Safety of Transcranial Direct Current Stimulation: Evidence Based Update 2016

Abstract

This review updates and consolidates evidence on the safety of transcranial Direct Current Stimulation (tDCS). Safety is here operationally defined by, and limited to, the absence of evidence for a Serious Adverse Effect, the criteria for which are rigorously defined. This review adopts an evidence-based approach, based on an aggregation of experience from human trials, taking care not to confuse speculation on potential hazards or lack of data to refute such speculation with evidence for risk. Safety data from animal tests for tissue damage are reviewed with systematic consideration of translation to humans. Arbitrary safety considerations are avoided. Computational models are used to relate dose to brain exposure in humans and animals. We review relevant dose-response curves and dose metrics (e.g. current, duration, current density, charge, charge density) for meaningful safety standards. Special consideration is given to theoretically vulnerable populations including children and the elderly, subjects with mood disorders, epilepsy, stroke, implants, and home users. Evidence from relevant animal models indicates that brain injury by Direct Current Stimulation (DCS) occurs at predicted brain current densities (6.3-13 A/m(2)) that are over an order of magnitude above those produced by conventional tDCS. To date, the use of conventional tDCS protocols in human trials (≤40 min, ≤4 milliamperes, ≤7.2 Coulombs) has not produced any reports of a Serious Adverse Effect or irreversible injury across over 33,200 sessions and 1000 subjects with repeated sessions. This includes a wide variety of subjects, including persons from potentially vulnerable populations.

Keywords: Electrical stimulation; Mood disorders; Safety; Transcranial Direct Current Stimulation; tDCS; tDCS safety.

Figure 1. Computational models predict skin current density to brain current density, and so their ratio

An exploratory analysis compared a range of montages on extremes of head anatomy (e.g. pediatric to obese, healthy and stroke). Additional models were solved to increase the depth of the study (methods of computational forward modeling are described in detail in Figure 3). Models included some or all of the following tissue masks: skin (0.465 S/m), fat (0.025 S/m), bone (0.01 S/m), CSF (1.65 S/m), gray matter (0.276 S/m), white matter (0.126 S/m), intervertebral discs (0.16133 S/m), ligament (0.250922 S/m), spinal cord (0.171267 S/m), air (1×10⁻¹⁵ S/m), electrode (5.99×10⁷ S/m), sponge (1.4 S/m), and gel (4.0 S/m). The review involved nineteen combinations of six different head types (pediatric, small adult, medium adult, medium adult stroke, large adult, and obese adult) and ten different electrode montages (two using HD electrodes and eight using 5×7cm sponge pad electrodes). (Top) Because electrodes are placed on the scalp during tDCS, and due to the conductivity and anatomy of the underlying tissue, a majority of the current does not reach the brain and the fraction that does reach the brain is diffused. The current density in skin is thus invariably higher than in the brain. (Bottom) The maximum current density in the skin and the brain (and their ratio) depends on several factors including the electrode montage. For a single head, the ratio is predicted for various conventional and HD montages. The skin to brain ratio varies from greater than 10:1 to 400:1. The maximum brain current density was 0.23 A/m2 for a small adult head and 0.32 A/m2 for pediatric head (Image adapted from Tyler et al, in preparation)

Figure 2. Number of Sessions by Duration (a) and Number of Subjects in Repeated Sessions by Duration (b)

Interim analysis of total sessions and dose in tDCS literature of published trials meeting our inclusion criterion. We searched the Pubmed database with the key words “transcranial direct current stimulation” limiting to papers published in English. We only included studies that met the following inclusion criteria: (1) used tDCS, (2) tested on human subjects, (3) reported original research, (4) used an electrolyte-soaked absorbent material, (5) clearly reported dosage information, and (6) published before July 2013. 488 of 1072 papers were considered. Of those, tDCS dosage and number of sessions were extracted. tDCS dosage includes current intensity, electrode size, duration, and position (not reported here). Number of sessions refers to the number of tDCS procedures completed (e.g. number of subjects times sessions per subject). If one subject underwent more than 4 sessions in one week, it was further classified as a repeated session. Data from ~55% of all original tDCS publications extracted using a combination of automated lexica analysis and manual screening (solid bar) and extrapolated based on update tDCS publication volume (lighter bar) – therefore the solid bars represent verified statistics until July 2013 while lighter bar represents prediction given fixed distribution of sessions in all papers to-date. We extracted the following dose parameters: Current applied, duration of session, current density at electrodes, and for studies with repeated sessions (three to seven treatments per week for at least one week) the number of sessions. If a clinical trial tested more than one tDCS condition per subject (e.g. anode vs. cathode over M1 separated by one week) these were considered separate sessions since the study design intended no carry-over effects and since the results of each session provide evidence for safety. From each qualified trial the total number of sessions and the parameters (intensity, duration, current density) for each session where determined. Therefore the total number of sessions applied at any given parameter (e.g. current of 2 mA) or combination of parameters (e.g. current of 2 mA and duration of 20 minutes) for tDCS trials is known. For repeated sessions, we aggregated data by number of subjects. Cumulative data is plotted under the assumption that increasing intensity or duration does not decrease risk, such that sessions at 2 mA support safety at 1 mA. This assumption presumes a monotonic dose-safety response curve and does not apply to efficacy.

Figure 3. Finite Element models comparing a common tDCS montage in human and three DCS montages in animal model

Animal data indicates possible injury at electric field thresholds over an order of magnitude above those generated by conventional tDCS protocols. (A) The evident difference in gross anatomy between human and rat is considered in computational models based on high-resolution MRI. (B, C, D) Lesion threshold in rat brain as reported by three different groups using modestly varied methods. The predicted minimum induced current density for brain lesions ranged from 12, 17, 6.3 A/m² (corresponding to electric fields of 42, 61, to 23 V/m) for the montages used by Liebetanz et al. (B) Fritsch et al (C) and Jankord et al. (D) respectively. In contrast to human tDCS, epicranial stimulation of the rat brain (stimulation applied over skull) resulted in higher cortical electric fields for the same input current, magnified by the smaller head anatomy. (E) Typical human tDCS driven by 2 mA at the electrodes resulted in 0.096 A/m² (0.35 V/m) on the cortex. (F) To match the cortical lesion threshold found in Liebetanz et al., 120 mA would have to be applied in human. We note however that current density in the skin would be much higher than in the brain, such that skin injury would potentially manifest well before the risk of brain injury. (Simulations and data adopted from Liebetanz et al 2009, Frisch unpublished observations, Jankord unpublished observations, Truong unpublished observations).

Figure 4. Endogenous oscillation activity does not significantly reduce epileptic threshold compared to quiescent state in brain slices

The electric field intensity needed to generate epileptiform activity can be systematically investigated in bran slices. Uniform electric field up to 125 V/m were tested, but note tDCS produces electrics fields <1 V/m. a) Rat hippocampal slices were prepared according to Reato et al. 2010 and stimulated with electric fields (E) of varying intensity (10 s duration) along the somatodendritic axis of CA3 pyramidal neurons. 20 µM carbachol was added to the aCSF superfusate to induce spontaneous neural activity. Slices were either quiescent (normal aCSF superfusate; Black traces) or oscillating (+carbochal ACSF, Red traces). Inset is a computational model showing the membrane polarization profile of a pyramidal neuron in an electric field (Dep, depolarized; Hyp, hyperpolarized). b) Epileptiform activity was quantified as the average variance of the voltage signal during stimulation, normalized to the average variance during the 10 seconds before stimulation. Mean +/− SEM across slices is plotted for each electric field intensity. Epileptic threshold for each slice was quantified as the electric field intensity required to produce a normalized variance >10. There was no significant difference in epileptic threshold between carbcahol and control conditions (p = 0.48; carbachol 98.3 +/− 15.61 V/m n = 10; control 104 +/− 14.97 V/m n = 9). c) Sample traces during stimulation at each electric field intensity (Left column black-traces normal aCSF Right column red-traces +Carbachol). Highlighted traces at 40,80, and 120 V/m/ Even at 40 V/m (>40× electric fields generated by tDCS) no epileptiform activity compared to baseline (quiescent or oscillation) was detected (Kronberg at al. unpublished data).

Figure 5. Dose response curves and assumptions

While all existing data from animal models of epicranial DCS suggest injury thresholds are well above conventional tDCS protocols, data across intensities is limited and various models (e.g. animal species and strain) and metrics (e.g. histological stains) of injury are used (represented schematically by color of X). (A) The approach taken in this review was limited to empirical data based on the lowest reported injury threshold as a current intensity. This approach is limited by the sensitivity of the experimental measure for injury and other details of the experiment and limitations of animal models. But, none-the-less, the single lowest reported injury intensity represents at this time a condition clinical tDCS should not approach. (B) Alternatively, various dose response curves can be fit to one or more data sets. This approach is especially sensitive to assumptions about threshold, even when existing data from a given trial appears to fit a specific curve reliably. Using any extrapolated function will inevitably lead to a projected injury threshold below the lowest experimentally measured threshold. In summary, in the absence of an established mechanism for injury and/or a justified dose-response curve, it is difficult to reliably extrapolate below existing experimental determined thresholds for injury in animals – which are substantially above intensities generated by conventional tDCS.

Figure 6. Visual display of original Research papers published (a), along with Sessions performed and Subjects tested by annum (b). Quantitative data for both are listed separately (c).The ratio of sessions/paper per year is also given (d) and subjects/paper per year (e)

Both ratios were calculated using the average amount of sessions or subjects per year. Lighter colors indicate projections (see rationale in Figure 2).

Figure 7. tDCS subject Demographic Charts

Medical conditions of subjects treated with tDCS, as reported in the papers analyzed (a). Papers published by medical conditions of subjects (b). Number of sessions given by average age (c) for all sessions that reported age, as well as sessions with subjects with depression, stroke/aphasia, only healthy subjects. Age was calculated by taking the average age of a subject group or using the age range to calculate an approximate average age.

Figure 8. Computer simulation of brain current flow produced by tDCS in the presence of an idealized Deep Brain Stimulation lead

Finite Element models of tDCS with and without burrhole defects typical in subthalamic nucleus deep brain stimulation. Common sponge (conventional) and HD-tDCS montages for motor and cerebellar stimulation are compared. Four montages are simulation M1-SO (top left) and Inion-Zygoma (top right), 4×1 HD-tDCS over M1 (bottom left) and 4×1 HD-tDCS over Iz (bottom right). In each care current flow produced through the head in the “healthy” case (intact skull and brain, no implant) is compared to a worst-case scenario where two “burr holes” through the skin, skull and brain are created and fully filled with CSF. Fluid filled burr holes draw a greater amount of current density than what would normally exist with healthy tissue (dashed images). However, peak current density and electric field are minimally affected (less than two fold). HD configurations have lower deep brain electric field intensities in general in addition to being more confined. Thus, even under these worst-case burr-hole conditions, there are only moderate changes in overall brain current flow patterns, and no large change in peak electric field. This suggest that under more realistic conditions (e.g. with the implant in place instead of CSF, skin present and the skull largely sealed) the presence of a DBS lead would not significantly change resulting brain current flow. Evidently, all models are limited to the assumption made and this simulation (Truong, Bikson et al., in preparation).