning was re-confirmed in a subset of trial participants using anatomical brain MRI scans. The atDCS electric field (V/m) distribution was modeled using a freely available software package called SimNIBS  (Fig. 1). The MNI ICBM 152 1mm T1 weighted image was used to perform simulation . As shown in Fig. 1, the electric field distribution (V/m) is influenced by the shape, location, and current applied through the electrodes.
The High-Dose tDCS group received 2mA atDCS twice daily for a total of 20 minutes (total charge density 16 C/cm2); Low-Dose group received 2 mA alternating with sham stimulation daily for a total of 20 minutes (total charge density 8 C/cm2) and Sham group received sham stimulation twice daily.
All stimulation sessions were conducted concurrently with standardized effortful swallowing exercises . All study SLPs and investigators were trained by an expert (SL) on eliciting these maneuvers. Occurrence of an effortful swallow was verified by a laryngeal microphone attached to the patient’s throat. We aimed to obtain 40 effortful swallows during each session, which were recorded and entered into the study website.
We planned to randomize 99 subjects- 33 in each arm, estimated to detect a difference of 1.0 and 1.15 standard deviations (SD) between groups in the mean primary outcome measure with a type I error rate of 2.5% and power 80% and 90%, respectively.
The primary efficacy analysis was performed in an intent-to-treat approach. Our outcome variable was a change in the arithmetic mean PAS scores (averaged for 5 different swallows at each time point), before and after intervention and compared across groups. We used a linear model using PROC GLM in SAS to analyze the mean change in PAS scores with the intervention after adjusting for baseline NIHSS scores and age, which were included as covariates based on clinical considerations after screening for other confounders. Similar analyses were performed on the dietary outcome (mean change in FOIS scores at 30-days) and other secondary analysis on changes in swallowing physiology (PDT, PCR, and HLPE). The incidence of safety outcomes were captured as a whole and compared separately across groups. Intraclass correlation (ICC) for PAS scores was performed on a subset of VFSS (randomly selected 303 swallow evaluations) between 2 reviewers (JP and SL). Regression models using interaction terms were used to assess for heterogeneity of intervention effects across the CBT-LL.
In addition, we conducted the following post-hoc analysis: 1) PRESS score, a recently validated prognostic tool for spontaneous swallowing recovery after an AIS, was computed for each participant and incorporated in an adjusted analysis for the efficacy outcomes ; 2) the total number of experimental sessions (10 or < 10) for every subject was used as a covariate in an adjusted analysis for our primary outcome; 3) Fisher-exact test was conducted to compare the proportion of subjects with a FOIS score ≥ 5 in the combined atDCS group versus sham. Treatment comparison p-values were considered statistically significant at the two-sided 0.05 level of significance; interaction p-values were considered statistically significant at the 0.15 level of significance given the relatively low power for tests of interaction to detect a true interaction. SAS Version 9.4 was used to carry out all analyses.