Neuromodulation Techniques in Phantom Limb Pain: A Systematic Review and Meta-analysis

Abstract

Objective

To evaluate the effects of neuromodulation techniques in adults with phantom limb pain (PLP).

Methods

A systematic search was performed, comprising randomized controlled trials (RCTs) and quasi-experimental (QE) studies that were published from database inception to February 2019 and that measured the effects of neuromodulation in adults with PLP. Hedge’s g effect size (ES) and 95% confidence intervals were calculated, and random-effects meta-analyses were performed.

Results

Fourteen studies (nine RCTs and five QE noncontrolled studies) were included. The meta-analysis of RCTs showed significant effects for i) excitatory primary motor cortex (M1) stimulation in reducing pain after stimulation (ES = −1.36, 95% confidence interval [CI] = −2.26 to −0.45); ii) anodal M1 transcranial direct current stimulation (tDCS) in lowering pain after stimulation (ES = −1.50, 95% CI = −2.05 to 0.95), and one-week follow-up (ES = −1.04, 95% CI = −1.64 to 0.45). The meta-analysis of noncontrolled QE studies demonstrated a high rate of pain reduction after stimulation with transcutaneous electrical nerve stimulation (rate = 67%, 95% CI = 60% to 73%) and at one-year follow-up with deep brain stimulation (rate = 73%, 95% CI = 63% to 82%).

Conclusions

The evidence from RCTs suggests that excitatory M1 stimulation—specifically, anodal M1 tDCS—has a significant short-term effect in reducing pain scale scores in PLP. Various neuromodulation techniques appear to have a significant and positive impact on PLP, but due to the limited amount of data, it is not possible to draw more definite conclusions.

Introduction

Phantom limb pain (PLP) is a chronic pain condition that is perceived in a limb that no longer exists [1]. PLP is a serious public health problem that can affect the physical, psychological, and functional health of amputees [2]. After limb amputation, 50–80% of patients experience PLP [2–6], which deteriorates their quality of life [3,7,8]. The occurrence of long-term PLP can lead to comorbidities, such as depression, sleep disturbances, and substance abuse, and affect the use of prostheses [5,9,10].

The current treatment options (pharmacological and behavioral) for PLP are not entirely effective [11]. The pathophysiology of PLP is related to cortical reorganization [11,12]. Nevertheless, no existing PLP treatment targets these maladaptive brain modifications.

Neuromodulation is a potential treatment option for chronic pain [13,14], which could alter maladaptive neuroplasticity and enhance descending inhibitory pathways [15,16]. Such methods include invasive central (deep brain stimulation [DBS], motor cortex stimulation [MCS], and spinal cord stimulation [SCS]) and peripheral (dorsal root ganglion stimulation [DRGS]) interventions and noninvasive central (transcranial magnetic stimulation [rTMS], transcranial direct current stimulation [tDCS]) and peripheral (transcutaneous electrical nerve stimulation [TENS], neuromuscular electrical stimulation [NMES], and peripheral nerve stimulation [PNS]) techniques.

Systematic reviews have attempted to evaluate these options but have encountered several limitations: They analyzed the efficacy of noninvasive brain stimulation without comparing invasive or peripheral techniques [17,18] focused on chronic pain conditions in general, including PLP patients, but failed to target PLP specifically [17]. Therefore, the efficacy of neuromodulation in PLP and the ideal option for treating PLP are unknown.

This study systematically reviewed the evidence on the efficacy and safety of neuromodulation techniques in the treatment of PLP to provide initial data on the type of neuromodulation that has significant effect sizes for further research and future clinical applications.

Methods

Study Design

We performed a systematic review. The study protocol was registered at PROSPERO, number CRD42018117998. This study follows the Cochrane handbook [19] and PRISMA guidelines [20] (Supplementary Data).

Eligibility Criteria

Our inclusion criteria were as follows: i) studies that evaluated any beneficial or adverse effect of the use of neuromodulation techniques in the treatment of adults (>18 years old) with a PLP diagnosis. We included i) invasive central nervous system (CNS) stimulation (DBS, MCS, and SCS), noninvasive CNS stimulation (rTMS and tDCS), and peripheral nervous system stimulation—TENS, PNS, and NMES; ii) randomized controlled trials (RCTs), including parallel-group and crossover designs, and quasi-experimental (QE) studies, including noncontrolled and nonrandomized studies; iii) studies with pain scales (visual analog scale [VAS], numeric rating scale [NRS], McGill Pain Questionnaire, or universal pain score [UPS]) as the primary outcome (and this information had to include mean and standard deviation before and after the intervention); iv) without language restriction. We excluded case–control studies, cohort studies, case series, review articles, conference abstracts, case reports, letters, and editorials. We excluded the studies with full text not available after trying to contact the authors. The most recent or the largest sample size publication was included when the authors published several studies using the same database.

Literature Search and Study Selection

We performed a literature search using five databases: Medline, Central Cochrane Library, Embase, Scopus, and Web of Science. The last updating search was run in February 2019. No additional filters (e.g., language or publication year) were set. The complete search strategy is available in the Supplementary Data. A manual search was also conducted to find other potential articles based on identified references in the individual articles.

Before titles and abstract screening, two reviewers (XM and KP-B) agreed on a standard approach. Two random samples of 50 search results were selected for training purposes. Reviewers screened these titles and abstracts, and inter-rater agreement and kappa estimator were computed, aiming for an inter-rater agreement of at least 90% (Supplementary Data). After this standardization process, duplicate records were removed, and two reviewers (XM and KP-B) screened all titles and abstracts following the prespecified framework and selection criteria. Discrepancies were resolved by a third reviewer independently (FF). After the title and abstract selection, full texts of selected reports were sought and analyzed; differences were resolved by consensus between all authors. These selection processes—titles and abstracts as well as full texts—were conducted using Endnote software. The complete list of excluded articles and the reasons for exclusion at this full-text stage is available in the Supplementary Data.

Data Extraction

Two authors systematically extracted data from each included study using a standardized form. The form used for data extraction documented the most relevant items, including sample characteristics, study design, and intervention characteristics. The PLP intensity (visual analog scale), quality of life, anxiety and depression scales, and adverse events were extracted to evaluate the efficacy and safety of the use of neuromodulation techniques.

Risk of Bias Assessment

To assess the risk of bias (RoB) of RCTs, we used the Jadad Scale [21] following the standard score system from 0 to 5 (high number means low risk of bias). To assess the risk of bias of QE studies, we used the Methodological Index for Non-Randomized Studies (MINORS) tool [22]. For QE controlled studies, the MINORS tool was scored from 0 to 2; 0 for not reported information, 1 for inadequately reported information, and 2 for well-reported information. We considered a low risk of bias when the information was well reported, a high risk of bias when it was not reported, and an unclear risk of bias when it was reported inadequately [22]. The RoB evaluation was assessed by two reviewers (XM and KP-B), and discrepancies were solved by a third reviewer independently (FF).

Data Synthesis

First, we reported the articles in a narrative approach. We decided to present results separately according to study design (RCTs vs QE studies), given the differences in the quality of evidence between these two types of studies.

Interventions were categorized into one of four groups: i) excitatory noninvasive CNS stimulation, ii) inhibitory noninvasive CNS stimulation, iii) invasive CNS stimulation, iv) PNS stimulation. Pain outcome was categorized according to time, based on a previous study [17], in three groups: i) short term (end of intervention or less than one week), ii) medium term (more than one week to less than six weeks), and iv) long term (more than six weeks).

Then, with the RCT data, we performed an exploratory meta-analysis using Hedge’s g for pain intensity. Although within the treatment categories are interventions with different parameters, we decided to do an exploratory synthesis to compare across the spectrum of the available neuromodulation techniques. When possible, we used pre and post scores of the pain analog scales for each outcome to calculate the mean difference between groups. The difference was then converted to an effect size (ES). Given that Cohen’s d has a slight bias to overestimate in small sample sizes, we adjusted Cohen’s d to Hedge’s g by applying a correction factor.

With the QE data, we pooled the rate of pain reduction per individual study by implementing a proportions meta-analysis approach. Briefly, we estimated 95% confidence intervals (CIs) using the exact binomial method and incorporated the Freeman-Tukey double arcsine transformation for computation of pooled proportions [23].

We assessed heterogeneity using the I² statistic, and we considered low heterogeneity I² < 40% [24]. We considered it appropriate to use random-effects models due to the overall heterogeneity of the evaluation (in population and intervention) [25]. Metaregression and publication bias were not assessed, as the number of studies pooled for each meta-analysis was <10 [26]. The data were processed using Stata, version 15.0.

Management of Missing Data

When the primary outcome data (i.e., PLP visual analog scale) were missing or unclear, we contacted the authors. We used Web Plot Digitizer, version 3.11, to extract data from relevant graphs. If we could not reach the authors or obtain the data graphically, we excluded the study from the quantitative analysis.

Results

Search Results

The search retrieved 1,573 results. After removing duplicates, 865 titles and abstracts were screened, and of these, 776 were excluded. Seventy-two studies were evaluated in full text, and 14 were included (Figure 1) [27–40].

Figure 1.

Prism flowchart (study selection).

Studies Characteristics

Regarding the included studies, nine were RCTs [27–31,33,35,36,38], including five crossover RCTs [28–30,33,36], and five were noncontrolled QE studies [32,34,37,39,40]. We included 236 PLP patients in this review. The number of participants ranged from eight to 54 in RCTs, and from three to 10 in QE studies.

Regarding the interventions, they consisted of invasive or noninvasive neuromodulation techniques:

• Three studies used tDCS: All were RCTs and used anodal primary motor cortex (M1)/cathodal supraorbital [29,30,33]. Also, one used anodal posterior parietal cortex (PPC)/cathodal supraorbital as a part of the experimental arms [30].

• Three studies used rTMS: All were RCTs and used excitatory frequencies (20 Hz, 10 Hz, 5 Hz) with intensities from 80% to 95% of the motor threshold [27,28,31]. Additionally, one used an inhibitory frequency (1 Hz) [28]. All of the studies targeted the primary motor cortex (M1).

• Two studies used DBS: Both were noncontrolled QE studies. One used contralateral ventroposterior-lateral nucleus (VPL) stimulation [34], and the other used contralateral periventricular gray stimulation [39].

• Four studies used TENS: Two were RCTs [36,38], and two were noncontrolled QE studies [37,40]. Two used stimulation in the PLP location on the nonamputated limb [37,38]; one study used the stump location [40], and the other used auricular TENS stimulation [36].

• One QE study used PNS (femoral nerve and sciatic nerve trunk stimulation on the amputee side) [32], and one QE study used NMES (stimulation of quadriceps muscles of both legs) [35].

Intervention Characteristics

Interventions were heterogeneous in terms of stimulation parameters (area of stimulation, current intensity, pulse frequency, duration of the session, frequency, and the number of sessions) (Table 1). Regarding the control group, most of the studies used sham stimulation as a comparator, and the other few studies used, respectively, 2-Hz TMS [28], mirror therapy [38], and placebo (TENS device off) [36]. One study reported the use of phantom hand movements during brain stimulation (anodal tDCS) [33]; no other intervention combinations in the active group were reported.

Table 1.

Study characteristics of individual studies

Author Year (Country)	Design	Population, Age, Sex, Cause of Amputation	No. of Patients	Intervention	Stimulation Location and Procedure	Control	Outcome (VAS or NRS) and P Value
Ahmed 2011 (Egypt)	RCT	Outpatients Age: I: 52 ± 12.7 y, C: 53 ± 13.3 y Sex: I: M = 13 (76.5%), F = 4 (23.5%), C: M = 6 (60%), F = 4 (40%) Cause: I: traumatic = 2 (11.8%), ischemic = 6 (35.3%), diabetic = 17 (100%), C: traumatic = 4 (40%), ischemic = 2 (20%), diabetic = 10 (100%)	I: 17, C: 10	rTMS: High frequency: 20 Hz, 10-sec trains (200 pulses) every minute, intensity of stimulation: 80% of resting motor threshold, 10-min sessions daily for 5 consecutive d	Optimal scalp position determined from where transcranial magnetic stimulation evoked motor potentials of maximum peak-to-peak amplitude in the muscle proximal to the stump	Sham stimulation: coil elevated and angled away from the head	I = 3.4 ± 1.2, C = 7.4 ± 0.84 (P = 0.001)
Malavera 2016 (Colombia)	RCT	Outpatients Age: I: 33.1 ± 6.6 y, C: 34.7 ± 9.9 y Sex: I: M = 25, F = 2, C: M = 25, F = 2 Cause: traumatic	I: 27, C: 27	rTMS: Frequency: 10 Hz, 20-min sessions, 20 trains of 6 sec (54-sec intertrain interval), intensity of stimulation: 90% of motor threshold, daily sessions for 10 d during a 2-wk period	M1 contralateral to the amputated leg (corresponding to the first dorsal interosseous muscle of the hand contralateral to pain)	Sham stimulation (sham coil)	I = 2.28 ± 2.51, C = 3.71 ± 2.97 (P = 0.03)
Irlbacher 2006 (Germany)	RCT – crossover	Outpatients Age: 46.6 ± 12.6 y Sex: M = 8, F = 6 Cause: not mentioned	14	rTMS: Frequency: rTMS 1 Hz, 5 Hz, 500 pulses per session, intensity of stimulation: 95%, daily sessions for 5 consecutive d	M1 area corresponding to affected phantom limb. Optimal placement defined by maximal motor response	Sham stimulation: 2 Hz, identical placement of coil that looks and sounds identical and produces same scalp sensation	I (1Hz) = 4.22 ± 2.79, I (5Hz) = 4.99 ± 2.33, C (2Hz) = 4.37 ± 2.97 (P = 0.02)
Bolognini 2013 (Italy)	RCT – double-blind, sham-controlled crossover	Outpatients Age: 35–81 y Sex: M = 6, F = 2 Cause: ischemic = 5, traumatic = 2, cancer = 1	8	Trial 1: anodal tDCS, frequency NR, 15-min sessions, intensity: 2 mA Trial 2: anodal tDCS, cathodal tDCS, frequency NR, 15-min sessions, intensity: 2 mA	Trial 1: M1 Anodal electrode placed over C3 or C4 to target hemisphere contralateral to amputation, cathode electrode placed over contralateral supraorbital area Trial 2: PPC Hemisphere contralateral to amputation, active electrode placed over P3 or P4, reference electrode placed over contralateral supraorbital area	Trial 1: Sham stimulation (stimulator turned off after 30 sec) Trial 2: Sham stimulation (stimulator turned off after 30 sec)	I = 0.8 (−69%), C = 2.6 (−21%, P = 0.02)
Bolognini 2015 (Italy)	RCT – double-blind, sham-controlled crossover	Outpatients Age: 22–76 y Sex: M = 3, F = 5 Cause: ischemic = 6, traumatic = 2	8	Anodal tDCS, 15-min session, ramping period of 10 sec at beginning and end, intensity of 1.5 mA, 5 consecutive d	M1 Anodal electrode placed over C3 or C4 to stimulate M1 contralateral to the amputation, cathode electrode over the contralateral supraorbital area	Sham stimulation (current lasted for 30 sec)	I = 3.3 (−41%), C = 4.7 (−17%, P = 0.04)
Kikkert 2018 (UK)	RCT – double-blind sham-controlled crossover	Outpatients Age: 47 ± 3 y Sex: M = 11, F = 6 Cause: ischemic = 1, traumatic = 13, cancer = 1	8	tDCS (anodal, amputation contralateral M1, cathodal, amputation ipsilateral supraorbital area) 3 sessions, active 1 session, 20 min, 1mA	M1 Electrode placement followed standard procedures and was determined according to the EEG 10-20 system	Sham:The stimulator was turned off after the impedance was stable (after ∼30 sec)	I = 1.73 ± 3.05 (−29.5%), C = 2.50 ± 2.56 (1.2%, P = 0.01)
Katz 1991 (Canada)	RCT – controlled crossover	Outpatients Age: 52.8 (23–73) y Sex: M = 18, F = 10 Cause: peripheral vascular disease (including diabetes mellitus) = 12, accident = 9, arterial thrombosis in 3, tumor = 2, radiation damage = 1, polio = 1	11/28	TENS, 30 min (each session was divided into three consecutive 10-min periods, including an initial resting baseline [B1], bilateral ear stimulation [BES], and a final resting baseline [B2]). Throughout the 30-min session, the subject monitored changes in (painful and/or nonpainful) phantom limb intensity by turning the dial, 1 d	Outer ears (auricular, bilateral ear stimulation)	Placebo	I = 5.1 ± 1.05, C = 9.0 ± 2.16 (P < 0.01)
Tilak 2016 (India)	RCT	Outpatients Age: I = 36.38 ± 9.55 y, C = 42.62 ± 10.69 y Sex: I: M = 11 (84.6%), F = 2 (15.4%), C: M = 12 (92.3%), F = 1 (7.7%) Cause: not mentioned	26	TENS, (contralateral leg) 20 min, one session of TENS for 4 consecutive d	Contralateral leg at the site exactly where they have PLP on the amputated leg	Mirror therapy	I = 2.46 ± 1.561, C = 2.08 ± 1.621 (P = 0.003)
Brede 2017 (USA)	RCT	Outpatients Age: I: 25.7 ± 5.9 y, C: 26.1 ± 5.9 y Sex: I: M = 22 (96%), F = 1 (4%), C: M = 21 (100%), F = 0 (0%) Cause: Blast injuries from improvised explosive devices and landmines were the most common source of trauma	44	MARP + NMES (15 min/session on 5 days/wk) to stimulate contractions for 12 wk	Quadriceps muscles of both legs	MARP only	I = 0.8 ± 0.7, C = 1.6 ± 0.7 (P = 0.005)
Mulvey 2013 (UK)	QE	Outpatients Age: 22–72 (54 ± 13.8) y Sex: M = 4, F = 6 Cause: vascular = 4, trauma = 5, BKN reamputation = 1	10	TENS (60 min) one time	Above stump = 8, on stump = 2	–	−1.8 ± 1.6 (P < 0.05)
Kawamura 1997 (Japan)	QE	Outpatients Age: 22–72 (54 ± 13.8) y Sex: M = 4, F = 6 Cause: vascular = 4, trauma = 5, BKN reamputation = 1	10	Arteriosclerosis obliterans, malignant tumor, or trauma	The sites of the contralateral limb that exactly corresponded to the sites of the amputated limb where the patients felt	–	−1.2 ± 1.9 (P < 0.001)
Rauck 2014 (USA)	QE	Outpatients Age: 47 ± 12 y Sex: M = 14 (88%), F = 2 (12%) Cause trauma = 11 (69%), vascular disease or dysfunction = 4 (25%), cancer = 1 (6%)	9/16	Peripheral nerve stimulation for two weeks	Needle electrode was inserted into the trunk of a major peripheral nerve (i.e., the femoral nerve trunk and/or the sciatic nerve trunk)	–	−81% ± 28% (P < 0.002)
Pereira 2013 (UK)	QE	Outpatients Age: 45–71 y Sex: M = 3, F = 2 Cause: traumatic	5	DBS Initial frequency: 10 Hz in three patients, 20 Hz in two patients, 30 Hz in one patient. Frequency at 3 years: 10 Hz in four patients, 15 Hz in one patient, 20 Hz in one patient	Contralateral, ventroposterolateral nucleus of the sensory thalamus targeted. Effect of macrostimulation assessed from 2 mm above to 5 mm below the calculated target to elicit paresthesia or analgesia Quadripolar electrodes. Final electrode position determined by intraoperative clinical assessment that relied on subjective reporting by awake patient during 48-h period of postoperative clinical assessment before decision on whether to permanently implant electrodes	–	−2.8 ± 2.6 (−65.3% ± 25%, P = 0.001)
Bittar 2005 (Australia)	QE	Outpatients Age: 38–76 y Sex: M = 3 Cause: vascular insufficiency = 1, traumatic = 2	3	DBS Implanted electrodes had four exposed contacts, each 1.5 mm long, arranged linearly with 1.5 mm in between. Location was adjusted to find site of greatest pain relief. Several days of testing before generator implanted	DBS (contralateral) of PVG and somatosensory thalamus. Two patients: PVG only. One patient: PVG and thalamic stimulation	–	−6.17% ± 7.4% (range = 55–70%, P = 0.02)

BKN = below knee; C = control; DBS = deep brain stimulation; EEG = electroencephalography; I = intervention; F = female; M = male; MARP = military amputee rehabilitation program; NMES = neuromuscular electrical stimulation; NR = not reported; NRS = numeric rating scale; PLP = phantom limb pain; PVG = ; QE = quasi-experimental; RCT = randomized controlled trial; rTMS = transcranial magnetic stimulation; tDCS = transcranial direct current stimulation; TENS = transcutaneous electrical nerve stimulation; VAS = visual analog scale.

Risk of Bias

The RoB for nine RCTs was assessed using the Jadad scale. The range of points was from 2 to 5 (TMS studies had low RoB, tDCS studies had low to moderate RoB, and the other techniques had moderate to high RoB). Five studies (56%) did not report the blinding procedures or reported incomplete information about the blinded personnel. Three studies (33.3%) did not report how they generated the randomization sequence and allocation concealment, even though they specified that the allocation of the subjects was done randomly (Figure 2a). The risk of bias for the five QE studies was assessed using the MINORS tool. The studies have a moderate to high RoB. All studies did not report the sample size calculation and the procedure and personnel in charge to assess the primary outcome. Also, all the studies included consecutive patients (Figure 2b).

Figure 2.

Risk of bias assessment. a) Risk of bias of randomized controlled trials using the Jadad scale. b) Risk of bias of quasi-experimental studies using Methodological Index for Non-Randomized Studies scale.

Neuromodulation Effects on Pain

We first evaluated the pooled effect of noninvasive excitatory M1 stimulation (including high-frequency rTMS and anodal tDCS). We found a significant and large effect size for short-term pain reduction (immediately after the end of stimulation sessions—six RCTs, two parallel and four crossover studies, N = 123, ES = −1.36, 95% CI = −2.26 to −0.45) (Figure 3.1a). We also found a significant medium-term pain reduction after one week to one month of stimulation (four RCTs, two parallel and two crossover studies, N = 106, ES = −1.24, 95% CI = −2.18 to −0.31) (Figure 3.2b).

Figure 3.

Forest plots on excitatory M1 noninvasive brain stimulation effects in phantom limb pain patients. a) Short-terms effects. b) Medium-term effects.

Considering the tDCS studies only, we found also a significant and large effect size of anodal tDCS in M1 for pain reduction immediately after the end of stimulation sessions (three crossover RCTs, N = 33, ES = −1.50, 95% CI = −2.05 to 0.95) (Figure 3.2a) and a significant pain reduction at one-week follow-up (two RCTs, N = 25, ES = −1.04, 95% CI = −1.64 to 0.45) (Supplementary Data). On the other hand, we found no significant pooled effect of anodal or cathodal tDCS in PPC on pain reduction immediately after the end of the stimulation sessions (one crossover RCT, N = 8, ES = −0.80, 95% CI = −1.82 to 0.22, ES = −0.40, 95% CI = −1.39 to 0.59) (Supplementary Data). These studies did not report any adverse effects related to tDCS stimulation.

Considering the rTMS studies only, we found no significant pooled effect of rTMS in M1 on pain reduction immediately after the end of the stimulation sessions (three RCTs, one parallel and two crossover, performed excitatory rTMS, N = 90, ES = −1.25, 95% CI = −3.05 to 0.56; one RCT performed inhibitory rTMS, N = 9, ES = −0.07, 95% CI = −1.00 to 0.85) (Supplementary Data) or pain reduction at one-month follow-up (two RCTs performed excitatory rTMS, N = 81, ES = −1.46, 95% CI = −3.79 to 0.86) (Supplementary Data). These studies do not report any adverse effects related to rTMS stimulation.

For the QE studies, we calculated a pooled response rate of TENS after stimulation sessions and found a significant pooled response rate (two studies, N = 20, pooled response rate = 67%, 95% CI = 60% to 73%) (Supplementary Data) and similar results for DBS after one year of follow-up (two studies, N = 8, rate = 73%, 95% CI = 63% to 82%) (Supplementary Data).

Discussion

Our review comprised 14 studies (nine RCTs and five QE studies) that measured the effects of neuromodulation techniques (tDCS, rTMS, DBS, PNS, NMES, TENS) in patients with PLP. These studies were heterogeneous with regard to stimulation parameters and had small sample sizes and variable risk of bias (RCTs: low to moderate; QE: moderate to high). The pooled effects showed that PLP patients who underwent excitatory M1 stimulation, especially anodal tDCS, experienced a significant reduction in pain scale scores at the short-term and midterm time points. Motor cortex rTMS effected no significant difference between the active and sham groups. DBS and TENS elicited a significant decrease in PLP (noncontrolled QE studies).

Effects of tDCS and rTMS

Our exploratory meta-analysis of tDCS included three RCTs [29,30,33], demonstrating a significant decline in visual analog scale (VAS) scores for pain at the short-term and midterm end points with anodal M1 tDCS. Only cathodal tDCS to PPC decreased nonpainful phantom sensations [30]. There was no benefit for stump pain or telescoping. These results implicate a separation between the effects on painful and painless phantom sensations, likely due to the different neural circuits that are involved. None of the three trials reported moderate to severe side effects, confirming the safety of and tolerance to this technique [41].

The tDCS stimulation parameters were homogeneous; all studies targeted the motor cortex using excitatory polarization with a low current intensity (1–2 mA). Also, the session duration was 15–20 minutes. The main difference was the number of sessions [1–5]; the study with more sessions [33] reported a larger effect size, but due to the small number of included studies, we cannot suggest a potential dose response of this intervention, necessitating further studies to determine the influence of the number of sessions in PLP patients. Only one study [33] used a concomitant task (imaginary movements of the amputated limb) during tDCS stimulation, reporting a large effect on pain, indicating that the inclusion of concomitant motor tasks in the tDCS protocol may help to reactivate the deafferented motor cortex (amputated side) [42].

All of the studies on rTMS were RCTs [27,28,31]. Although Ahmed et al. [31] noted benefits of rTMS in reducing PLP, the exploratory pooled effect showed no significant difference between the active and sham groups.

One explanation of this negative result is the frequencies in these three rTMS experiments vs tDCS. Yet, all of the studies targeted the motor cortex; the forest plot results (Supplementary Data) showed a clear effect of stimulation frequency, with 1 Hz and 5 Hz showing no effect, 10 Hz having a marginal effect, and 20 Hz eliciting a significant result, suggesting that higher frequencies should be examined in future studies. This heterogeneity of the intervention makes it difficult to determine the reference stimulation parameters and might be a source of variability in the clinical effects.

This variability of parameters contrasts with the tDCS studies that had more homogeneous stimulation parameters (M1 anodal tDCS, cathoda l supraorbital, 2 mA, 20 minutes). Further, the baseline characteristics of the rTMS studies included patients who had undergone amputation less than three months earlier, which is not the best time to begin treating PLP. Conversely, more than three months had passed after amputation in most of the patients in the TDCS studies.

These trials were limited by their small sample sizes and short follow-up periods. Only well-powered trials with longer follow-ups can determine the true potential of tDCS and rTMS for PLP.

Effects of Invasive Neuromodulation Techniques

There are four reviews on DBS in patients with chronic pain, including small studies with PLP patients; three studies noted pain relief [43–45], but one showed no significant reductions in pain scores [46]. However, most of these reports were small mixed studies or case series. All available evidence has come from small trials, especially noncontrolled QE studies, and there is no consensus on how to perform DBS surgery or the parameters of stimulation [47], hampering the studies’ comparability. Although the pooled effect of the QE studies is significant in the short term and at follow-up, uncontrolled research results are inherently unreliable, and the certainty of the results should be considered low. Moreover, the risk for complications of invasive neuromodulation surgery is high; it is usually used only for severe PLP and is often chosen as a late treatment option, likely rendering it unsuitable for widespread use.

We could not include any study on DRG or SCS, for which there are only case reports, case series, and small retrospective cohorts (Supplementary Data). There are limited data on invasive neuromodulation (DBS, DRG, and SCS), and further research is needed.

TENS Effects

Three reviews on TENS [48–50] in mixed chronic pain populations, including few PLP patients, have been unable to judge the effectiveness of TENS for this condition. The main problem of TENS studies is the difference in stimulation site, including auricular areas [36], the contralateral leg [37,38], and above the stump [40]. All of these studies reported a significant reduction in PLP, based on the VAS, McGill Pain Questionnaire, universal pain score (UPS), and numerical rating scale (NRS). We pooled the results of the two quasi-experiments with the same stimulation parameters (contralateral to the site at which PLP was experienced on the amputated leg), and a significant reduction was found immediately after stimulation. However, the limitations of the noncontrolled QE designs prevented us from making a definitive conclusion.

Neurophysiological Hypothesis of Motor Cortex Stimulation in PLP

An important finding of this meta-analysis is that excitatory (anodal) primary motor cortex (M1) tDCS stimulation elicits a statistically and clinically significant reduction in pain intensity (by 1.5 points on the VAS) in patients with PLP [51]. The most frequently accepted hypothesis states that after amputation, PLP is associated with maladaptive reorganization of the sensory–motor cortex [52] and lower cortical inhibition [53], based on functional magnetic resonance imaging and TMS studies. Thus, PLP could be associated with decreased afference to M1, resulting in maladaptive plasticity in this area [54].

The resulting pain relief can be explained by the potential effects of tDCS on central pathophysiological mechanisms of pain that is associated with PLP [11,16]. It has been suggested that anodal M1 tDCS regulates in a top-down manner, sending signals toward the thalamocortical connections, prefrontal cortex, cingulate gyrus, and periaqueductal gray [55–57]. Based on this assumption, Lang et al. [58] reported significant activation of cortical and subcortical pain-related areas with anodal stimulation in healthy subjects. Further, one study on M1 anodal tDCS noted an increase in endogenous opioid release in the thalamus, insula, cingulate, and nucleus accumbens in a chronic pain patient [59].

Based on animal models, we hypothesized that the motor cortex not only is a passive target but also could be actively involved in the modulation of pain networks through disinhibition of the periaqueductal gray, enhancing the endogenous pain modulation system and the pain pathway of the dorsal horn in the spinal cord [60,61]. This phenomenon could result in the widespread modulation of other pain-related neural areas, such as the thalamus, thus modulating the upstream nociceptive signaling pathway. However, more mechanistic human studies are needed to truly understand the function of motor cortex stimulation as a target in the treatment of PLP patients and its influence on pain perception.

Limitations and Strengths

Due to the heterogeneity in stimulation parameters in the studies that were analyzed, it can be argued that meta-analyses do not compare similar interventions. These techniques are too heterogeneous to pool into a single unique meta-analysis; thus, we performed separate meta-analyses for similar techniques (mechanism of action and similar stimulation parameters) to reduce heterogeneity. The range in the variability of parameters was evaluated carefully to avoid a meta-analysis of opposing mechanistic effects (such as excitatory vs inhibitory stimulation, central vs peripheral intervention, and invasive vs noninvasive). For instance, the main pooled estimate in our study included noninvasive excitatory M1 stimulation. In addition, because effect estimates are needed for decision-making to guide future research and clinical applications, we found meta-analyses to be useful in providing a better overview of the results. Further, our meta-analysis recognizes the limitations in the primary studies, such as insufficient detail on the evaluation of the outcome, concomitant interventions, and the treatment that was received by the control group.

However, this study has important strengths. It followed the PRISMA guidelines and was registered in the PROSPERO database. In addition, we performed a comprehensive search strategy using multiple databases, without any language restrictions and across the articles that were cited by each of the identified studies, allowing us to retrieve all articles in previous systematic reviews [17,18] and others that were not found in these reviews. These strengths allowed us to report the state of the art in RCTs and QE controlled research on neuromodulation interventions in PLP and examine the efficacy and safety of these interventions.

Conclusions

The evidence from RCTs suggests that excitatory M1 stimulation—specifically, anodal M1 tDCS—significantly reduces pain scale scores in PLP patients after the stimulation sessions and at the midterm follow-up (one week). There are limited data on the efficacy of invasive neuromodulation techniques (DBS, DRG, and SCS). More RCTs are needed to evaluate the benefits and risks of neuromodulation techniques in patients with PLP. These studies need to be well powered, with longer follow-up (more than six weeks), and adequately reported (stimulation parameters and co-intervention description) to provide a definitive conclusion.

 

Funding sources: This study was supported by an National Institutes of Health (NIH) RO1 grant (1R01HD082302-01A1 to Felipe Fregni).

Conflicts of interest: The authors declare that they have no compelling interests related to this article.

Protocol registration number: PROSPERO 2018 CRD42018117998.