Effectiveness and Safety of Non-Invasive Neuromodulation for Vision Restoration: A Systematic Review and Meta-Analysis

ABSTRACT

We carried out a systematic review and meta-analysis to determine the effectiveness and safety of non-invasive electrical stimulation (NES) for vision restoration. We systematically searched for randomised controlled trials (RCTs) comparing NES with sham stimulation, for vision restoration between 2000 and 2022 in CENTRAL, MEDLINE, EMBASE, and LILACS. The main outcomes were as follows: visual acuity (VA); detection accuracy; foveal threshold; mean sensitivity as the parameter for the visual field; reading performance; contrast sensitivity (CS); electroencephalogram; quality of life (QoL), and safety. Two reviewers independently selected studies, extracted data, and evaluated the risk of bias using the Cochrane risk of bias 2.0 tool. The certainty in the evidence was determined using the GRADE framework. Protocol registration: CRD42022329342. Thirteen RCTs involving 441 patients with vision impairment indicate that NES may improve VA in the immediate post-intervention period (mean difference [MD] = −0.02 logMAR, 95% confidence intervals [CI] −0.08 to 0.04; low certainty), and probably increases QoL and detection accuracy (MD = 0.08, 95% CI −0.25 to 0.42 and standardised MD [SMD] = 0.09, 95% CI −0.58 to 0.77, respectively; both moderate certainty). NES likely results in little or no difference in mean sensitivity (SMD = −0.03, 95% CI −0.53 to 0.48). Compared with sham stimulation, NES increases the risk of minor adverse effects (risk ratio = 1.24, 95% CI 0.99 to 1.54; moderate certainty). The effect of NES on CS, reading performance, and electroencephalogram was uncertain. Our study suggests that although NES may slightly improve VA, detection accuracy, and QoL, the clinical relevance of these findings remains uncertain. Future research should focus on improving the available evidence’s precision and consistency.

Introduction

The visual pathway includes the retina, optic nerve, optic chiasm, optic tract, lateral geniculate nucleus, and visual cortex in the occipital lobe of the brain.¹ Any injury that affects this circuit at any level can result in visual field defects (VFD) and may impact central vision, including visual acuity (VA).¹ The most common causes of VFD are glaucoma, retinal disorders, and diseases that compress, inflame, or affect the blood supply to the visual pathway (e.g., optic neuritis, tumours, or stroke).² When the condition involves macula vision, such as age-related macular degeneration (AMD) or axons that transmit visual information from this central area of the retina, VA may be compromised.³

Globally, the number of individuals with vision impairment worldwide has increased as life expectancy has also increased.^4,5 In 2020, 33.6 million adults aged ≥50 years were blind, with 12.3% of these cases attributed to glaucoma.^4,6 Additionally, the incidence of visual impairment and VFD in stroke was 60% and 24.8%, respectively.⁷ Furthermore, based on high-quality evidence, vision impairment has been associated with substantial economic burden, lower quality of life (QoL), highest disability burden, and elevated mortality.^8–11

Vision impairment resulting from ganglion cell or other neuron damage in the visual pathway has been considered irreversible for a long time. However, the ”residual vision activation theory”¹² proposes that areas of residual vision could induce dynamic changes, leading to considerable potential for vision recovery.¹³ Furthermore, there have been reports of spontaneous rehabilitation of some lost visual functions even without intervention, possibly related to neuronal plasticity.¹⁴ These hypotheses are not new, as training programmes for patients with vision loss caused by neurological conditions were proposed as early as the 1980s.^15,16 These programmes mainly involved repeated testing of the visual field (VF) border,¹⁷ prism therapy,^17,18 visual restorative therapy (VRT),^19,20 and compensatory therapy.¹⁹ Although these rehabilitation options were effective for some patients, many showed persistent vision impairment.²¹ For those patients with irreversible vision impairment, non-invasive electrical stimulation (NES) has garnered interest as a potential treatment option in the last two decades.^22–26

Since the early 2000s it has been known that weak current flows on the healthy visual system induce phosphenes²⁷ and increased activity in electroencephalography (EEG).²⁸ At the molecular level, animal models of optic nerve injuries have shown that NES treatment leads to increased delivery or production of growth factors, such as insulin-like growth factor, vascular endothelial growth factor, and activity-dependent neurotrophic factor peptide.²⁹ In addition, in both animals and humans, electrical stimulation and exposure to electromagnetic fields induce various potentially regeneration-enhancing effects, such as neuroplasticity, anti-inflammatory effects, blood–brain barrier recovery, and an increase in blood flow.^30,31

At present, NES has been evaluated to treat vision impairment with non-standardised protocols to deliver the current.³² The following different types of NES have been proposed in the literature for vision restoration: (i) repetitive transorbital alternating current stimulation (rtACS); (ii) transpalpebral electrical stimulation (tpES); (iii) transcorneal electrical stimulation (TES); (iv) transcranial random noise stimulation (tRNS); (v) high-frequency tRNS (hf-tRNS); and (vi) transcranial direct current stimulation (tDCS).³³

The first study using NES was conducted by Shinoda et al. in patients with AMD, suggesting an increase in VA after TES.³⁴ Subsequently, other studies have been performed in different aetiologies of vision impairment and with diverse types of NES; Bola et al. conducted rtACS on subjects with optic neuropathy that resulted in improved patient-reported vision-related QoL.³⁵ Plow et al. assessed tDCS combined with VRT in patients with unilateral post-chiasmal VFD following a stroke or brain damage, observing higher improvement in VF and vision-related activities of daily living compared with the VRT and sham groups. However, these differences were not statistically significant.³⁶ Gall et al. compared rtACS with sham stimulation in patients with optic nerve damage; they found a higher improvement in VF of the intervention group and concluded that rtACS is a safe and effective method for partially restoring VFD. Again, none of the differences in functional parameters were statistically significant.³⁷

There has been a rapid increase in the number of publications focusing on this area in recent years; however, the effectiveness and safety of NES to improve VDF have not been systematically assessed and reported. Only two narrative reviews have been published in the medical literature.^32,33 Therefore, this systematic review and meta-analysis (SR/MA) aim to determine the effectiveness and safety of NES versus sham stimulation for patients with vision impairment.

Materials and methods

We report this SR/MA of interventional studies following Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines.³⁸ The protocol is available in PROSPERO ID: CRD42022 329342.

Search strategy

We conducted a comprehensive electronic search including the Cochrane Central Register of Controlled Trials (CENTRAL; searched 26 April 2022) via Ovid, MEDLINE via Ovid (searched 26 April 2022), EMBASE via Elsevier (searched 25 April 2022), and LILACS via BIREME (searched 25 April 2022) using the following MeSH search terms: Optic Nerve Diseases; Macular Degeneration; Retinal Artery Occlusion; Glaucoma; Retinitis Pigmentosa; Optic Neuropathy; Ischaemic; Amblyopia; Vision Disorders; Electric Stimulation Therapy; Electric Stimulation; and Transcranial Direct Current Stimulation with database-specific limiters for clinical trials. We have also searched clinical trial registries for ongoing or recently completed trials on ClinicalTrials.gov (searched 25 April 2022) and the World Health Organization International Clinical Trials Registry Platform (WHO-ICTRP; searched 25 April 2022). To identify any additional articles that met our inclusion criteria, we checked the literature cited in the included studies and searched grey literature in the following electronic resources: (i) GreySource; (ii) Health Management Information Consortium (HMIC); and (iii) The US National Technical Information Service (NTIS). Search strategies are detailed in supplementary materials, Table S1.

Selection criteria

Two independent reviewers (PAN and MDV) screened the titles and abstracts to identify studies for our SR/MA. If a study was deemed potentially relevant, we obtained the full text, and the same reviewers, working in pairs and independently, made the final eligibility decision. Discrepancies were solved with a third reviewer (WOC, ATH, JYN, LKM, or PLC). We contacted the article authors to see if any additional information or clarification was required. The selection process was performed using the Rayyan web tool.³⁹

We included RCTs published between 2000 and 2022, regardless of language or publication status (published, unpublished, or in press), that evaluated the effectiveness of NES on the visual pathway with rtACS, tpES, TES, tRNS, hf-tRNS, or tDCS compared with sham stimulation. Our inclusion criteria were as follows: (i) patients over 18 years of age; (ii) diagnosis of vision impairment due to ophthalmological and neurological diseases at any level of the visual pathway; (iii) sufficient visual fixation capacity; and (iv) residual vision. The exclusion criteria were as follows: (i) refractive errors as the main diagnosis to be treated; (ii) electrical devices such as cardiac pacemakers or deep brain stimulation; (iii) concomitant diagnosis of epilepsy; (iv) diagnosis of any cancer; (vi) consumption of psychotropic medications, such as anticonvulsants, antidepressants, and/or antipsychotics; (vii) studies in which dual intervention was performed with a different type of non-invasive neuromodulation; and (viii) open-label clinical trials.

The main outcomes were the measurement of VA and assessment detection accuracy, foveal threshold, and mean sensitivity as the parameter for VF analysis using perimetry. The secondary outcomes were as follows: (i) neuronal electrical activity of the visual pathway evaluated with EEG or visual evoked potentials; (ii) QoL evaluated with generic or specific tools, such as the 25-item National Eye Institute Visual Function Questionnaire (NEI-VFQ-25); (iii) contrast sensitivity (CS); (iv) reading performance; and (v) safety profile consisting of evaluation of any adverse effect (AE) related to the intervention.

Data extraction

Two paired reviewers independently extracted data using an Excel form (Microsoft Corp., Redmond, WA, USA). The following variables were extracted: author name; year of publication; journal name; corresponding author information; scope; the location where the trial was conducted (city and country); study design; inclusion and exclusion criteria; study population; number of participants; characteristics of the population (age, gender, aetiology of vision loss and comorbidities); type of intervention and comparator; stimulation parameters; main and secondary outcomes; study funding; and conflicts of interest.

Risk of bias assessment

Two review authors (PAN and MDV) independently assessed the risk of bias for each included study using the Cochrane ‘Risk of Bias 2.0’ (RoB 2) tool (version 22 August 2019).⁴⁰ We resolved disagreements by consensus or consulting a third review author (JYN). We contacted the study authors if relevant information was unavailable from publications, trial protocols, published articles, or other sources. RoB 2 evaluates the following domains: bias arising from the randomisation process; bias due to deviations from the intended interventions; bias due to missing outcome data; bias in the measurement of the outcome; and bias in the selection of the reported results. We focus on assessing the effect of the intervention per outcome at a specific time point.

Data analysis

We used Review Manager (RevMan) Version 5.4 for the statistical analysis.⁴¹We calculated the risk ratio (RR) with 95% confidence intervals (CI) for dichotomous outcomes. In the case of continuous outcomes, we extract the mean change from the baseline for each group or the mean post-intervention and standard deviation (SD). We calculated the standardised mean difference (SMD) with 95% CI if outcomes were measured with different scales or the mean difference (MD) with 95% CI if similar scales were used. If outcomes reported MD data, we pooled endpoint and mean change data.⁴²

For VA calculations, we converted the Snellen notation to logMAR.^43,44 In cases where the SD or the mean was not reported, we tried to obtain them by contacting the authors. If we did not receive an answer from the authors, we imputed the missing data with replacement values based on the standard error (SE), CI, p-value, t-value, median, and/or interquartile range (IQR) following the recommendations and guidelines described in the Cochrane Handbook for Systematic Reviews of Interventions.⁴²

We evaluated clinical and methodological heterogeneities considering differences in population, interventions, and outcome measurements based on the overlap of point estimates and degree of CIs and I2. When high heterogeneity is detected, we pool the data using the random effects model using DerSimonian and Laird methods. If we did not detect substantial heterogeneity, we carried out a fixed-effects model using the inverse variance method.⁴⁵ We also assessed statistical heterogeneity with the I2 statistic; values greater than 50% were considered significantly heterogeneous.⁴² For outcomes with significant heterogeneity, we conducted prespecified subgroup analyses with tests for interaction and appraised statistically significant findings using ICEMAN.⁴⁶

Sensitivity analysis

We explore the robustness of the effect by including studies with a high risk of bias versus a low risk of bias.

Certainty of evidence

We use the GRADE framework to assess the certainty of evidence for each outcome in the body of evidence. According to GRADE, RCTs start as high-certainty evidence but can then be rated down to moderate, low, or very low certainty based on study limitations, inconsistency, imprecision, indirectness, and publication bias.⁴⁷ In case we obtained more than 10 studies, the publication bias was assessed by inspecting the effect of small studies through funnel plots. We created tables summarising the findings using GRADEpro GDT.⁴⁸

Results

Search results

Initially, we identified 2,106 records in electronic databases and registers. We also found 72 additional references using other methods: 70 in websites of grey literature and 2 in citation searching. Based on their title and abstracts, 64 studies were selected. After a full-text evaluation, 13 studies reported in 16 records met our inclusion criteria.^{22–26,36,37,49–54} The PRISMA flowchart is shown in Figure 1.

Figure 1.

Preferred reporting items for systematic review search strategy according to the PRISMA 2020 flow diagram³⁸.

Included studies

The main characteristics of each study are summarised in Table 1. Out of 13 studies, involving 441 participants, 92% (n = 12) were parallel RCTs, and one (0.08%, n = 1) was a crossover RCT. For the crossover study, we extracted data outcomes of the last follow-up just before changing the group participants. The NES procedures carried out for the treatment of vision impairment included rtACS,^24,26,49,50 tRNS,⁵¹ tDCS,^22,36,53 hf-tRNS,⁵² TES,^24,49,50,54 and tpES.²⁵ Silja Räty et al. reported three RCTs from different centres.²³ Experiment 1 (Magdeburg) had three groups: (i) active rtACS; (ii) active tDCS with active rtACS; and (iii) sham-tDCS with sham rtACS. Experiment 2 (Helsinki) consisted of the following groups: (i) rtACS and (ii) sham. Experiment 3 (Rome) was composed of (i) tDCS and (ii) sham tDCS. Data from all groups in experiments 2 and 3 were extracted, while only groups I and III were included from experiment 1. Group II was excluded from this SR/MA due to the use of dual stimulation with different techniques of NES, which did not meet the inclusion criteria.

Table 1.

Summary of study characteristics.

Reference	Study design	Sample size (n)	Population				Intervention (n)	Comparison (n)	Outcomes		Last FU, (months)
			Aetiology	Age (years)	Sex (M: F)	Time from onset (months)			Main outcomes	Secondary outcomes
Räty et al., 2021- Exp. 123	Parallel RCT	16	Homonymous hemianopia due to ischaemic or haemorrhagic stroke	Median 58; IQR: 51–66	13:3	>6	rtACS (8)	Sham rtACS (8)	Change in visual field assessed by high resolution and standard perimetries.	Visual acuity, reading speed, contrast sensitivity, dynamic visual acuity, and adverse effects	2
Räty et al., 2021- Exp. 223		18		Median: 59; IQR: 36–67	13:5	>6	rtACS (9)	Sham rtACS (9)			2
Räty et al., 2021- Exp. 323		14		Median: 68; IQR: 60–73	10:4	>6	tDCS (7)	Sham tDCS (7)			2
Donkor et al.51	Parallel RCT	19	Amblyopia	Mean: 42; SD:14.48	10:7	NR	tRNS (9)	Sham tRNS (10)	Visual acuity and contrast sensitivity	Adverse effects	1
de Venecia et al.53	Parallel RCT	22	Proliferative diabetic retinopathy	Mean 55.2; SD: 11.3	13:9	NR	tDCS (11)	Sham tDCS (11)	Visual acuity	Number acuity and accuracy rates	No FUd
Moret et al.52	Parallel RCT	20	Amblyopia	Mean: 44,4; SE: 10,8	8:12	NR	hf-tRNS with perceptual training (10)	Sham hf-tRNS with perceptual training (10)	Contrast sensitivity and Visual acuity	NR	6
Schatz et al.49	Parallel RCT	52	Retinitis pigmentosa	Mean: 46; SD: 15	27:25	NR	TES (32)	Sham TES (20)	Visual field area	Visual acuity, electroretinography, rod, and cone full-field stimulus threshold, IOP, and adverse effects	2.75
Haller et al.54	Parallel RCT	97	NAION, multiple sclerosis, and ocular trauma	Mean: 50.8; SD: 10.9	60:37	>3	TES (65)	Sham TES (32)	Visual acuity	IOP, visual field, optical coherence yomography, quality of life and adverse effects	2
Gall et al.37	Parallel RCT	88	Optic nerve damage due to glaucoma, AION, post-acute inflammation, optic nerve compression, congenital and unknown aetiology of optic atrophy, or Leber’s hereditary optic neuropathy.	Mean: 57.8; SD: 14.2	51:37	NR	rtACS (49)	Sham rtACS (39)	Visual acuity, and change in visual field assessed by high resolution and standard perimetries	EEG power spectra, and adverse effects	2
Spiegel et al.22	Crossover RCT	16	Amblyopia	Mean: 22,5 SD: 4,46	6:10	NR	tDCS (8)	Sham tDCS (8)	Visual acuity and stereopsis	Adverse effects	5 days
Anastassiou et al.25	Exploratory and parallel RCT	22	AMD	Mean: 76.2	NR	NR	tpES (12)	tpES (12)	Visual acuity	Contrast sensitivity, macular sensitivity, fixation stability using microperimetry, and adverse effects	6
Naycheva et al.50	Pilot, Parallel RCT	13	Retinal artery occlusion	Median: 74; IQR: 25–84	4:9	10 days to 17 (median 11)	TES (10)	Sham TES (3)	Visual acuity	Adverse effects, IOP, visual field with perimetry	2.75
Plow et al.36	Pilot	12	Hemianopia and quadrantanopia caused by stroke or surgical trauma	Mean: 59.58; SD 3.47	5:7	39.83 ± 16.16	VRT with tDCS (6)	VRT with sham tDCS (6)	Changes in visual field	Quality of life and adverse effects	6
Sabel et al.26	Parallel RCT	22	Optic nerve damage	Mean: 52.1; SD 15.4	13:9	68 ± 100	rtACS (12)	Sham rtACS (10)	Detection accuracy change in percent over baseline within defective visual field	Visual field with perimetry, EEG, visual acuity, and adverse effects	2
Schatz et al.24	Parallel RCT	24	Retinitis pigmentosa	NR	NR	NR	TES	Sham TES	Visual field area	Visual acuity, electroretinography, rod, and cone full-field stimulus threshold, IOP, and adverse effects	2.75

AION = anterior ischaemic optic neuropathy; AMD = age-related macular degeneration; EEG = electroencephalogram; FU = follow up; IOP = intraocular pressure; IQR – interquartile range; NR = not reported; RCT = randomised control trial; rtACS = repetitive transorbital alternating current stimulation; SD = standard deviation; SE = standard error; TBI = traumatic brain injury; tDCS =tTranscranial direct current stimulation; TES = transcorneal electrical stimulation; tpES = transpalpebral electrical stimulation; tRNS = repetitive visual cortex transcranial random noise stimulation; VFD = visual field defect; VRT = vision restoration therapy.

^aWe excluded the third arm of experiment 1 as they did dual stimulation with cathodal direct current stimulation and alternating current stimulation.

^bThis study has not been published; the results are reported in ClinicalTrials.gov.

^cWe extracted the data since the first follow-up before the change of the group.

^dMeasures were taken immediately before and after the stimulation; there was no further follow-up.

All the studies included male and female participants, with 52% (n = 229) male participants. Across studies that reported participants’ mean or median age, the age ranged from 22.5 to 84 years. A range of ophthalmological and neurological diseases were investigated, including ischaemic or haemorrhagic stroke,^23,36 amblyopia,^22,51,52 proliferative diabetic retinopathy,⁵³ retinitis pigmentosa,^24,49 AMD,²⁵ retinal artery occlusion,⁵⁰ and three studies included subjects with various diagnoses related to optic nerve disease (Table 1).^26,37,54

Table 2 provides the stimulation parameters and specific neurostimulator characteristics for each type of NES and study. Regarding pre-chiasmatic stimulation, the RCTs reported the placement of electrodes in three different structures: (i) eyelids (n = 4, 27%); (ii) cornea (n = 3, 20%); and (iii) periorbital zone (n = 2, 13.3%). Post-chiasmatic stimulation mainly involved the occipital lobes (n = 6, 40%) with tDCS, tRNS, or hf-tRNS. Stimulation parameters, including the number of sessions, time of stimulation per session, total treatment duration, frequency, and amplitude of electric current applied, were, respectively, within the following limits: 6–52 sessions, 5.3–50 min, 5–364 days, 5–30 Hz, and 0.18–1.5 mA for pre-chiasmatic stimulation; and 1–36 sessions, 10–25 min, 1–84 days, 0.1–640 Hz, and 1.5–2.0 mA for post-chiasmatic stimulation. For the RCTs (n = 9, 60%) that applied pre-chiasmatic stimulation,^{23–26,37,49,50,54} the amplitude was determined according to the phosphene threshold, while for the RCTs (n = 6, 40%) that performed post-chiasmatic stimulation the phosphene threshold was not considered to determine it. Additionally, Räty et al., Experiment 1 and Experiment 2, which included patients with homonymous hemianopsia who were stimulated with rtACS, did not provide information regarding whether phosphenes were detected in one or both hemifields.²³

Table 2.

Summary of procedure data of non-invasive neuromodulation for vision restoration.

Reference	Electrode position	Stimulation parameters	Sham parameters	Neurostimulator reference	Time of stimulation (minutes)	No. of sessions (total)	Treatment duration
Pre-chiasmatic stimulation
Räty et al., 2021- Exp. 123	Transpalpebral	● Amplitude: 1.5 mA ● Frequency gradually increased from 5 to 30 Hz in a 48 seg ‘block’	Occasional current bursts (one 5 Hz burst every 1 min) at 100% of the phosphene threshold that induced weak phosphenes to ensure blinding	MC4, NeuroConn GmbH, Ilmenau, Germany	20	10	2 weeks
Räty et al., 2021- Exp. 223	Transpalpebral	● Amplitude: 0.45–1.5 mA ● Frequency: gradually increased from 5 to 15 Hz in a 48 seg ‘block’	Occasional current bursts of 10 pulses (one 5-Hz burst every 5 min) at 100% of the phosphene threshold to induce weak phosphenes to ensure blinding	MC4, NeuroConn GmbH, Ilmenau, Germany)	30 min during treatment days 1–5 and for 40 min during days 6–10	10	2 weeks
Schatz et al.49	Transcorneal in lower eyelid on the ocular surface	● Frequency: 20 Hz ● Current-balanced 5 ms positive immediately followed by 5 ms negative deflections. ● 150% or 200% phosphene threshold levels	NR	OkuStim, OkuSpex, and OkuEl (CE approved; Okuvision GmbH, Reutlingen, Germany)	30	52	52 weeks
Haller et al.54	Transcorneal placed on the subject’s conjunctiva	● Current pulse, biphasic square output with a pulse duration of 1 ms ● Frequency: 20 Hz150% phosphene threshold levels	The same protocol as the intervention without stimulation	Okuvision Stimulation Set manufactured by Okuvision GmbH, Reutlingen, Germany	30	6	6 weeks
Gall et al.37	Transorbital	● Frequencies: 8–25 Hz ● 125% of phosphene threshold level	Minimal dose stimulation	SAFELEADTM, Astro-Med, Inc, USA	50	10	10 days
Anastassiou et al.25	Transpalpebral through eight contact points, four on the upper and four on the lower eye lid	● The current applied varied individually between 150 and 220 µA according to phosphene threshold ● Frequency 5–80 Hz	Electrostimulation but with no flow of current	TheraMacTM: Acuity Medical International, Min- neapolis, U.S.A	5.3b	10	5 days
Naycheva et al.51	Transpalpebral	●10 ms rectangular biphasic current pulses (5 ms positive, directly followed by 5 ms negative) at 20 Hz ● 66% or 150% phosphene threshold levels	Electrostimulation but with no flow of current	TwisterÒ; Dr. Langer GmbH, Waldkirch, Germany	30	6	6 weeks
Sabel et al.26	Transorbital with four stimulation electrodes placed at or near the eyeball	●Amplitude: below 1000 µA adjusted according to patients perceived phosphenes threshold ● Current pulses in firing bursts of 2 to 9 pulses	rtACS with no current stimulation	Ag/AgCl ring electrode, Easycap, Germany – EBS Technologies, Klein- machnow, German	30	10	10 days
Schatz et al., 201126	Transcorneal in lower eyelid on the ocular surface	● Frequency: 20 Hz ● Amplitude: 0.36 to 0.18 mA ● 66% or 150% phosphene threshold levels	NR	OkuStim, OkuSpex, and OkuEl (CE approved; Okuvision GmbH, Reutlingen, Germany)	30	6	6 weeks
Post-chiasmatic stimulation
Räty et al., 2021- Exp. 323	Transcranial in bilateral occipital lobes	● Amplitude: 2 mA	Occasional current bursts (one 5 Hz burst every 1 min) at 100% of the phosphene threshold that induced weak phosphenes to ensure blinding.	MC4, NeuroConn GmbH, Ilmenau, Germany)	20	10	2 weeks
Donkor et al.51	Transcranial in visual cortex	● Amplitude: 2.0 mA ● Frequency range 0.1–640 Hz	The 30 s ramp-up was immediately followed by the ramp-down out	DC-stimulation MC device (Eldith, Neuro- Conn GmbH, Germany)	25	5	5 days
de Venecia, et al.53	Transcranial in V1a	● Amplitude: 1 mA	Stimulation for 30 seg using the same stimulation parameters	DC-STIMULATOR PLUS, NeuroConn Gmbh, Ilmenau, Germany	10	1	1 day
Moret et al.52	Transcranial in occipital cortex, with the centre at ~3 cm above the inion	● Amplitude: Current linearly increased in intensity up to 1.5 mA during the first 30 s of stimulation ● Frequency: 100–600 Hz	The current linearly increased for the first 30 s up to a 1.5 mA and then decreased to 0 mA in the next 30 s	Battery-driven stimulator (BrainSTIM, EMS)	NR	8	2 weeks
Spiegel et al.22	Transcranial on Oz and Cza	● Amplitude: 2 mA	The tDCS electrodes were placed over the motor cortex and stimulation was only delivered for 30s	Chattanooga Ionto DJO International, Guildford, Surrey, UK	15	5	5 days
Plow et al.36	Transcranial on Oz and Cza	● Amplitude: constant current of 2 mA	tDCS with no current stimulation	IOMED Inc, Salt Lake City, Utah	NR	36	12 weeks

ms = millisecond; NR = not reported; rtACS = repetitive transorbital alternating current stimulation; s = seconds; tDCS = transcranial direct current stimulation.

^aFollowing the international 10–20 EEG guidelines.

^bEvery session included eight spots (40 sec/spot) around the eye globe.

Risk of bias within studies

Figure S1, in supplementary materials, shows the RoB assessment. Concerning VA, three RCTs were rated as ‘some concerns’ due to the unavailability of the protocol,^22,25,50 and four RCTs were marked as ‘high risk’ due to the lack of blinding during VA evaluation.^24,51,52 For the VF outcome, three RCTs were rated as having ‘some concerns’ due to the unavailability of the pre-specified analysis plan.^25,26,50 For CS, three RCTs were rated as having ‘some concerns’ due to the unavailability of the protocol.^25,51,52 Concerning adverse effects, four RCTs had ‘some concerns’ about selecting the reported results due to the researchers’ pre-specified intention and the unavailability of measurement and analysis.^25,50–52

Effect of interventions

Visual acuity

Twelve RCTs (n = 429) evaluated VA; seven studies reported findings at 1 month or less of follow-up,^{22,24,26,37,49,50,54} while three studies reported findings for more than 1 month of follow-up.^24,37,50 Of these, 50% (n = 6) reported VA with logMAR notation,^{22,24,37,50,53,54} 25% (n = 3) with Snellen notation,^23,26 8.3% (n = 1) with MNREAD,²³ 8.3% (n = 1) with Early Treatment Diabetic Retinopathy Study (ETDRS) letters number,²⁵ and 8.3% (n = 1) with uncrowded and crowded VA in logMAR based on Landolt-C optotypes.⁵¹

The use of NES in seven RCTs was found to slightly reduce logMAR (i.e. improve VA) at 1 month or less of follow-up (MD = −0.02, 95% CI −0.08 to 0.04, I2 = 73%, n = 312, Figure 2), while at more than 1-month follow-up, NES in three RCTs was reported to increase logMAR (i.e. diminish VA) (MD = 0.01, 95% CI 0.00 to 0.02, I2 = 0%, n = 127, Figure 3). The certainty in the evidence was low and moderate, respectively, owing to unexplained heterogeneity, wide CIs that included important benefits and harm, and a small sample size.

Figure 2.

The forest plot shows the mean visual acuity difference at 1 month or less of follow-up evaluated with logMAR between active stimulation and sham stimulation in each study.

Figure 3.

The forest plot shows the mean difference in visual acuity after 1 month of follow-up evaluated with logMAR between active stimulation and sham stimulation in each study.

Visual field

Three RCTs (n = 158) evaluated the detection accuracy of VF with a high-resolution, super-threshold visual detection test (HRP).^23,26,37 This test has lower test-to-test variability than standard static perimetry due to above-threshold target luminance. Results showed that NES may increase detection accuracy at 1-month follow-up or less (SMD = 0.09, 95% CI −0.58 to 0.77, n = 158, I2 = 0%, Figure 4a); however, at more than 1-month follow-up, NES may reduce the detection accuracy (SMD = −0.22, 95% CI −1.40 to 0.96, n = 136, I2 = 86%, Figure 5a). The certainty in the evidence was rated as moderate and low, respectively, owing to wide CIs, small sample size, and unexplained heterogeneity (see Table S2 in supplementary materials). Regarding conventional perimetry, five RCTs (n = 232) addressed the foveal threshold and mean sensitivity as a mean VF parameter.^{23,26,37,49,50} The pooled effect suggests that NES likely reduces the foveal threshold at 1 month follow-up or less (SMD = −0.07, 95% CI −0.47 to 0.33, I2 = 6% n = 110, Figure 4b), with low certainty of evidence due to small sample size and wide CIs. Similarly, NES may reduce mean sensitivity at 1 month follow-up or less (SMD = −0.03, 95% CI −0.53 to 0.48, n = 85, I2 = 0%, Figure 4c) and at more than 1 month follow-up (SMD = −0.33, 95% CI −0.51 to −0.14, n = 85, I2 = 0%, Figure 5b). The certainty in the evidence for NES compared with sham stimulation on mean sensitivity is moderate due to the small sample size and wide CIs, which indicate serious imprecision (See Table S2 in supplementary materials).

Figure 4.

Forest plots of visual field evaluation with perimetry at 1 month or less of follow-up. (a) The forest plot shows the standardised mean difference in detection accuracy between active stimulation and sham stimulation in each study. (b) The forest plot shows the standardised mean difference of the foveal threshold between active and sham stimulation in each study. (c) The forest plot shows the standardised mean difference of mean sensitivity between active and sham stimulation in each study.

Figure 5.

Forest plots of visual field evaluation with perimetry after 1 month of follow-up. (a) The forest plot shows the standardised mean difference in detection accuracy between active stimulation and sham stimulation in each study. (b) The forest plot shows the standardised mean difference of mean sensitivity between active stimulation and sham stimulation in each study.

Contrast sensitivity

Four RCTs (n = 109) evaluated CS using the Mars Letter CS test and logMAR,²³ ETDRS, and Pelli-Robson CS chart,^25,52 and Landolt-C optotypes⁵¹ before and after stimulation. The main findings on CS improvement can be summarised as follows: (i) Räty et al. reported that rtACS and tDCS did not differ significantly in improving CS²³; (ii) Moret et al. found similar and significant improvements in CS with both active stimulation (31.2% improvement) and sham stimulation (49.9% improvement)⁵²; (iii) Anastassious et al. observed rapid improvement (mean = +4.4 optotypes) after 1 week of follow-up with active stimulation, but CS declined (mean = +1.5 optotypes) after 6 months. No significant differences were observed over time in the sham stimulation group²⁵; and (iv) Donkor et al.⁵¹ reported findings similar to Anastassious et al.,²⁵ with significant improvement observed after 1 day of active stimulation and no significant effect observed after 2–5 days of follow-up.

Reading performance

One RCT (n = 48) assessed the effect of tDCS and rtACS on reading performance in patients with homonymous hemianopia due to ischaemic or haemorrhagic stroke.²³ They utilised an international reading speed test and found no significant difference between the two stimulation methods in terms of improving reading performance (Experiment 2: MD = 4.6, 95% CI −6.0 to −13.4, p = .489; Experiment 3: MD = −1.5, 95% CI −15.0 to −6.5, p = .710).

Neuronal electrical activity of the visual pathway

Two RCTs (n = 110) evaluated the effect of stimulation on the electrical activity of the visual pathway using EEG signals at occipital areas (O1, O2).^26,37 The power spectra at O1 and O2 were estimated using Fast Fourier Transformation. Gall et al. evaluated five spectral bands, including delta, 1–3 Hz; theta, 3–7 Hz; alpha, 7–14 Hz; and beta, 14–30 Hz; and found that the alpha and beta power increases were significantly greater after 10 days of rtACS (Δ = 3.7%) than after sham stimulation (Δ = 1.5%).³⁷ On the other hand, Sabel et al. only recorded alpha-rhythm in the 7.513 Hz frequency range and reported that ACS increased alpha power after 10 days of stimulation at O1 (Δ = 11.5%) and O2 (Δ = 30.4%). No changes were observed with sham stimulation.^26,37 No studies reporting visual evoked potential were found.

Quality of life

Three RCTs (n = 151) assessed the impact of NES versus sham stimulation on health-related QoL, as measured by the NEI-VFQ-25^54,55 and the Veterans Affairs Low-Vision Visual Functional Questionnaire (VA LV VFQ).³⁶ One study involving 97 subjects with optic nerve disease reported an estimated mean change from a baseline of 29.3 in the NES group and 15.4 in the sham group, resulting in a between-group SMD of −0.16 (95% CI −0.59 to 0.26, Figure 6).⁵⁴ In contrast, the other two studies reported SMDs of 0.44 (95% CI, −0.18 to 1.06) and 0.69 (95% CI, −0.49 to 1.87) between the NES and sham groups, respectively.^36,37 The corresponding pooled SMD among patients receiving the intervention compared with sham stimulation was 0.08 (95% CI −0.25 to 0.42, n = 151, I2 = 44%, Figure 6). The evidence was assessed as having moderate certainty due to imprecision resulting from a small sample size and wide CIs that included both important benefits and harms (see Table S2 in supplementary materials).

Figure 6.

The forest plot shows the standardised mean difference in the quality of life after stimulation between active stimulation and sham stimulation in each study.

Adverse effects

Ten RCTs (n = 274) addressed AEs.^{22–25,36,37,49–51} Three RCTs (n = 50) did not report any AEs during the study period,^22,25,36 and no studies reported serious adverse events. Compared with sham stimulation, NES increased the risk of minor AEs (RR 1.24, 95% CI 0.99 to 1.54, n = 224, I2 = 0%, Figure 7). However, all AEs were transient, including foreign body sensations, skin irritation, vertigo, dizziness, and mild headache. The evidence was assessed as having moderate certainty due to imprecision resulting from a small sample (see Table S2 in supplementary materials). The GRADE summary of the findings is presented in Table 3.

Figure 7.

Comparison of the relative risk of adverse effects with active and sham stimulation.

Table 3.

Summary of findings table of non-invasive neurostimulation compared to sham stimulation for vision impairment.

Outcomes	No. of participants (studies) Follow-up	Certainty of the evidence (GRADE)	Anticipated absolute effects
			Risk with sham stimulation	Risk difference with non-invasive neurostimulation
Visual acuity at ≤1 month FU assessed with: logMAR	312 (7 RCTs)	⊕⊕◯◯ Lowa,b	The mean was 0.2	MD 0.02 lower (0.08 lower to 0.04 higher)
Visual acuity at >1 month FU assessed with: logMAR	127 (3 RCTs)	⊕⊕⊕◯ Moderatec	The mean was 0.4	MD 0.01 higher (0 to 0.02 higher)
Detection accuracy at ≤1 month FU assessed with: high resolution perimetry	158 (5 RCTs)	⊕⊕⊕◯ Moderateb	The mean was 0.9	SMD 0.09 higher (0.58 lower to 0.77 higher)
Detection accuracy at >1 month FU assessed with: high resolution perimetry	136 (4 RCTs)	⊕⊕◯◯ Lowa, b	The mean was 0.2	SMD 0.22 lower (1.4 lower to 0.96 higher)
Foveal threshold at ≤1 month FU assessed with: perimetry	110 (2 RCTs)	⊕⊕◯◯ Lowa,d	The mean was −0.75	SMD 0.07 lower (0.47 lower to 0.33 higher)
Mean sensitivity at ≤1 month FU assessed with: perimetry	69 (4 RCTs)	⊕⊕⊕◯ Moderateb	The mean was 3.1	SMD 0.03 lower (0.53 lower to 0.48 higher)
Mean sensitivity at >1 month FU assessed with: perimetry	85 (5 RCTs)	⊕⊕⊕◯ Moderateb	The mean was 3.1	MD 0.33 lower (0.51 lower to 0.14 lower)
Safety assessed with: no. of adverse effects	224 (7 RCTs)	⊕⊕⊕◯ Moderatee	289 per 1,000	69 more per 1,000 (3 fewer to 156 more)
QoL	151 (3 RCTs)	⊕⊕⊕◯ Moderateb	The mean was 9.5	SMD 0.08 higher (0.25 lower to 0.42 higher)

CI = confidence interval; FU = follow up; MD = mean difference; QoL = quality of life; RR = risk ratio; SMD = standardised mean difference.

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

GRADE Working Group grades of evidence:

High certainty: we are very confident that the true effect lies close to that of the estimate of the effect.

Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.

Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect.

Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

Explanations.

^aUnexplained significant statistical heterogeneity was indicated.

^bDowngraded once due to we noted wide CIs that included important benefit and harm (all instances of note) and small simple size.

^cDowngraded once due to small simple size.

^dDowngrade the quality of evidence for this outcome by two levels due to wide CIs and small simple size.

^eDowngraded once due to few events.

Subgroup and sensitivity analyses

The results related to VA were further validated by an exploratory subgroup analysis, which revealed that the heterogeneity observed for VA at less than 1 month follow-up and at 1 month follow-up could not be explained by the differences in the type of intervention or aetiology (Figure 2). Moreover, the sensitivity analysis demonstrated that there was no effect modification on the pooled outcome when studies with a high risk of bias were excluded (see Figure S2 in supplementary materials).

Discussion

‘Visual neurorehabilitation’ refers to the complex neurophysiological processes involved in activating residual visual function after being affected by neurological diseases. It has been suggested that a necessary condition for this type of rehabilitation is the existence of remaining vision.⁵⁶ Furthermore, complete rehabilitation of visual function is rarely achieved, reflecting the narrow limits of neural plasticity and functional recovery after visual pathway damage or our limited knowledge, which constrains the design of new treatment methods.⁵⁷ Nonetheless, compensatory improvements in functionally relevant visual skills for daily living, such as visual search and spatial orientation, reading, and fusion, can be achieved in most visually impaired patients using VRT, indicating some potential for recovery, development, and relearning of visuomotor strategies.⁵⁸ An additional alternative, NES, which involves electrical current delivery, has been suggested as a novel treatment for vision rehabilitation after retinal or visual pathway diseases.^{22–26,36,37,49–54}

In our SR/MA, we found a significant clinical heterogeneity regarding the magnitude of visual impairment, aetiologies (including multiple ophthalmological and neurological diseases), neuromodulation techniques, stimulation parameters, duration of follow-up, outcomes used to measure efficacy, and the way outcome data were reported. As a result, pooling the data to address the question of whether NES is an effective treatment for vision restoration in adults poses a challenge, despite being able to perform a meta-analysis.

This SR/MA of 13 RCTs involving 441 patients with vision impairment resulted in the following findings: (i) low-certainty evidence that NES may improve VA at 1 month or less of follow-up; (ii) moderate-certainty evidence suggesting that NES increases QoL and detection accuracy at 1 month or less of follow-up; (iii) the impact of NES on mean sensitivity likely results in little or no difference; (iv) the effect of NES in the foveal threshold, CS, reading performance, and EEG remains uncertain; and (v) about safety, NES for the treatment of vision impairment is considered safe based on the low risk of serious AEs and the low level of concern for reported minor AEs.

Comparing outcomes across different NES techniques, which might potentially result in different effects, becomes challenging. However, various NES modalities, such as transcorneal, transorbital, and transpalpebral, may be deemed more suitable for addressing macular and optic nerve lesions,^{25,26,37,49,50} while transcranial stimulation may be more appropriate for the treatment of cortical lesions (refer to subgroup analysis in Figure S2 in the supplementary materials).^23,36

The current intensity of stimulation is one of the parameters that exhibited the most heterogeneity among the studies, possibly because in the majority of them (although not in all), the amplitude was related to the phosphene threshold. For instance, Anastassiou et al. reported that the current applied varied individually between 150 and 220 µA,²⁵ whereas Sabel et al. applied up to 1000 µA current depending on how well the patients perceived the phosphenes.²⁶ Additionally, the total treatment duration also showed significant heterogeneity. Schatz et al. performed TES for retinitis pigmentosa in a total of 52 sessions across 364 days.⁴⁹ In contrast, de Venecia et al. carried out only one session of tDCS stimulation for proliferative diabetic retinopathy over 1 day.⁵³

Most of the included studies had a follow-up period of less than 1 month after the stimulation protocol ended, which precluded the evaluation of the long-term results of neuromodulation. Electric current-induced effects in the central nervous system can be seen up to a year after invasive or non-invasive methods of stimulation, according to studies performed on non-neuro-ophthalmic diseases.^59,60

Additionally, the evaluation of VA and VF was conducted using various notations for the former and multiple types of perimetry (such as HRP, static, and kinetic, among others) for the latter, resulting in difficulties when interpreting the outcomes. These tests are psychophysical, so they have a certain subjectivity that limits the interpretation and quantification of outcomes.⁶¹ An alternative to assess visual function more objectively may be the neuronal electrical activity of the visual pathway evaluated with EEG or the evoked visual potential, which measures the electrical signal generated at the visual cortex in response to visual stimulation, taking into account that the visual cortex is primarily activated by the central VF and there is a large representation of the macula in the occipital cortex.^62,63 As we mentioned, two RCTs evaluated EEG analyses of functional brain networks as a marker for visual function recovery.^26,37 However, whether this approach is preferable to the use of evoked visual potentials is still debatable. The neuro-ophthalmologists of this SR/MA (LKM and PLC) suggested that visual evoked potentials have more sensitivity and specificity for diagnosis and monitoring of vision impairment.

This SR/MA has several strengths. Firstly, it is the first comprehensive, evidence-based analysis of the reported clinical efficacy and safety of the different types of NES, which could lead to therapeutic applications in the visual domain. Secondly, the study used structured GRADE and Cochrane methods to incorporate multidisciplinary perspectives. Finally, it included performing a systematic search to synthesise 13 RCTs with more than 400 patients.

However, there are potential limitations of this study. (i) Not all populations had the same aetiology and severity of vision impairment. We address this issue using the changes from the baseline, which is a common measure included in RCTs. (ii) The data are sparse for most outcomes resulting in wide CIs, which we considered as serious imprecision. Additionally, we could not evaluate publication bias; as a result, it may not have been appropriate to create funnel plots for meta-analyses with fewer than 10 studies.⁴⁰ We stress the need for conducting future RCTs to provide more robust evidence. (iii) There were significant clinical differences in follow-up. We address this by grouping studies by those that carried out follow-ups at 1 month or less and those that carried out follow-ups at more than 1 month. However, future studies must standardise both the length and time points of follow-up. (iv) No study assessed the potential long-term neuromodulation effects of NES in the visual pathway.

Regarding future research, as of 25 April 2022, we found six ongoing clinical trials registered with www.clinicaltrials.gov (Table 4). These studies focus on the efficacy of different electrostimulation modalities: transcorneal; transorbital; transpalpebral; and transretinal; in treating ophthalmological diseases of interest, such as amblyopia, optic neuropathy, and AMD. We hope that the results of these studies will help us understand the long-term potential therapeutic effects of these modalities in treating vision impairment, as well as provide more information on the most effective stimulation protocol and methods of objective evaluation of the outcomes based on each vision loss aetiology.

Table 4.

Ongoing clinical trials.

Study	Author	Place	Status	Clinical trial ID
Transcorneal Electrical Stimulation (TES) for the Treatment of Amblyopia	Harold P. Koller,	Wills Eye Hospital	Unknown	NCT02495935
Effect of Transorbital Electrical Stimulation of Optic Nerve on Remyelination After an Acute Optic Neuritis (ONSTIM)	Hayet Serhane	Centre Hospitalier National d’Ophtalmologie des Quinze-Vingts	Unknown	NCT04042363
Transpalpebral Micro-Current Electrical Stimulation for the Treatment of Dry Age-Related Macular Degeneration	Kevin Parkinson	NR	Completed, no results posted	NCT02540148
Optimal Non-invasive Brain Stimulation for Peripheral Vision	Allen MY Cheong	The Hong Kong Polytechnic University	Recruiting	NCT04846140
Efficacy of Electrical Micro-current Retinal Stimulation for Treatment of Dry Age-related Macular Degeneration	Richard Beauchemin	NR	Completed, no results posted	NCT01600300
Paraorbital-Occipital Electric Stimulation in Patients with Optic Neuropathy (BCT_optnerve)	Bernhard A. Sabel,	University of Magdeburg	Completed, no results posted	NCT01282827

ID = identification; NR = not reported.

Although we discovered a wide variety of non-invasive neuromodulation techniques, it is necessary to continue investigating new treatment alternatives for vision impairment. Globally, almost 1.1 billion people are living with vision loss due to different causes, and particularly, aetiologies related to diseases of the visual pathway have the worst prognosis, which significantly affects QoL and functionality.^4,64 Other possible treatment and research alternatives include invasive neuromodulation using neurosurgical approaches to target specific areas such as the optic nerve, lateral geniculate ganglion, and visual cortex, depending on the aetiology of the vision impairment.^65,66

Funding Statement

The authors reported that there is no funding associated with the work featured in this article.

Disclosure statement

No potential conflict of interest was reported by the authors.

Ethical approval statement

The study was designed in accordance with the Declaration of Helsinki and received approval from the Andes Ethics Committee (Evaluation Report Number: 20220407).

Supplementary data

Supplemental data for this article can be accessed online at https://doi.org/10.1080/01658107.2023.2279092.