Epidural spinal cord stimulation for motor recovery in spinal cord injury: A systematic review

2.1 Search strategy

The methodology described by the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement (Moher et al., 2016) was adhered to for the current review. In consultation with a research librarian (DM), the most suitable electronic databases were chosen, and key search terms agreed upon. An exhaustive literature search was performed using four electronic databases (CINAHL, Embase, Medline, Web of Science). A search strategy and Medical Subject Headings (MeSH) was constructed around the following key terms: spinal cord stimulation, spinal cord injury and motor response generation. These search terms were expanded using a wide array of alternative terminologies, truncations and abbreviations. An inverse manual search of the bibliographies of all included full-text articles was also conducted. Filters were used to restrict the search to specific dates (January 1995 to June 2020), English language studies and human studies.

2.2 Study selection procedure

Eligibility criteria were formulated in accordance with the PICO model (Population, Intervention, Comparison, Outcome). In brief, the following inclusion criteria were established requiring: (1) Participants >18 years of age with a primary diagnosis of SCI and an Epidural Spinal Cord Stimulator surgically implanted, (2) Epidural stimulation aimed primarily at producing a motor response, (3) ESCS of a pulsed or continuous nature, (4) A clinical intervention design (Case study, Case series, Cohort study or Randomized controlled trial), (5) Studies including at least one reasonable measure of motor output (e.g. EMG, force), (6) Studies reporting data pertaining to stimulation parameters, (7) Primary original data from an interventional study, and (8) Publication in English between January 1995 to June 2020. We excluded (1) Animal studies, (2) Participants <18 years of age, (3) Healthy participants, (4) Participants with other neurological disorders, (5) All other types of interventional electrical spinal cord stimulation, (6) Pain as the primary outcome, (7) Studies that failed to specify stimulation parameters, (8) Review articles, conference proceedings, expert opinions or any other secondary publications, (9) Full text or abstract unavailable and (10) Articles published before 1995. After retrieval of the initial search results, duplicates were removed by review-specific software (www.covidence.org). Following this, two independent reviewers (CM, CT) began pilot screening of 150 randomly selected articles to ensure clear interpretation of the exclusion criteria. This pilot screening process was repeated until a Cohen’s Kappa >0.75 was achieved. Subsequently, the abstracts were screened by both reviewers using the criteria listed above and the reasons for exclusion were documented. Inter-operator agreement for abstract screening resulted in a Cohen’s Kappa of 0.80. Following this, full texts were reviewed for inclusion and again, all reasons for exclusion recorded. The decisions of both reviewers were compared, and the inclusion or exclusion of disputed studies was discussed with a third reviewer (NF) until a clear consensus was reached. The literature search was last performed on the 30 June 2020.

2.3 Data extraction

Two independent reviewers completed a standardized spreadsheet designed in Excel in order to systematically extract the relevant data for review. Information was extracted regarding (1) Patient demographics, (2) Stimulation parameters, (3) Outcome measures related to motor output, (4) EMG methods and results, (5) Adverse events and (6) Repeat participants. Repeat participants were identified if the following three criteria were fulfilled: 1) The study referenced the participant in a previous study, 2) Age, ASIA classification and time since injury data matched the referenced participant in both studies and 3) majority of authors on the duplicate papers were consistent. Data extraction results were evaluated for accuracy and a third reviewer adjudicated to resolve any lack of consensus.

2.4 Methodological appraisal

The Modified Downs and Black Quality Checklist was used to assess the methodological quality of the included full texts (Downs & Black, 1998). Two independent reviewers conducted the quality appraisal. When disagreements occurred a third reviewer adjudicated until a consensus was reached. The Downs and Black Quality Checklist is a 27-item checklist that evaluates methodological strengths and weaknesses of articles based on the categories of (1) Reporting, (2) Internal validity (Bias), (3) Internal validity (Confounding), (4) External validity and (5) Power. The modified version differs from the original only regarding the final question of statistical power. Instead of rating according to a post-hoc calculation of study powers, a binary rating is used based on whether the study performed a power calculation or not. Downs and Black scores were subsequently awarded a quality rating based on previously published descriptors: excellent (26 to 28); good (20 to 25); fair (15 to 19); and poor (≤14) (Hanada et al., 2020; Hooper, Jutai, Strong, & Russell-Minda, 2008; Munn, Sullivan, & Schneiders, 2010).

3.1 Study selection

An initial screening yielded 3435 articles. A total of 1213 duplicate articles were subsequently removed. Following title and abstract screening, 40 articles were identified for full-text review. Of the 40 articles reviewed at full-text, 18 fulfilled all inclusion criteria. Of these 18 included studies, all were interventional case studies or case series. The selection of the studies followed the PRISMA guidelines (Moher et al., 2016). See Fig. 1 for the PRISMA diagram detailing the search process.

OPEN IN VIEWER

Fig. 1

PRISMA Diagram (Moher et al., 2016).

The total number of study participants evaluated in the current review was 40. However, 7 of these were identified as repeat participants in a minimum of two and a maximum of four studies, resulting in cumulative data presented on only 24 individuals. The majority were of a young age (34±12 yr) and of male gender (73.7%). The mean (±SD) time between SCI and study enrolment was 4.2 (±2.1) yr. In terms of injury severity, the most studied sub-group were those with an initial classification of ASIA-A (n = 11), followed by ASIA-B (n = 8), ASIA-C (n = 4) and ASIA-D classifications (n = 1).

3.3 Methodological approach

All of the studies in this review were case studies or case series. Due to the heterogeneity of the outcome variables reported, a quantitative synthesis of methodological parameters was not possible. Therefore, a qualitative table was constructed outlining the intervention type, sample sizes, levels of injury, ASIA classifications, parameters selected and implant devices (Table 1). Training data were also included in this table (frequency, intensity, time and type as available), as well as the last follow-up evaluation post-ESCS training, see Table 1. Described stimulation parameters (frequency, amplitude/ and mode of delivery) varied greatly. All but one of the included studies targeted the lower limbs via the CPG (T12-L2) to enable locomotor type activity. Lu et al. (2016) was the only study to use ESCS for the purposes of upper limb motor recovery.

The motor outcome measures selected by the studies in this review varied greatly. EMG was employed by seventeen of the eighteen studies. Gait-related measures; namely, velocity and/or distance were collected by eight studies (Angeli et al., 2018; Carhart et al., 2004; Ganley et al., 2005; Gill et al., 2018; Herman et al., 2002; Huang et al., 2006; Moshonkina et al., 2012; Wagner et al., 2018). However, only two studies utilized clinically validated walking tests; namely the 6-minute walk test or 10 metre walk test (Herman et al., 2002; Wagner et al., 2018). Formalised muscle force evaluations were reported in four studies (Angeli et al., 2018; Angeli et al., 2014; Lu et al., 2016; Wagner et al., 2018). A full list of outcome measures is documented in Table 2.

ASIA classification both pre- and post-intervention was reported by five studies (Angeli et al., 2018; Gill et al., 2018; Grahn et al., 2017; Lu et al., 2016; Wagner et al., 2018). Examining these five studies, three outlined the motor and sensory scores in detail (Table 3). Re-categorisation of ASIA score post-ESCS was achieved by four participants. It should be noted however, that both participants in the study reported by Lu et al. (2016) were re-categorised via an assessment with ESCS applied, an approach which was not reported at baseline assessment.

Table 4 outlines the level of functional motor recovery achieved by each study. The categories reflect the common goal in the majority of the reviewed studies, that is, to achieve ambulation. The most basic category is volitional motor activity (VMVT) such as voluntary isometric force or joint movement, followed by assisted function (ambulation and standing) and finally independent function (ambulation and standing). The number of participants and their ASIA classifications are also provided to give context to the reader. Independent ambulation with a gait aid (crutches/walker) was reported by four of the 18 studies (Carhart et al., 2004; Ganley et al., 2005; Herman et al., 2002; Wagner et al., 2018).

With the exception of Herman et al. (2002), all other studies (n = 17) presented some form of EMG data as an indirect estimate of ESCS’s impact on motor function. Table 5 provides a description of the EMG recording and signal processing methods employed, along with a synopsis of the corresponding output presented by each study. Overall, the high degree of variability in recording and signal processing techniques was notable. Despite a large number of studies reporting some form of signal rectification and averaging (n = 12), an equal number (n = 12) presented exemplary raw EMG traces performed during VMVT (Angeli et al., 2014; Calvert et al., 2019; Lu et al, 2006; Moshinka et al., 2012; Rejc et al., 2015; Rejc et al., 2017a), assisted gait and/or standing (Gill et al., 2018; Rejc et al., 2017b, Wagner et al., 2018), or a combination of VMVT and gait/standing (Angeli et al., 2018; Grahn et al., 2017; Harkema et al. 2011). This output provides a useful, if somewhat limited, qualitative comparison of ESCS on and off (n = 8), level of muscle contraction (n = 2) or baseline compared to follow-up visits (n = 3). A total of four studies presented muscle activity during gait movements, using the more widely accepted format of EMG linear envelopes which have been amplitude normalised and averaged over a number of gait cycles (Carhart et al., 2004; Ganley et al., 2005; Gorgey et al., 2020; Huang et al., 2006). The resultant traces offer a clearer qualitative description of muscle activity during stepping movements with and without ESCS. Five studies did attempt some form of quantitative comparison in EMG response (Gill et al., 2018; Calvet et al., 2019; Rejc et al., 2015; Rejc, Angeli, Bryant, & Harkema, 2017; Wagner et al., 2018). However, given the very small sample size (n = 1 to 4) and in some cases the quantitative methods used, caution should be aired when interpreting these results.

Results from the modified Downs and Black Quality Checklist are presented in Table 6. While all 18 of the studies reviewed were categorised into the poor range (<14) (Hanada et al., 2020; Hooper et al., 2008; Munn et al., 2010), a chronological trend towards improving quality was observed. Lu et al. (2016) achieved the highest score and represented the only upper limb study (13/28). The highest score achieved by lower limb studies (10/28) was shared between four studies (Darrow et al., 2019; Rejc et al., 2017a; E. Rejc et al., 2017b; Wagner et al., 2018).

Reported adverse events in this review were very rare, with just one study reporting a hip fracture (Angeli et al., 2018). However, it should be noted that fourteen of the included studies failed to report any information regarding adverse events or lack thereof.

4.1 Summary of findings

This review identified a total of 18 studies that evaluated ESCS for improvement in motor function in 24 separate individuals with SCI. Thirteen of these studies included motor-complete SCI patients, with the remaining five reporting on motor incomplete patients. All studies reported some level of functional improvement, with eleven studies describing improved locomotor function, eight studies reporting improved standing ability and one study (Lu et al. 2016) describing improved upper limb function (Table 4). The ASIA impairment scale is currently the most widely accepted clinical classification system for SCI and forms the standard basis for measuring neurological outcomes (Alizadeh, Dyck, & Karimi-Abdolrezaee, 2019; Steeves et al., 2007). Clinical evaluations of sensorimotor functioning using this scale were reported post-intervention in five studies. A total of four participants (n = 4), in three of these reviewed studies, improved their overall classification (Table 3). One participant from Angeli et al. (2018) and two participants from Lu et al. (2016) were reclassified from ASIA-B to ASIA-C, while one participant from Wagner et al. (2018) was reclassified from ASIA-C to ASIA-D. Changes in EMG motor response were reported by all studies, with the exception of Herman et al. (2002). However, inconsistencies in the reported methods and presentation of EMG data limit any meaningful interpretation. The overall quality of the literature was poor (Downs & Black scores ranging from 4/28 to 13/28), with all papers being either case studies or case series. Considering the technological and logistical feasibility of research of this type, the reported lack of quality is unsurprising. However, chronological analysis is encouraging (Table 6) and suggests a trend towards improving quality, as study designs become more refined and robust.

4.2 Evidence of efficacy

The variability in outcome measurement and the limited sample size preclude a categorical declaration of efficacy at this point. Despite this, there are promising elements found by this review that warrant further attention. Namely, improvements in ASIA scoring reported in three studies (Angeli et al., 2018; Lu et al., 2016; Wagner et al., 2018). The total motor scores of Wagner’s three participants rose by 17, 13 and 6 points (Wagner et al., 2018). In Angeli et al. (2018) one participant out of three reported on, bettered their baseline motor score from 23 to 24/25 (Table 3). In contrast, Lu et al. (2016) chose to reassess their participants with ESCS switched on. Both participants enhanced their total motor scores from 9 to 39 and 17 to 37, respectively. Of these three studies (Angeli et al., 2018; Lu et al., 2016; Wagner et al., 2018), the improvements in motor scores were sufficient for the ASIA classification to be changed for four participants.

While clinical evaluations are useful for researchers and clinicians, functional abilities for daily activities have been noted as the most meaningful and valued outcomes for individuals living with SCI (Steeves et al., 2007). Improvement in EMG output without translation into function for example, offers little to improve the life of an individual with SCI. In this respect, ESCS has displayed some positive results. Four studies achieved independent ambulation with gait aids in the community and/or home setting (Carhart et al., 2004; Ganley et al., 2005; Herman et al., 2002; Wagner et al., 2018). While these impressive results were achieved in motor incomplete individuals, it still represents a significant easing of ADL difficulty for each participant. This is perhaps the strongest evidence to support efficacy of ESCS currently.

4.3 Study participants

The majority of participants were male (n = 14), in keeping with the higher incidences of SCI in the men (Smith et al., 2019; Smith et al., 2018). Study participants varied most notably in terms of ASIA classification and time since injury. Despite early and impressive results reported in motor incomplete SCI participants (Carhart et al., 2004; Ganley et al., 2005; Herman et al., 2002), most participants in the current review were motor complete (ASIA-A and B, n = 19), while only five participants with residual motor abilities were studied (ASIA-C and D, n = 5). Despite the imbalanced number of studies which have thus far evaluated ESCS in motor-complete patients (Table 3 and 4), the results from the current review would generally point towards greater efficacy in individuals with some residual motor function.

Time since injury (4.2±2.1 yr) was highly variable between studies; however, all participants at the time of recruitment were in the chronic stage of SCI, where substantial muscle atrophy has already occurred (Gorgey & Dudley, 2007; Kern et al., 2008). As of yet, no study has examined the use of ESCS in the acute or sub-acute stages of injury. Acutely, it may not be possible to safely implant an ESCS device. However, interventions in the sub-acute setting, before muscle atrophy has taken hold, may offer a more robust musculoskeletal environment for ESCS to function. Previous research has recognised that functional changes tend to occur earlier in the injury timeline, with the extent of motor improvements plateauing from 6 months onwards (Sumida et al., 2001; Waters, Yakura, Adkins, & Sie, 1992). Additionally, animal research has documented that early implementation of rehabilitation interventions improves recovery of function compared with delayed training (Brown, Woller, Moreno, Grau, & Hook, 2011; Norrie, Nevett-Duchcherer, & Gorassini, 2005). Hence, if the ideal study design were possible, based on the results of the current review, it would seem that ESCS applied to a large sample of participants classified as ASIA-C and ASIA-D in the sub-acute setting may offer the greatest potential for significant motor recovery.

4.4 Quality appraisal

The relatively poor Downs and Black results reported in the current review can primarily be attributed to the inherent design limitations of the included studies (case studies and case series). The entirety of the studies in the current review fell into the poor category (score of < 14/28). The single participant studies of Ganley et al. (2005) and Harkema et al. (2011) shared the lowest score with 4/28. Conversely, the evaluation of upper limb function by Lu et al. (2016) scored highest with 13/28. The external validity, selection bias and power categories were identified as the poorest performers amongst most studies. Encouragingly, however, quality scores did trend higher chronologically, as sample size and methods improved. Further increases in study quality could be achieved in the future, with a few simple changes. Firstly, sufficient details of the training intervention along FITT principle (frequency, intensity, time, and type) and a reporting of study compliance could be expounded upon easily. Secondly, simple reporting of adverse events (whether they occurred or not). Thirdly, more robust statistical methods could be reported in greater detail. These adjustments in reporting would have significantly enhanced study quality across many of the included papers, with no change to either their methodology or sample size.

In order for ESCS to advance to the point of wide scale clinical acceptance as a therapeutic technology, higher level study designs are urgently required. RCT’s may prove challenging for logistical and ethical reasons since finding a large number of participants at the same study site with suitably similar SCI’s (and support infrastructure) is highly unlikely. Greater time and resources should therefore be expended on implementing multi-site studies to overcome these logistical constraints and greater transparency surrounding recruitment methods will also enhance the external validity of ESCS studies.

4.5 Reported outcome variables

Given the breadth of patient motor abilities reported in the current review, it is not surprising that a wide range of outcome measures were used to evaluate ESCS efficacy (Table 2). For the studies involving motor complete participants, the goal of training was largely to improve standing and basic volitional movements. Therefore, the outcome measures utilised broadly reflected these goals. Time standing, level of physical assistance and percentage body weight support during standing were commonly used in such circumstances (Gill et al., 2018; Harkema et al., 2011; Rejc et al., 2015; Rejc et al., 2017a; Rejc et al., 2017b). A number of other studies with motor incomplete participants focused on gait related outcome measures such as gait velocity, distance achieved and the number of unassisted steps (Angeli et al., 2018; Carhart et al., 2004; Gorgey et al., 2020; Herman et al., 2002; Huang et al., 2006; Wagner et al., 2018). A modest selection of studies used established measures of muscle or joint force (Angeli et al., 2018; Angeli et al., 2014; Lu et al., 2016; Wagner et al., 2018). Regaining voluntary muscle force production forms the basis of motor recovery and so the omission of specific objective measurements of force by the majority of studies is notable. Likewise, muscle cross-sectional area is strongly correlated to force production (Jones, Bishop, Woods, & Green, 2008), yet only one study measured this variable pre- and post-intervention (Wagner et al., 2018). Previous reviews have highlighted the importance of selecting standardised and validated outcome measures to assess clinically meaningful changes in motor function for individuals with SCI (Alexander et al., 2009; Tomaschek, Gemperli, Rupp, Geng, & Scheel-Sailer, 2019). While one can sympathise with the difficulty of selecting appropriate outcome measures in the SCI/ESCS context, this review highlights the need for greater consistency between future studies, in order to allow a more formal synthesis of results at a later date.

4.6 Evidence of muscle activity

The most common outcome measure reported by almost all studies in the current review, was muscle activity via surface (and in some cases fine-wire) EMG. The reported methods were, in many cases, not reflective of best practice for recording (Hermens, Freriks, Disselhorst-Klug, & Rau, 2000), normalisation (Besomi et al., 2020) or presentation (Basmajian & De Luca, 1985) of dynamic EMG signals. A recent consensus statement on EMG signal normalization highlighted its importance for comparing muscle activity between measurement sessions and/or experimental conditions (Besomi et al., 2020). Amplitude normalisation procedures were not specified in 9 papers and in others are difficult to interpret. While lapses in methodological rigour may in part be explained by the qualitative and descriptive nature of the EMG data presented, five studies did attempt to quantitatively evaluate changes in EMG, using inferential statistics. Again, the approaches utilised were in some cases questionable. Notably, Darrow et al. (2019) reported a significant change in the “raw-noise response magnitude”, a novel EMG outcome variable, which appears to have originated with this group.

Several studies did present normalised EMG envelopes averaged across the gait cycle, providing a more comprehensive qualitative description of muscle activity with and without ESCS during assisted and unassisted stepping (Carhart et al.,2004; Ganley et al., 2005; Gill et al., 2018; Gorgey et al., 2020; Huang et al., 2006). Of note is the detailed qualitative description by Huang et al. (2006) of ESCS effects on muscle activity relative to the gait cycle and Gill et al. (2018) who quantitatively compared normalised EMG changes during discrete phases of the gait cycle. These highlights were arguably exceptions to the trends seen in the majority of studies. While presentation of exemplary un-normalized and un-rectified EMG traces does provide some qualitative evidence of motor engagement, it does little to move the field towards a consensus as to the efficacy of ESCS. Adherence to recommended recording, signal processing and normalization techniques would greatly improve the quality and consistency of this outcome measure in future ESCS studies.

4.7 Stimulation protocols used

Parameter details such as frequency, amplitude and placement of electrodes varied greatly between studies, and often only examples of selected parameters were provided. It must be appreciated that parameter selection and optimisation in practice is a constantly moving target. Daily ESCS training reported in Angeli et al. (2014) reduced the threshold intensity to produce force over the course of four months training, illustrating the need to adapt the stimulation parameters applied as the patient improves.

Similarly, there appears to be a task-specific nature to training. For instance, when the same parameters were applied to participants, minimal EMG activity was displayed in sitting but large amplitudes were generated in standing (Rejc et al., 2017a). Likewise, research reported by the same group observed that training with ESCS optimized for stepping subsequently impaired standing ability (Rejc et al., 2017a). Earlier work by Rejc et al. (2015) established that parameters optimised for standing in one individual resulted in poor rhythmical EMG activity, insufficient to support standing, in another. This may be due to the distinct pathological fingerprint of each individual SCI and the co-existing interplay of propriospinal connectivity (Eisdorfer et al., 2020; Taccola et al., 2018).

In something as dynamic and reactive as ESCS, one can understand the difficulty of expressing real-time parameters in the static medium of a research paper. Authors such as Rejc et al. (2015) and Gill et al. (2018) should be commended for providing the location of each stimulator, stimulation configuration, frequency, and amplitude in graphic form. In future studies, further detailed descriptions are warranted concerning individual ranges, task specific parameters and how fatigue may affect parameter selection.

4.8 Limitations

In order to adequately interpret the summary findings presented, some limitations of the current review must be considered. Firstly, all data extracted was garnered through the texts themselves or the supplementary data available as an accompaniment. As the duration of many studies lasted years, it stands to reason that it would be difficult for the authors to relay every detail in publication form. Secondly, the current review focused exclusively on studies which evaluated ESCS for improving motor function in SCI. Consequently, a number of technical ESCS studies or other non-motor related studies; namely, studies examining autonomic function were not included. The combined therapeutic value of ESCS may therefore not have been fully evaluated. Finally, the authors acknowledge the logistical and technical constraints of conducting larger RCT studies of a surgically implanted device in such a diverse patient cohort. The current review therefore provides the best available evidence at the time of writing.