Scolaris Content Display Scolaris Content Display

Spinal cord stimulation for low back pain

Collapse all Expand all

Background

Spinal cord stimulation (SCS) is a surgical intervention used to treat persistent low back pain. SCS is thought to modulate pain by sending electrical signals via implanted electrodes into the spinal cord. The long term benefits and harms of SCS for people with low back pain are uncertain.

Objectives

To assess the effects, including benefits and harms, of SCS for people with low back pain.

Search methods

On 10 June 2022, we searched CENTRAL, MEDLINE, Embase, and one other database for published trials. We also searched three clinical trials registers for ongoing trials.

Selection criteria

We included all randomised controlled trials and cross‐over trials comparing SCS with placebo or no treatment for low back pain. The primary comparison was SCS versus placebo, at the longest time point measured in the trials. Major outcomes were mean low back pain intensity, function, health‐related quality of life, global assessment of efficacy, withdrawals due to adverse events, adverse events, and serious adverse events. Our primary time point was long‐term follow‐up (≥ 12 months).

Data collection and analysis

We used standard methodological procedures expected by Cochrane.

Main results

We included 13 studies with 699 participants: 55% of participants were female; mean age ranged from 47 to 59 years; and all participants had chronic low back pain with mean duration of symptoms ranging from five to 12 years. Ten cross‐over trials compared SCS with placebo. Three parallel‐group trials assessed the addition of SCS to medical management.

Most studies were at risk of performance and detection bias from inadequate blinding and selective reporting bias. The placebo‐controlled trials had other important biases, including lack of accounting for period and carryover effects. Two of the three parallel trials assessing SCS as an addition to medical management were at risk of attrition bias, and all three had substantial cross‐over to the SCS group for time points beyond six months. In the parallel‐group trials, we considered the lack of placebo control to be an important source of bias.

None of our included studies evaluated the impact of SCS on mean low back pain intensity in the long term (≥ 12 months). The studies most often assessed outcomes in the immediate term (less than one month). At six months, the only available evidence was from a single cross‐over trial (50 participants). There was moderate‐certainty evidence that SCS probably does not improve back or leg pain, function, or quality of life compared with placebo. Pain was 61 points (on a 0‐ to 100‐point scale, 0 = no pain) at six months with placebo, and 4 points better (8.2 points better to 0.2 points worse) with SCS. Function was 35.4 points (on a 0‐ to 100‐point scale, 0 = no disability or best function) at six months with placebo, and 1.3 points better (3.9 points better to 1.3 points worse) with SCS. Health‐related quality of life was 0.44 points out of 1 (0 to 1 index, 0 = worst quality of life) at six months with placebo, and 0.04 points better (0.16 points better to 0.08 points worse) with SCS. In that same study, nine participants (18%) experienced adverse events and four (8%) required revision surgery. Serious adverse events with SCS included infections, neurological damage, and lead migration requiring repeated surgery. We could not provide effect estimates of the relative risks as events were not reported for the placebo period.

In parallel trials assessing SCS as an addition to medical management, it is uncertain whether, in the medium or long term, SCS can reduce low back pain, leg pain, or health‐related quality of life, or if it increases the number of people reporting a 50% improvement or better, because the certainty of the evidence was very low. Low‐certainty evidence suggests that adding SCS to medical management may slightly improve function and slightly reduce opioid use. In the medium term, mean function (0‐ to 100‐point scale; lower is better) was 16.2 points better with the addition of SCS to medical management compared with medical management alone (95% confidence interval (CI) 19.4 points better to 13.0 points better; I2 = 95%; 3 studies, 430 participants; low‐certainty evidence). The number of participants reporting opioid medicine use was 15% lower with the addition of SCS to medical management (95% CI 27% lower to 0% lower; I2 = 0%; 2 studies, 290 participants; low‐certainty evidence). Adverse events with SCS were poorly reported but included infection and lead migration. One study found that, at 24 months, 13 of 42 people (31%) receiving SCS required revision surgery. It is uncertain to what extent the addition of SCS to medical management increases the risk of withdrawals due to adverse events, adverse events, or serious adverse events, because the certainty of the evidence was very low.

Authors' conclusions

Data in this review do not support the use of SCS to manage low back pain outside a clinical trial. Current evidence suggests SCS probably does not have sustained clinical benefits that would outweigh the costs and risks of this surgical intervention.

PICOs

Population
Intervention
Comparison
Outcome

The PICO model is widely used and taught in evidence-based health care as a strategy for formulating questions and search strategies and for characterizing clinical studies or meta-analyses. PICO stands for four different potential components of a clinical question: Patient, Population or Problem; Intervention; Comparison; Outcome.

See more on using PICO in the Cochrane Handbook.

Spinal cord stimulation for low back pain

Background

Low back pain is a leading cause of disability around the world. Spinal cord stimulation, a surgical treatment involving implantation of a device that applies electric impulses to the spinal cord, has been suggested to improve pain in people with long‐term low back pain. This study aimed to review evidence regarding the benefits and harms of this procedure for people with low back pain.

Study characteristics

We searched online databases and registries for relevant studies on 10 June 2022. We found 13 trials with 699 participants. Of these, 55% were female and the average age of study participants ranged from 47 years to 59 years. The average duration of low back pain amongst study participants varied from 5 to 12 years. Ten of the 13 studies had financial ties to manufacturers of spinal cord stimulation systems.

Key findings

No studies have tested whether spinal cord stimulation surgery is better than placebo (sham or 'dummy' treatment) in people followed up for longer than 6 months. This means that the benefits of the treatment in the long term are unknown. Most of the available studies only measured outcomes at less than 1 month after treatment, and only 1 study measured outcomes at 6 months after treatment:

Pain intensity (0 to 100, lower scores mean less pain)

At 6 months, the only available study found no benefit of spinal cord stimulation on back pain compared with placebo (1 trial, 50 participants; moderate‐certainty evidence). At 6 months, participants given placebo treatment reported that their average pain was 61 points, and those given spinal cord stimulation reported that their pain was 4 points better (8.2 points better to 0.2 points worse).

Function (0 to 100, lower scores mean better function)

At 6 months, one study found no benefit of spinal cord stimulation on function (that is, people's general physical function) compared with placebo (1 trial, 50 participants; moderate‐certainty evidence). Participants given placebo treatment reported that their functioning was 35.4 points at 6 months, and those given spinal cord stimulation reported that their functioning was 1.3 points better (3.9 points better to 1.3 points worse).

Health‐related quality of life (0 to 1, higher scores mean better quality of life)

At 6 months, one study found no benefit from spinal cord stimulation on health‐related quality of life compared with placebo (1 trial, 50 participants; moderate‐certainty evidence). Participants given placebo treatment reported that their health‐related quality of life was 0.44 points at 6 months, and those given spinal cord stimulation reported that their health‐related quality of life was 0.04 points better (0.16 points better to 0.08 points worse).

Global assessment of efficacy (number of participants with a 50% improvement in pain or better)

None of the placebo‐controlled studies measured this outcome.

Withdrawals due to adverse events (i.e. an unwanted event that causes harm)

We are uncertain whether spinal cord stimulation caused people to withdraw from studies due to adverse events because there were few studies and the evidence was based on only a few cases.

Adverse events (e.g. increased pain)

One study that followed people for 12 months found 9 participants (18%) experienced adverse events such as infections, damage to the spine or nerves, bladder problems, and movement of very small parts of the devices that deliver the electrical impulses to the spinal cord (known as 'lead migration').

Serious adverse events (e.g. an infection requiring hospitalisation)

Some studies reported serious adverse events in people receiving spinal cord stimulation that required repeated surgery. The only placebo‐controlled study that followed people for 12 months found 4 participants (8%) required repeated surgery. In the five other studies of people receiving a new spinal cord stimulation implant, the number of people requiring repeat surgery, due to adverse events such as infection or device problems, ranged from 4.1% at 8 weeks to 30.9% at 24 months. However, it was not possible to estimate how common these events were compared with placebo or no treatment, as limited information was available.

Limitations of the evidence

For people with low back pain, we are moderately confident that, at 6 months, spinal cord stimulation probably does not lead to lower pain, better function, or higher quality of life compared with placebo. We are uncertain whether spinal cord stimulation can improve outcomes in the immediate term compared with placebo. Little to no information is available regarding long‐term efficacy or the risk of side effects and complications.

Authors' conclusions

Implications for practice

There is currently no evidence on the benefits and harms of spinal cord stimulation (SCS) compared with placebo in the long term for people with low back pain. Moderate‐certainty evidence suggests there is probably no benefit of SCS over placebo on pain, function, or health‐related quality of life in the medium term. Most placebo‐controlled trials to date have examined immediate‐term outcomes (less than one month) only, and although there appeared to be initial benefits on pain and function in some studies, the evidence was of very low certainty. Taken together, our findings suggest SCS probably has little to no sustained benefit over placebo for people with low back pain.

It is uncertain whether adding SCS to medical management improves low back pain intensity or health‐related quality of life, or whether it increases the number of people reporting a 50% improvement or better, because the certainty of the evidence is very low. Low‐certainty evidence suggests adding SCS to medical management may slightly improve function and slightly reduce opioid use in the medium term. Harms of SCS included infection, lead migration, and dural tear, some of which required repeat surgery. However, adverse events and serious adverse events were poorly documented and the magnitude of risk therefore remains uncertain.

The data in this review do not support the use of SCS for people with low back pain outside a randomised, placebo‐controlled trial.

Implications for research

The long‐term benefits and harms of SCS for people with low back pain are essentially unknown. Future trials should compare back pain outcomes with SCS versus placebo in the long term (i.e. at 12 months' follow‐up or longer), in people naive to the intervention, and use robust methods to minimise risk of bias. We cannot see value in new trials comparing different SCS types or comparing SCS to uncontrolled and unreported medical management, until any efficacy over placebo has been proven. To allow better evaluation of risk for adverse events and comparison with other treatment options, future trials should clearly document the number of people in each group who experience any adverse event, a serious adverse event, or withdraw due to an adverse event. Based on the data in this review, we cannot say if a specific subset of people with low back pain benefit from SCS and others do not benefit or are harmed; this could be investigated in future high‐quality trials.

Summary of findings

Open in table viewer
Summary of findings 1. Spinal cord stimulation versus placebo for low back pain in adults

Spinal cord stimulation (SCS) versus placebo for low back pain in adults

Patient or population: adults with low back pain

Setting: outpatient

Intervention: conventional, burst, or high‐frequency SCS

Comparison: placebo

Outcomes

Anticipated absolute effects (95% CI)

Relative effect

No. of participants (studies)

Certainty of the evidence (GRADE)

Comments

Risk with placebo

Risk with SCS

Pain intensity

VAS, translated to a 0‐ to 100‐point scale, where 0 is no pain

Medium‐term follow‐up (≥ 3 months to < 12 months)

Mean back pain during placebo period was 61 points

Mean back pain was 4 points better (8.2 points better to 0.2 points worse)

50 participants (1 study)

Moderatea

SCS probably does not improve back or leg pain in the medium term. Data are based on a single trial of burst SCS at low risk of bias. The CIs excluded clinically important benefits.

Eight of 10 available placebo‐controlled trials measured low back pain outcomes in the immediate‐term only. Based on those trials, it was uncertain whether SCS improves low back pain more than placebo in the immediate term (8 studies, 139 participants; very low‐certainty evidence).

Two trials measured leg pain in the immediate term. Based on those two trials, it was uncertain whether SCS improves leg pain more than placebo in the immediate term (2 studies, 39 participants; very low‐certainty evidence).

Function

Roland‐Morris Disability Questionnaire &

Oswestry Disability Index translated to a 0‐ to 100‐point scale, where 0 is no disability or best function

Medium‐term follow‐up (≥ 3 months to < 12 months)

Mean disability during placebo period was 35.4 points

Mean disability was 1.3 points better (3.9 points better to 1.3 points worse)

50 participants (1 study)

Moderatea

SCS probably does not improve function in the medium term. Data are based on a single trial of burst SCS at low risk of bias. The CIs excluded clinically important benefits.

One other study measured function in the immediate‐term only. Based on that trial, it was uncertain whether SCS improves function more than placebo in the immediate term (1 study, 20 participants; very low‐certainty evidence).

Health‐related quality of life

EQ‐5D, index from 0 to 1 where 0 is worst quality of life

Medium‐term follow‐up (≥ 3 months to < 12 months)

Mean quality of life during placebo period was 0.44 points out of 1

Mean quality of life was 0.04 points better (0.16 points better to 0.08 points worse)

50 participants (1 study)

Moderatea

SCS probably provides little to no benefit for health‐related quality of life in the medium term. Data are based on a single trial of burst SCS at low risk of bias. The CIs excluded clinically important benefits.

Two other trials measured health‐related quality of life in the immediate‐term only. Both suggested no benefit, though we were unable to pool the results of those studies (2 studies, 52 participants; very low‐certainty evidence).

Global assessment of efficacy

≥ 50% improvement in pain

Medium‐term follow‐up (≥ 3 months to < 12 months)

Not estimable

Not estimable

(0 studies)

No data available

Withdrawals due to adverse events

Follow‐up: longest measuredb

Not estimable

Not estimable

(0 studies)

Very lowe

Poorly reported in included studies. We are uncertain whether SCS results in more people withdrawing due to adverse events.

One small cross‐over RCT with 6‐week follow‐up reported 2 withdrawals with placebo versus 1 withdrawal with SCS (1 study, 19 participants; very low‐certainty evidence).

Adverse eventsc

Follow‐up: longest measuredb

Not estimable

Not estimable

(0 studies)

Very lowe

Poorly reported in included studies. One cross‐over study at low risk of bias found 9 out of 50 (18%) people who received SCS experienced an adverse event over a 12‐month period, but did not specify whether events occurred during the placebo or active SCS period.

Serious adverse eventsd

Follow‐up: longest measuredb

Not estimable

Not estimable

(0 studies)

Very lowe

Poorly reported in included studies. Although the incidence was uncertain, serious adverse events included infections, neurological damage, and lead migration requiring repeated surgery. One placebo‐controlled study at low risk of bias found 4 out of 50 (8%) people who received SCS required surgical revision within 12 months.

In the six trials in this review that followed people receiving a new SCS implant, surgical revision rates in the SCS group due to adverse events ranged from 4.1% at 8 weeks to 30.9% at 24 months.

CI: confidence interval; VAS: visual analogue scale

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.

aDowngraded one level for indirectness due to possible differences between the burst SCS regimen provided in the trial and other SCS regimens provided internationally.
bLong‐term efficacy and safety were not estimable as no data were reported.
cAdverse events included increased pain, infection, unpleasant paraesthesia, incorrectly implanted electrode causing shocks, pain at internal pulse generator/incision site, neurostimulator pocket fluid collection.
dSerious adverse events included unintentional dural tears during lead placement, revision of leads, infection requiring surgery, pulse generator replacement, and micturition problems requiring explant or revision surgery.
eDowngraded one level for risk of bias, one level for imprecision, and one level for indirectness.

Background

Description of the condition

Low back pain is the leading cause of years lived with disability worldwide (Global Burden of Disease Study 2018). Low back pain typically refers to pain between the twelfth rib and the buttock crease (Dionne 2008). Sometimes low back pain is associated with radiating leg pain or sciatica. In many cases, the source of low back pain cannot be established (Hartvigsen 2018). Instead, low back pain is classified in terms of duration: acute (fewer than six weeks' pain duration), subacute (six to 12 weeks' pain duration), or chronic (more than 12 weeks' pain duration). Some consider chronic low back pain that persists following back surgery to be a distinct syndrome known as 'failed back surgery syndrome' (FBSS) (Thomson 2013).

The mechanisms of chronic low back pain and associated leg pain are uncertain. Theories have suggested that persistent pain states, such as chronic low back pain, occur in part because of dysfunctional processing of pain‐related information in the spinal cord (Nijs 2015). However, the clinical importance of abnormal spinal cord processing in people with chronic low back pain remains uncertain (Roussel 2013).

Description of the intervention

Spinal cord stimulation (SCS) involves implanting a device in the low back/trunk that generates electrical pulses and delivers them to the spinal cord via electrodes within the posterior epidural space (Kemler 2000). The ‘leads’, containing sets of electrodes, can be implanted via laminectomy or percutaneously. Depending on the location and intensity of the person's pain, a clinician may select from a varying number and type of leads (uni‐, bi‐, or multipolar), and parameters of stimulation (amplitude, pulse width, frequency). Parameters of stimulation can be adjusted wirelessly using a remote control (Mailis‐Gagnon 2013).

Before a surgeon implants the device, current protocols usually require a trial screening period. Leads are temporarily placed percutaneously, and the clinician assesses the individual’s response to the stimulation while they continue with usual activities. The screening phase lasts from days to weeks. A positive response is often defined as at least 50% pain relief (Kemler 2000). If the screening phase is positive, a surgeon may offer a laminectomy to permanently implant a paddle lead or percutaneous leads which are anchored. Internal pulse generators (IPGs) are connected to the implanted leads via a tunnelling device such that the entire system is most often implanted under the skin. IPGs use rechargeable or primary cell batteries, depending on patient preference. The lifetime of an IPG is dependent upon multiple variables, including a person's use of the device. Replacement of IPGs is required for both rechargeable and primary cell types, though the former likely have a greater longevity.

How the intervention might work

The mechanism of action of SCS for low back pain is poorly understood. SCS was originally thought to work via the gate‐control mechanism (Melzack 1965); that is, stimulation of part of the spinal cord would interrupt transmission of pain‐related information to the cortex. However, evidence of the effects of SCS on the relay of pain‐related information at the spinal cord in humans is limited (Meyerson 2000). In addition, SCS does not appear to influence pain in response to an experimentally induced noxious stimulus (Meyerson 2000). Other suggested mechanisms have included inhibition of the sympathetic nervous system (sympatholytic effect) (Kemler 2000), and interrupted transmission of pain‐related nerve impulses by the brain (supraspinal inhibition) (Meyerson 2000). It is unclear whether the mechanism of action differs in people with chronic low back pain, compared to those with leg pain, or those diagnosed with failed back surgery syndrome (FBSS) (Meyerson 2000).

Why it is important to do this review

SCS is thought to be helpful for chronic low back pain, sciatica, and FBSS. The National Institute for Health and Care Excellence (NICE) recommends SCS for refractory neuropathic pain (NICE 2020). In 2014, the SCS market was estimated to be valued at 1.3 billion (i.e. 1300 million) US dollars (USD) (PRWeb 2015). In the USA, the average cost of implanting a stimulator is USD 30,000, plus USD 10,000 per annum for maintenance care if the person experiences complications. One study estimated that 12% of people who had SCS experienced at least one complication, such as lead migration or wound infection (Shamji 2015).

Evidence on the benefits and harms of SCS, compared with placebo or no treatment, is limited. A Cochrane Review of efficacy in chronic pain was withdrawn because the review was out of date (Mailis‐Gagnon 2013). Grider 2016 conducted a systematic review of SCS for low back pain and focused on a wide range of trials, including those that compared SCS with different stimulation regimens and various other control treatments of unknown efficacy. This made the true efficacy of the procedure difficult to determine. Grider 2016 did find three small trials that compared SCS to no treatment or placebo/sham (160 participants in total). The trials had mixed results. One small trial (40 participants) found no effect on pain intensity at four weeks compared with placebo SCS (device switched off) (Perruchoud 2013). One hallmark 2007 trial by Kumar and colleagues ('PROCESS'; 100 participants), investigating SCS as an addition to 'conventional medical management', found a large effect on leg pain at six months (‐26.7 points (95% confidence interval (CI) ‐40.4 to ‐13.0) on a 100‐point scale) (Kumar 2007, primary reference). Because the 'conventional medical management' was not standardised or provided in a controlled way, this effect is challenging to interpret.

There have been additional trials since the Grider 2016 review. In 2019, Rigoard and colleagues reported on the PROMISE trial (Rigoard 2019). Similar to the Kumar 2007 trial, PROMISE compared SCS plus 'optimal medical management' with 'optimal medical management' alone. The 'optimal medical management' was not standardised or controlled by the investigators. At six months, the between‐group difference in low back pain was 1.1 points (95% CI 0.6 to 1.6) on a 0 to 10 scale. The large effect on leg pain observed in the PROCESS trial by Kumar and colleagues in 2007 was not replicated: at six months the effect was 1.3 points (95% CI 0.7 to 1.9) on 0 to 10 scale (Rigoard 2019). In the Rigoard trial, 18% of participants experienced a stimulator‐related adverse event. The SCS Frequency Study, a small study (24 participants) that compared SCS treatment at three different frequencies against 'sham' SCS treatment (device is switched on but not delivering any stimulation), found that some SCS regimens were not superior to sham (Al‐Kaisy 2018). New trials are also underway (e.g. MODULATE‐LBP (Al‐Kaisy 2020)) or have overdue results.

To date, the evidence from trials of SCS suggests that, compared with placebo or no treatment, or as an addition to medical management, the effects on low back pain and leg pain are uncertain. A recent Cochrane Review of SCS interventions for any pain condition concluded that SCS may have clinically important effects when added to conventional medical management or physical therapy, but that effects over placebo may be much smaller and unimportant (O'Connell 2020). The certainty of evidence was low to very low. However, that review did not examine the evidence on SCS specifically for people with low back pain. A focused Cochrane Review will help resolve some of the uncertainty regarding efficacy of SCS for people with low back pain, and help clinicians, people with low back pain, and policymakers make decisions based on the best available evidence.

Objectives

To assess the effects, including benefits and harms, of SCS for people with low back pain.

Methods

Criteria for considering studies for this review

Types of studies

We sought randomised controlled trials (RCTs), quasi‐randomised trials (e.g. trials that use alternate allocation), and cross‐over trials (e.g. trials in people with implanted stimulators that compare active stimulation with a period or periods where the stimulator is turned off or is inactive to act as a placebo stimulation) for this review. We considered studies published as full texts, abstracts only, and data found from unpublished sources. We did not limit inclusion by date or language of publication.

Types of participants

We considered studies in adult participants (≥ 18 years) of any gender with chronic low back pain (> 12 weeks' pain duration), with or without leg pain, including people classified as having FBSS. We excluded studies in participants who had pain conditions other than chronic low back pain, with or without leg pain, unless we could obtain separate data for the effects of treatment on participants with chronic low back pain, with or without leg pain, either from the published report or through contacting authors. We excluded studies in participants who had chronic low back pain caused by serious spinal pathology (e.g. fracture, cancer, infection). We did not place restrictions on study setting or the demographic characteristics of participants.

Types of interventions

We considered studies that compared SCS to placebo or no treatment or assessed SCS as an addition to medical management. We excluded studies that only compared different forms of SCS. We included studies using SCS procedures of any kind (e.g. using an implanted rechargeable or conventional (not rechargeable) pulse generator (IPG) or an older design of radiofrequency stimulator), and using any stimulation protocol. For analysis, we considered 'conventional' SCS to be tonic stimulation below 1 kHz, 'high‐frequency' SCS to be tonic stimulation at 1 kHz to 10 kHz, and 'burst' SCS to be intermittent bursts of stimulation.

Comparator arms had to include a placebo or no treatment, or assess SCS as an addition to medical management. If no treatment was delivered by trial staff, we considered this a 'no treatment' group. Participants may have received co‐interventions that could be considered usual care, such as oral medicines (i.e. opioids, non‐steroidal anti‐inflammatories, antidepressants, anticonvulsants, and other analgesics), physical therapies (e.g. massage, acupuncture, spinal manipulation), psychological therapies (e.g. cognitive behavioural therapy), and injection therapies (e.g. nerve blocks, epidural corticosteroids) (Kumar 2007). Although not strictly a 'no treatment' comparison, we included trials assessing the addition of SCS to medical management that was provided (at least in part) by investigators.

The following are examples of acceptable placebo SCS interventions that we considered for inclusion: i) the stimulator is switched off; ii) the stimulator is switched on initially for programming then switched off; iii) the stimulator is switched on but emits no electrical impulse to the spinal cord. There is debate in the field about whether very low‐amplitude stimulation could also act as a placebo SCS stimulation (Tjepkema‐Cloostermans 2016). However, because of uncertainty around the precise level of stimulation that should be considered 'subtherapeutic,' we excluded studies comparing SCS intervention to very low‐amplitude stimulation, and considered studies that use such a comparator to be evaluating different forms of SCS.

Types of outcome measures

Major outcomes

For each outcome, we considered the hierarchy of pain and physical function outcomes provided by the Cochrane Musculoskeletal Group and the ranking of core outcome measures relevant to low back pain provided by Chiarotto and colleagues (Chiarotto 2018). Accordingly, where multiple outcomes were reported, we gave preference to the highest on the list. For each outcome, the hierarchy of outcomes is provided below in order of preference.

Outcomes assessing benefits

  • Pain intensity: numeric rating scale (NRS); visual analogue scale (VAS); pain severity subscale of Brief Pain Inventory

  • Function: Oswestry Disability Index version 2.1a or 24‐item Roland‐Morris Disability Questionnaire for physical functioning; NRS; global disability score; 36‐item Short‐Form (SF‐36) (physical function); other validated functional scales

  • Health‐related quality of life: 12‐item Short‐Form questionnaire (SF‐12); Patient‐Reported Outcomes Measurement Information System Global‐10 (PROMIS‐GH‐10); health‐related quality of life survey (HRQoL); EuroQol‐5D (EQ‐5D); 36‐item Short‐Form questionnaire (SF‐36) (mental health); other validated quality of life scale

  • Global assessment of efficacy: participant‐rated improvement measured as per cent improvement or on a categorical scale

Outcomes assessing harms

  • Proportion of withdrawals due to adverse events

  • Proportion of participants with adverse events: any adverse events reported (e.g. cardiovascular events, worsening of pain, fatigue, etc.)

  • Proportion of participants with serious adverse events (defined as leading to hospitalisation, disability, or death)

Minor outcomes

  • Medication use: number and proportion of participants taking any pain medication, daily dose of opioids as a morphine equivalent dose, or as reported in trials

  • Health care use: number of visits to any healthcare provider for care related to participant's back pain or management of the SCS, or both

  • Work status: number and proportion of participants reported to have returned to work, work absences, or as reported in trials

Timing of outcome assessment

We grouped outcome measures for outcomes assessing benefit (pain, disability, quality of life, medication use, health care use, work status) by timing of measurement as: immediate‐term (< one month), short‐term (≥ one month to < three months), medium‐term (≥ three months to < 12 months), or long‐term (≥ 12 months) follow‐up. For cross‐over trials, we used the duration of an SCS treatment to categorise timing of measurement. For example, if a trial had three treatment periods of two weeks each (placebo for two weeks versus high‐frequency SCS for two weeks versus conventional SCS for two weeks, with outcomes collected at the end of each period), then we designated this as 'two‐week follow‐up' and it fell in the immediate‐term category. If a trial had outcomes from multiple periods from the same SCS treatment (for example, a trial had two three‐month periods of burst SCS and two three‐month periods of placebo, pooling outcomes from both periods), then we designated this 'three‐month follow‐up' and it fell in the medium‐term category. Long‐term follow‐up (≥ 12 months) was our primary time point. We chose this primary time point because SCS systems can degrade over time and require replacement. The impact of these events can only be captured with long‐term follow‐up. We collected adverse event outcomes at the last time point.

Search methods for identification of studies

Electronic searches

We searched the following databases, from their inception to 10 June 2022:

  • Cochrane Central Register of Controlled Trials (CENTRAL; 2022, Issue 6);

  • MEDLINE via Ovid (1946 to 10 June 2022);

  • Embase via Ovid (1947 to 10 June 2022);

  • CINAHL (Cumulative Index to Nursing and Allied Health Literature) Complete via EBSCOhost (1982 to 10 June 2022).

We also searched the following trial registries for registered studies for which results have not yet been published:

When we found unpublished studies, we contacted trialists to request data for inclusion if we deemed the studies complete. If we were unsuccessful in obtaining data, we listed these studies as 'awaiting classification'. Where studies were ongoing, we kept records and reported them as such. We did not limit our search by date or language. See Appendix 1 for our search strategy.

Searching other resources

To identify any additional references, we searched the reference lists of included studies and systematic reviews relevant to the treatment of low back pain. We included any references highlighted through discussion with experts in the field. We also used personal communication with experts working in the field of back pain or chronic pain and communicated directly with manufacturers of spinal cord stimulators (including Medtronic, Boston Scientific Corporation, Nalu Medical, and Saluda Medical) to identify unpublished reports. In addition, we searched grey literature sources, including Bielefeld Academic Search Engine (BASE), Open Grey (opengrey.eu), and e‐thesis online (ethos.bl.uk).

Data collection and analysis

Selection of studies

Two review authors (AT and SG) independently screened titles and abstracts of all the potentially‐relevant reports we identified from the searches. We coded them as 'retrieve' (eligible or potentially eligible/unclear) or 'do not retrieve'. We retrieved the full‐text study reports/publications. Two review authors (AT and SG) independently screened these to identify studies for inclusion, and recorded reasons for exclusion of the ineligible studies. We resolved any disagreements through discussion or, if required, we consulted a third author (CM). We identified and excluded duplicate reports and collated multiple reports of the same study so that each study, rather than each report, is the unit of interest in the review. We recorded the selection process in sufficient detail to complete a PRISMA flow diagram (PRISMA Group 2009), and a Characteristics of excluded studies table. For screening of non‐English language papers, we initially used Google Translate to assist eligibility assessment. We did not require translators to assist with assessing eligibility of studies or data extraction.

Data extraction and management

We built a custom data collection form using Covidence for study characteristics and outcome data, which we piloted on several studies. One review author (AT) extracted study characteristics from included studies. A second review author (SG) spot‐checked study characteristics for accuracy against the trial report. We extracted the following study characteristics if available.

  • Methods: study design, total duration of study, details of any 'run‐in' or pre‐implantation screening period, number of study centres and location, study setting, withdrawals, and date of study.

  • Participants: N, mean age, age range, sex, socioeconomic status, back pain duration, pain severity, diagnostic criteria, inclusion criteria, exclusion criteria, and baseline pain, function, quality of life, pain medication use, healthcare use, and work status.

  • Interventions: intervention (including brand and type of SCS device, duration of intervention, stimulation parameters), comparison, concomitant medications, excluded medications or procedures, and post‐procedure care, as outlined in the TIDieR checklist (Hoffmann 2014).

  • Outcomes: primary and secondary outcomes specified and collected, and time points reported.

  • Characteristics of the design of the trial as outlined below in the Assessment of risk of bias in included studies section.

  • Notes: funding for trial and notable declarations of interest of trial authors.

Two review authors (AT and SG) independently extracted outcome data from included studies. We extracted the number of events and number of participants per treatment group for dichotomous outcomes, and means and standard deviations and number of participants per treatment group for continuous outcomes. We noted in the characteristics of included studies table if outcome data were not reported in a usable way or if we had to transform data or estimate it from a graph. We used the PlotDigitizer program to extract data from graphs or figures (PlotDigitizer) and performed this step in duplicate. We resolved disagreements by consensus or by involving a third review author (CM). One review author (AT) transferred data from Covidence into a Review Manager file (RevMan Web 2020). We double‐checked that data were entered correctly by comparing the data presented in the analyses with the study reports.

We selected data to extract based on the following decision rules:

  • Extract outcome data in the order of preference outlined in the Types of outcome measures section above.

  • If both final values and change from baseline values are reported for the same outcome, extract the final values.

  • If both unadjusted values and values that have been adjusted for baseline are reported for the same outcome, extract the adjusted values.

  • For outcomes assessing benefits, give preference to intention‐to‐treat (ITT) analysis data rather than 'per protocol' or 'as treated' data, if available.

  • If multiple time points are reported, use the one closest to the mid‐point: two weeks for immediate term, two months for short term, eight months for medium term. For long‐term outcomes, use the time point closest to 12 months.

Assessment of risk of bias in included studies

Two review authors (AT and SG) used the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions to independently assess risk of bias for each study (Higgins 2011). We resolved disagreements by discussion or by involving another author (CM or IH). We assessed the risk of bias according to the following domains:

  • random sequence generation (selection bias);

  • allocation concealment (selection bias);

  • blinding of participants and personnel (performance bias);

  • blinding of outcome assessment (detection bias), for self‐reported outcomes;

  • incomplete outcome data (attrition bias);

  • selective outcome reporting (reporting bias);

  • other bias: included if trials were stopped early, if there were differences between groups at baseline or differences between groups in timing of outcome assessment, and if there were co‐intervention differences across groups.

For cross‐over trials, we considered additional issues such as the impact of carryover and period effects, as suggested in Table 23.2a of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2021a).

We graded each potential source of bias as high, low, or unclear risk. In our risk of bias table, we documented a quote from the study report, together with a justification for our judgement. For each of the domains listed, we summarised the risk of bias judgements across different studies. If information on risk of bias was based on unpublished data or correspondence with a trialist, we noted this in the risk of bias table. When evaluating treatment effects, we took into account the risk of bias for the studies that contribute to that outcome. To provide summary assessments of the risk of bias, we presented the figures generated by RevMan Web 2020.

Measures of treatment effect

We analysed dichotomous data as risk ratios, or Peto odds ratios when the outcome was a rare event (approximately less than 10%), and used 95% confidence intervals (CIs). Data were insufficient to calculate the number needed to treat for an additional beneficial outcome (NNTB) or the number needed to treat for an additional harmful outcome (NNTH).

We analysed continuous data as mean difference (MD) or standardised mean difference (SMD), depending on whether the same scale was used to measure an outcome, and 95% CIs. When studies used different scales to measure the same conceptual outcome (e.g. function), we calculated SMDs rather than MDs, with corresponding 95% CIs. We back‐translated SMDs to a typical scale (e.g. 0 to 100 for pain) by multiplying the SMD by a typical among‐person standard deviation (e.g. the standard deviation of the control group at baseline from the most representative trial) (Higgins 2021b). We entered data presented as a scale with a consistent direction of effect across studies. For analysis of cross‐over studies, we used the generic inverse variance (GIV) approach, which allowed us to adjust mean differences for cross‐over design and multiple comparisons to the placebo group (see Unit of analysis issues).

We defined effect sizes for continuous outcomes as small (MD < 10% of the scale), medium (MD 10% to 20% of the scale), or large (MD > 20% of the scale) (Rubinstein 2012). Because the evidence was of low or very low certainty, we did not calculate NNTB or NNTH. For all continuous outcomes (pain intensity, function, health‐related quality of life), we considered a medium effect size (a difference of 15%) to be the minimum clinically important difference (MCID).

Unit of analysis issues

For all trials, the unit of analysis was the participant. Where a single trial reported multiple trial arms, we included only the relevant arms. If we combined two or more comparisons from the same study in a meta‐analysis, we attempted to adjust the number of participants in the placebo period to avoid double‐ or triple‐counting. For example, some studies compared multiple types of SCS to placebo. In each of these cases, we attempted to adjust for multiple comparisons to the placebo group. We adjusted results from Al‐Kaisy 2018, Schu 2014, Sokal 2020, and Sweet 2016 by estimating the mean difference, where the n in the control arm is divided by the number of comparator groups used in our analysis. This method of accounting for multiple comparisons to the placebo period required studies to report either raw data or standard deviations. De Ridder 2013 and Eldabe 2020 reported insufficient information on variance and so we could not adjust the estimated mean difference for multiple comparisons. For studies where multiplicity could not be adjusted for, it is likely that uncertainty is underestimated, increasing the chance of a type 1 error. We avoided analysing cross‐over studies as parallel studies, in accordance with Chapter 23 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2021a). None of the cross‐over trials provided data from the first phase, so data extracted from these trials are at risk of bias from carryover effects. We recorded this as 'other bias' in our risk of bias assessment. For studies where the cross‐over design could not be accounted for, the uncertainty is likely to be overestimated. Further information on data transformations used in our analysis of cross‐over trials is available in Table 1.

Open in table viewer
Table 1. Summary data extracted from cross‐over trials and methods used to estimate mean difference and standard error

Analysis

Study

Mean (intervention)

SD (intervention)

N (intervention)

Mean (placebo)

SD (placebo)

N (Placebo)

Effect

size (mean

difference)

SE

Effect size adjusted for cross‐over design?

Effect size adjusted

for multiple

comparisons

to placebo group?

Notes

1.1 SCS versus placebo SCS, Outcome 1: Low back pain intensity (0‐100) at immediate‐term follow‐up (< 1 month)

1.1.1

Al‐Kaisy 2018 (High‐frequency SCS)(1)

45.1

18.7

24

48.3

24.5

8

‐3.2

9.47

No

Yes

Mean difference and SE estimated from 2 sample t‐test with unequal variance adjusted for multiplicity by dividing N of placebo period by 3.

Mean and SD were rescaled (x10)

1.1.1

Al‐Kaisy 2018 (High‐frequency SCS)(2)

45.7

20.7

24

48.3

24.5

8

‐2.6

9.66

No

Yes

Mean difference and SE estimated from 2 sample t‐test with unequal variance adjusted for multiplicity by dividing N of placebo period by 3. Mean and SD were rescaled (x10)

1.1.1

Al‐Kaisy 2018 (High‐frequency SCS)(3)

32.2

19.8

24

48.3

24.5

8

‐16.1

9.56

No

Yes

Mean difference and SE estimated from 2 sample t‐test with unequal variance adjusted for multiplicity by dividing N of placebo period by 3. Mean and SD were rescaled (x10)

1.1.1

Perruchoud

2013 (High‐frequency SCS)

43.5

19.2

33

42.6

21.4

33

‐0.9

3.93

Yes. Means were from within subjects model

N/A

Effect size = ‐0.09 (95% CI ‐0.68 to 0.86) reported in paper. SE calculated from CI. Effect size and CI were rescaled (x10). Adjustment for period effects not required

1.1.1

Sokal 2020 (High‐frequency SCS)

51.7

14

18

54.2

12.2

18

‐1.7

2.20

Yes, using regression

weights ( β = ‐0.17) and SD of individual

regression

weight (τ = 0.68) provided by authors in Table A1

No

Effect size = ‐0.17, SE = 0.22 reported in paper Table A1. Effect size and SE were rescaled (x10). Effect size estimates are adjusted for cross‐over. Unclear if multiplicity was accounted for.

1.1.1

Sweet 2016 (High‐frequency SCS)

22.9

4.1

4

63.1

12.2

2

‐50.1

6.44

Adjusted for cross‐over, period and sequence effects

Yes

Patient level scores were digitally extracted from Figure 3. To estimate effect size, a mixed‐effects model was fitted accounting for cross‐over, period and sequence effects

1.1.2

De Ridder 2013 (Conventional SCS)

51.5

15

59.5

15

‐7.8

12.30

No

No

Mean estimates digitally extracted from Figure 3. Mean difference was calculated, and SE was assumed equal to burst SCS estimate from De Ridder. Results were rescaled (x10).

1.1.2

Eldabe 2020 (Conventional SCS)

51.0

19

38.0

19

‐12.8

3.9

Yes

No

Means extracted from report. Mean percentage reduction and confidence interval were reported. These were converted to absolute values and rescaled (x10).

1.1.2

Schu 2014 (Conventional SCS)

71

19

20

83

11

10

‐12.0

5.49

No

Yes

Mean difference and SE estimated from 2 sample t‐test with unequal variance adjusted for multiplicity by dividing N of placebo period by 2.

1.1.2

Sokal 2020 (Conventional SCS)

41.8

17.6

18

54.2

12.2

6

‐9.9

5.6

Yes

Yes

Effect size = ‐0.99, SE = 0.56 reported in paper Table A1. Effect size and SE were rescaled (x10)

1.1.2

Sweet 2016 (Conventional SCS)

53.2

6.3

4

63.1

12.2

2

‐31.2

7.2

Yes

No

Patient level scores were digitally extracted from Figure 3. To estimate effect size, a mixed‐effects model was fitted, accounting for cross‐over, period and sequence effects

1.1.2

Wolter 2012 (Conventional SCS)

56.8

22.4

6

63.7

20

6

‐28.5

6.4

Yes

N/A

Patient level scores were reported in Table 3. A paired 2 sample t‐test was performed (accounts for carryover).

Estimates were rescaled (x10)

1.1.3

DeRidder 2013 (Burst SCS)

35.5

15

59.5

15

‐24.1

12.3

No

No

Mean estimates digitally extracted from Figure 3. Difference between burst SCS and placebo was reported statistically significant at 0.05 threshold. Mean difference was calculated, and conservatively assuming P = 0.05 allowed calculation of the standard error for back pain.

Results were rescaled (x10)

1.1.3

Eldabe 2020 (Burst SCS)

54

19

51

19

2.55

5.2

Yes

No

Means extracted from report. Mean percentage reduction and confidence interval were reported. These were converted to absolute values and rescaled (x10).

1.1.3

Schu 2014 (Burst SCS)

47

25

20

83

11

10

‐36

6.58

Yes

Yes

Mean difference and SE estimated from 2 sample t‐test with unequal variance adjusted for multiplicity by dividing N of placebo period by 2.

1.1.3

Sokal 2020 (Burst SCS)

52.7

13.3

18

54.2

12.2

6

‐0.3

3.7

Yes

Yes

Effect size = ‐0.03, SE = 0.37 reported in paper Table A1. Effect size and SE were rescaled (x10). Effect size estimates are adjusted for cross‐over. Unclear if multiplicity was accounted for.

1.4 SCS versus placebo SCS, Outcome 1: Low back pain intensity (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

1.4.1

Hara 2022 (Burst SCS)

57

50

61

50

‐4.0

2.14

Yes

N/A

Mean difference and confidence intervals were reported in Table 2. SE was calculated from confidence interval. Results were rescaled (x10)

1.2 SCS versus placebo SCS, Outcome 2: Function (0‐100) at immediate‐term follow‐up (< 1 month)

1.2.2

Schu 2014 (Conventional SCS)

49.2

14.6

20

59

20.6

10

‐9.8

7.29

Yes

Yes

Mean difference and SE estimated from 2 sample t‐test with unequal variance adjusted for multiplicity by dividing N of placebo by 2.

Results were rescaled (x2).

1.2.3

Schu 2014 (Burst SCS)

38.4

16

20

59

20.6

10

‐20.6

7.43

Yes

Yes

Mean difference and SE estimated from 2 sample t‐test with unequal variance adjusted for multiplicity by dividing N of placebo by 2.

Results were rescaled (x2).

1.5 SCS versus placebo SCS, Outcome 2: Function (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

1.5.1

Hara 2022 (Burst SCS)

34.0

50

35.4

50

‐1.3

1.33

Yes

N/A

Mean difference and confidence intervals are reported in Table 2. SE was calculated from confidence interval.

1.6 SCS versus placebo SCS, Outcome 3: Health‐related quality of life (0‐1) at immediate‐term follow‐up (<1 month)

1.6.1

Perruchoud 2013 (High‐frequency SCS)

0.48

33

0.46

33

0.017

0.0602

Yes

N/A

Means were from within subjects model. Effect size = 0.017 (95% CI ‐0.101 to 0.135) extracted from report. SE calculated from CI.

1.8 SCS versus placebo SCS, Outcome 3: Health‐related quality of life (0‐1) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

1.8.1

Hara 2022 (Burst SCS)

0.48

50

0.44

50

0.04

0.0632

Yes

N/A

Mean difference and confidence intervals were reported in Table 2. SE was calculated from confidence interval.

1.3 SCS versus placebo SCS, Outcome 4: Leg pain intensity (0‐100) at immediate‐term follow‐up (<1 month)

1.3.1

Al‐Kaisy 2018 (High‐frequency SCS)(1)

18.1

25.5

24

25.1

25.5

8

‐1.4

10.41

No

Yes

Mean difference and SE estimated from 2 sample t‐test with unequal variance adjusted for multiplicity by dividing N of placebo group by 3.

Mean and SD were rescaled (x10). Follow‐up SD values not reported; taken from baseline

1.3.1

Al‐Kaisy 2018 (High‐frequency SCS)(2)

23.7

25.5

24

25.1

25.5

8

‐3.1

10.41

No

Yes

Mean difference and SE estimated from 2 sample t‐test with unequal variance adjusted for multiplicity by dividing N of placebo group by 3.

Mean and SD were rescaled (x10). Follow‐up SD values not reported; taken from baseline

1.3.1

Al‐Kaisy 2018 (High‐frequency SCS)(3)

22

25.5

24

25.1

25.5

8

‐7.0

10.41

No

Yes

Mean difference and SE estimated from 2 sample t‐test with unequal variance adjusted for multiplicity by dividing N of placebo group by 3.

Mean and SD were rescaled (x10). Follow‐up SD values not reported; taken from baseline

1.3.2

DeRidder 2013 (Conventional SCS)

36

15

66

15

‐30.1

15.3

No

No

Mean estimates digitally extracted from Figure 3 in report. Mean difference was calculated, and SE was assumed equal to burst SCS leg pain estimate from De Ridder.

Results were rescaled (x10).

1.3.3

DeRidder 2013 (Burst SCS)

36

15

66

15

‐30.1

15.3

No

No

Mean estimates digitally extracted from Figure 3. Difference between burst and placebo was reported statistically significant at 0.05 threshold. Mean difference was calculated, and conservatively assuming P = 0.05 allowed calculation of the standard error for leg pain.

Results were rescaled (x10).

1.7 SCS versus placebo SCS, Outcome 4: Leg pain intensity (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

1.7.1

Hara 2022 (Burst SCS)

59

50

61

50

‐2.0

2.28

Yes

N/A

Mean difference and confidence intervals were reported in Table 2. SE was calculated from confidence interval. Results were rescaled (x10).

Dealing with missing data

We contacted investigators or study sponsors in order to verify key study characteristics and obtain missing numerical outcome data where necessary (e.g. when we identified a study published as an abstract only or when data were not available for all participants). We did not identify cases where we thought the missing data could introduce serious bias, and therefore did not conduct a planned sensitivity analysis to explore the impact of missing data in the overall assessment of results.

For dichotomous outcomes that measure adverse events (e.g. number of withdrawals due to adverse events), we calculated the proportion using the number of participants that received treatment as the denominator.

For dichotomous outcomes that measure benefits (e.g. proportion of participants reporting pain medication use), we calculated the proportion using the number of randomised participants as the denominator.

For continuous outcomes (e.g. mean change in pain score), we calculated the MD or SMD based on the number of participants analysed at that time point. If the study did not present the number of participants analysed for each time point, we used the number of randomised participants in each group at baseline.

Where possible, we computed missing standard deviations from other statistics, such as standard errors, CIs or P values, according to the methods recommended in Chapter 6 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2021c). If we could not calculate the standard deviations, we imputed them (e.g. from other studies in the meta‐analysis).

Assessment of heterogeneity

We used the information in the data extraction tables to assess the included studies' clinical and methodological diversity, in terms of participants, interventions, outcomes, and study characteristics, to determine whether a meta‐analysis was appropriate. To assess statistical heterogeneity, we visually inspected the forest plots to look for obvious differences in results between the studies; we also used the I2and Chi2 statistical tests.

As recommended in the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2021), we interpreted an I2 value of 0% to 40% as indicating that the heterogeneity 'might not be important'; of 30% to 60% as representing 'moderate' heterogeneity; of 50% to 90% as representing 'substantial' heterogeneity; and of 75% to 100% as representing 'considerable' heterogeneity. As noted in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2020), we kept in mind that the importance of I2 depends on: (i) the magnitude and direction of effects and (ii) the strength of evidence for heterogeneity.

When interpreting the Chi2 test, we took a P value of less than or equal to 0.10 to indicate evidence of statistical heterogeneity. If we identified substantial heterogeneity, we reported it and investigated possible causes by following the recommendations in the Cochrane Handbook.

Assessment of reporting biases

Because we were unable to pool more than 10 trials, we did not create funnel plots or undertake formal statistical tests to investigate funnel plot asymmetry, as planned (Page 2021). To assess outcome reporting bias, we checked published reports against trial protocols and registries, and prepared an Outcome Reporting Bias in Trials ('ORBIT') matrix (Table 2). For studies published after 1 July 2005, we screened the World Health Organization clinical trial register on the International Clinical Trials Registry Platform (trialsearch.who.int) to check for protocols.

Open in table viewer
Table 2. Outcome Reporting Bias In Trials (ORBIT) matrix

Study ID

Low back pain intensity

Function

Health‐related quality of life

Global assessment (≥ 50% better)

Withdrawals due to adverse events

% with adverse events

% with serious adverse events

Al‐Kaisy 2018

Partial

?

?

?

Partial

Partial

Partial

De Ridder 2013

Partial

?

?

?

?

?

?

Eisenberg 2015

Full

?

?

?

?

?

?

Eldabe 2020

Partial

?

Full

?

Full

Partial

?

Hara 2022

Full

Full

Full

Not measured

Full

Partial

Partial

Kumar 2007

Partial

Partial

Partial

Partial

Partial

Partial

Measured

Kapural 2022

Full

Full

Partial

Full

Full

Full

Full

Perruchoud 2013

Partial

?

Partial

?

Partial

?

?

Rigoard 2019

Full

Full

Full

Full

Partial

Full

Partial

Schu 2014

Full

Full

?

?

?

Partial

Partial

Sokal 2020

Full

Partial

Measured

?

?

Measured

?

Sweet 2016

Full

Partial

Partial

?

?

?

?

Wolter 2012

Full

Partial

?

?

?

?

?

'Full': sufficient data for inclusion in a meta‐analysis were reported (e.g. mean, standard deviation, sample size per group for continuous outcomes).
'Partial': insufficient data for inclusion in a meta‐analysis were reported (e.g. means only, with no measures of variance).
'Measured': outcome was measured but no outcome data were reported.
'Not measured': outcome was not measured by trialists.
'?': unclear whether the outcome was measured or not (as a trial protocol or prospective study registry entry was unavailable).

Data synthesis

We undertook meta‐analyses only where this was meaningful; that is, if the treatments, participants, and the underlying clinical question were similar enough for pooling to make sense. We pooled outcomes grouped by comparison; namely, SCS versus placebo and SCS plus medical management versus medical management alone. We used random‐effects models where there were sufficient studies. In addition to the planned SCS versus placebo comparison, we conducted separate meta‐analyses for each of the three distinct clinical types of SCS: conventional SCS (tonic stimulation at < 1 kHz), high‐frequency SCS (tonic stimulation at 1 kHz to 10 kHz), or burst SCS (intermittent bursts of stimulation).

For our meta‐analyses of cross‐over trials (Analysis 1.1; Analysis 1.5; Analysis 1.3), we used the methods suggested in section 23.2 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2021a): we conducted a paired analysis where possible and adjusted for multiplicity by dividing the number of participants in the placebo period by the number of comparisons (see Table 1). We included results from paired analyses from cross‐over studies where these were reported or calculable, and pooled studies using the generic inverse variance approach. We used paired results from Sokal 2020 and Perruchoud 2013, and conducted our own paired analysis using data reported by Wolter 2012. We excluded one study from the analyses because its approach to intervention and outcome collection (1‐hour outcomes only) was substantially different to the other trials (Eisenberg 2015).

Our primary planned comparison and outcome was SCS versus placebo on low back pain intensity at long‐term follow‐up, for which there were no trials available. The other comparison of interest was SCS versus 'no treatment' on low back pain intensity. The latter analyses pooled all studies that assessed the addition of SCS to medical management. In all analyses, we included trials regardless of their risk of bias.

Subgroup analysis and investigation of heterogeneity

We did not locate a sufficient number of trials to allow formal subgroup analysis. As an exploratory analysis, we pooled outcomes separately for three distinct clinical types of SCS: conventional SCS, high‐frequency SCS, or burst SCS. We explored heterogeneity in our analysis of SCS as an addition to medical management, by examining the impact of removing one study that reported very large effects (Kapural 2022).

Sensitivity analysis

To investigate the robustness of the treatment effect on pain intensity and function for all time points, we had planned to carry out the following sensitivity analyses for the main comparison of SCS versus placebo:

  • including only studies we judged as having a low risk of selection bias;

  • including only studies we judged as having a low risk of detection bias.

Only two analyses, both at the immediate‐term time point, had a sufficient number of studies to conduct this sensitivity analysis (Analysis 1.1; Analysis 1.3).

Summary of findings and assessment of the certainty of the evidence

We created a summary of findings table using the outcomes (as described under Types of outcome measures) below.

Outcomes assessing benefits:

  • pain intensity;

  • physical function;

  • health‐related quality of life;

  • global assessment of efficacy.

Outcomes assessing harms:

  • withdrawals due to adverse events;

  • proportion with adverse events; and

  • proportion with serious adverse events.

The main comparisons in the summary of findings table were SCS versus placebo in the medium‐term (i.e. the longest measured time points in our included studies) for outcomes assessing benefits (pain, function, quality of life, global assessment of efficacy), and last follow‐up for outcomes assessing harms (withdrawal due to adverse events, adverse events, serious adverse events). Because the intervention is a surgically‐implanted device with substantial potential for adverse events (including revision surgery within two years), we considered that long‐term outcomes were likely to be the most important to people undergoing spinal cord stimulation. However, because no long‐term data were available, we decided (post hoc) to present data for the longest available time point (medium‐term follow‐up, i.e. ≥ 3months to < 12 months), rather than provide an empty summary of findings table.

Two people (AT and SG) independently assessed the certainty of the evidence. We used the five GRADE considerations (study limitations, inconsistency, imprecision, indirectness, and publication bias) to assess the certainty of the body of evidence as it relates to the studies which contributed data to the meta‐analyses for the prespecified outcomes, and reported the certainty of evidence as high, moderate, low, or very low. We used methods and recommendations described in Chapters 14 and 15 of the Cochrane Handbook for Systematic Reviews of Interventions (Schünemann 2021a; Schünemann 2021b). We justified all decisions to downgrade the certainty of evidence for each outcome using footnotes, and we made comments to aid the reader's understanding of the review where necessary. Due to sparse data, we were unable to provide a NNTB or NNTH, absolute and relative per cent change in the summary of findings (SoF) table, as described in the Measures of treatment effect section above.

We considered the following when making judgements about the five GRADE considerations.

  • Study design and risk of bias: we made an overall judgement on whether the certainty of the evidence for an outcome warranted downgrading on the basis of study limitations. To assist our interpretation of these biases, we referred to Table 14.2a in Chapter 14 of the Cochrane Handbook for Systematic Reviews of Interventions (Schünemann 2021b). For example, we considered downgrading the certainty of the evidence by one level if most of the evidence came from individual studies either with a crucial limitation for one item, or with some limitations for multiple items.

  • Inconsistency: we evaluated each direct comparison for consistency in the direction and magnitude of the effect sizes from individual trials, considering the width of the confidence interval and magnitude of the heterogeneity parameter. We downgraded comparisons by one level if we identified important and unexplained heterogeneity.

  • Indirectness: although we used precise inclusion criteria to minimise the scope for this problem, indirectness in the evidence could still arise. We used Table 14.2b in Chapter 14 of the Cochrane Handbook for Systematic Reviews of Interventions to assist interpretation of issues with indirectness (Schünemann 2021b). For each outcome, we judged indirectness arising from, for example, differences in participant populations, SCS intervention parameters, and 'no intervention' comparator protocols.

  • Imprecision: in cases where studies included relatively few participants and few events, and thus had wide confidence intervals around the estimate of the effect, the results of meta‐analyses that include these studies are imprecise.

    • Dichotomous outcomes: when the 95% confidence interval around the pooled or best estimate of effect included benefits or harms that would lead to substantially different clinical decisions (e.g. the confidence interval includes both no benefit and large benefit), we downgraded the evidence.

    • Continuous outcomes: as with dichotomous outcomes, we downgraded the evidence if the confidence interval was so imprecise that it included effects that would lead to opposing clinical decisions. That is, if the lower and upper bounds of the confidence interval included effects that would lead a clinician or person undergoing spinal cord stimulation to make a substantially different clinical decision, we downgraded the evidence.

  • Publication bias: because we found fewer than 10 studies examining the same intervention comparison, we used methods such as checking for unpublished trials in trial registries, examining protocol papers for outcome switching, and constructing an ORBIT matrix.

Results

Description of studies

Results of the search

Our search, conducted up to 10 June 2022, yielded 6492 records across five databases and two clinical trials registers (CENTRAL = 921; MEDLINE = 1014; Embase = 2719; CINAHL = 54; Bielefield = 940; trials registers (WHO ICTRP, clinicaltrials.gov) = 844). After duplicates were removed, 4776 unique records remained. Of these, we retrieved 113 articles for full‐text screening on the basis of their titles and abstracts. We deemed 13 trials eligible for inclusion (Al‐Kaisy 2018; De Ridder 2013; Eisenberg 2015; Eldabe 2020; Hara 2022; Kumar 2007; Kapural 2022; Perruchoud 2013; Rigoard 2019; Schu 2014; Sokal 2020; Sweet 2016; Wolter 2012). Three trials are awaiting classification (see Characteristics of studies awaiting classification). We initially identified 14 relevant ongoing trials in clinical trials registries, one of which was published on 18 October 2022 and subsequently included in this review (Hara 2022). Thus, we have classified 13 studies as ongoing (see Characteristics of ongoing studies). We excluded 29 studies (see details in Excluded studies and Characteristics of excluded studies). We present a flow diagram of the study selection process in Figure 1.


PRISMA study flow diagram

PRISMA study flow diagram

Included studies

Study design and setting

All thirteen studies were randomised controlled trials (RCTs). Ten used a cross‐over design (Al‐Kaisy 2018; De Ridder 2013; Eisenberg 2015; Eldabe 2020; Hara 2022; Perruchoud 2013; Schu 2014; Sokal 2020; Sweet 2016; Wolter 2012), and three used a parallel‐group design (Kumar 2007; Kapural 2022; Rigoard 2019). Six studies had two intervention arms (Eisenberg 2015; Hara 2022; Kumar 2007; Kapural 2022; Rigoard 2019; Wolter 2012), four had three intervention arms (De Ridder 2013; Eldabe 2020; Schu 2014; Sweet 2016), and three had four intervention arms (Al‐Kaisy 2018; Perruchoud 2013; Sokal 2020).

Three studies were multinational (Kumar 2007; Perruchoud 2013; Rigoard 2019). The other ten studies were conducted in seven different countries: Belgium (De Ridder 2013), Germany (Schu 2014; Wolter 2012), Israel (Eisenberg 2015), Poland (Sokal 2020), the UK (Al‐Kaisy 2018; Eldabe 2020), Norway (Hara 2022), and the USA (Sweet 2016; Kapural 2022). The total duration of treatment with SCS in placebo‐controlled trials varied between 2.5 hours and six months. Some parallel trials followed the SCS group for 24 months.

Six studies were funded by manufacturers of spinal cord stimulators (Al‐Kaisy 2018; Eldabe 2020; Kumar 2007; Kapural 2022; Perruchoud 2013; Rigoard 2019); four did not report a funding source but had investigators with financial ties to manufacturers (De Ridder 2013; Schu 2014; Sokal 2020; Sweet 2016); and three appeared independent of industry funding (Hara 2022; Eisenberg 2015; Wolter 2012).

Participant characteristics

Thirteen studies randomised 699 participants with low back pain to receive spinal cord stimulation or a control intervention, with the sample size ranging from four to 218 participants per trial. The mean age of participants ranged from 47 years to 59 years. Six studies reported the mean duration of back pain symptoms before the trial (Al‐Kaisy 2018; Eisenberg 2015; Kapural 2022; Rigoard 2019; Sokal 2020; Wolter 2012), which ranged from five to 12 years. Females accounted for 55% of the participants.

Inclusion criteria varied between studies. Eight studies included participants with chronic pain following spinal surgery or a previous diagnosis of 'failed back surgery syndrome' (FBSS) (Al‐Kaisy 2018; De Ridder 2013; Eldabe 2020; Hara 2022; Rigoard 2019; Schu 2014; Sokal 2020; Sweet 2016), while one study only recruited participants who had not had any surgery for back or leg pain (Kapural 2022). Three studies stated participants should have stable medication for pain control (De Ridder 2013; Perruchoud 2013; Schu 2014). Seven studies required participants to already be implanted with an SCS and have achieved stable pain control (De Ridder 2013; Eisenberg 2015; Eldabe 2020; Perruchoud 2013; Schu 2014; Sweet 2016; Wolter 2012).

Interventions

Nine studies included an intervention arm delivering a conventional frequency stimulation (De Ridder 2013; Eisenberg 2015; Eldabe 2020; Kumar 2007; Rigoard 2019; Schu 2014; Sokal 2020; Sweet 2016; Wolter 2012), five studies included an intervention arm delivering high‐frequency stimulation (Al‐Kaisy 2018; Kapural 2022; Perruchoud 2013; Sokal 2020; Sweet 2016), and five studies included an intervention arm delivering burst stimulation (De Ridder 2013; Eldabe 2020; Hara 2022; Schu 2014; Sokal 2020) (see Table 3 for intervention characteristics). In 10 studies, the experimental arms were compared against a placebo/sham stimulation arm of the trial where an SCS was implanted but was switched off or not discharging (Al‐Kaisy 2018; De Ridder 2013; Eisenberg 2015; Eldabe 2020; Hara 2022; Perruchoud 2013; Schu 2014; Sokal 2020; Sweet 2016; Wolter 2012). Only three of the placebo‐controlled trials involved implantation of a new SCS device (Al‐Kaisy 2018; Hara 2022; Sokal 2020). Three studies assessed SCS as an addition to trial care, labelled as "optimal medical management" or "conventional medical management" (Kumar 2007; Kapural 2022; Rigoard 2019). In these parallel‐group trials, although guidelines were provided for medical management, it appears that the care was not clearly controlled or reported on by the trialists. As such, we considered this comparison to have been between SCS plus medical management and medical management alone. While this is not strictly a 'no intervention' comparison according to our prespecified entry criteria for the review, we decided to err on the side of including these studies.

Open in table viewer
Table 3. Characteristics of SCS interventions in included studies

Study ID

Type of stimulation given

Device details

Electrode type/number

Stimulation parameters

Comparator

Details of pre‐implantation trial period

Duration of stimulation

Al‐Kaisy 2018

3 high‐frequency stimulation waveforms

Rechargeable implanted pulse generator produced by Medtronic (Minneapolis, MN, USA).

Dual octapolar leads (Octad, Medtronic, Minneapolis, MN, USA).

High‐frequency stimulation 1 included 5882 Hz for 30 μs; high‐frequency stimulation 2 included 3030 Hz for 30 μs; high‐frequency stimulation 3 included 1200 Hz for 180 μs

Placebo stimulation with the generator turned on and discharging, but without electricity transmitted to the lead

"All the recruited subjects received a trial of HF10 therapy for 7–14 days to assess efficacy and tolerability to the treatment. For every subject we initially activated a single bipole corresponding to the vertebral area of T9–T10, titrating up the HF10 SCS amplitude (1–5 mA range) during the first two to three days of the trial. If significant relief was not obtained (50%, but usually >70%), we activated a new bipole below the tested one for the following two to three days and, if again not successful, we moved to a new bipole higher than the one initially tested. At the end of the trial period, only those subjects reporting at least 50% or greater back pain VAS reduction from baseline were permanently implanted”

4 treatment arms, all of 3 weeks' duration

De Ridder 2013

Burst and conventional

Nonsterile EON IPG System (St. Jude Medical)

Externalised extension wires, Lamitrode tripole, 88, penta or 44

“Burst stimulation consists of intermittent packets of closely spaced, high‐frequency stimuli, for instance, 40‐Hz burst mode with five spikes at 500 Hz per burst, with a pulse width of 1 ms and 1 ms interspike intervals delivered in constant current mode. The cumulative charge of the five 1 ms spikes is balanced during 5 ms after the spikes.”

Conventional stimulation included tonic stimulation of 40 Hz or 50 Hz

Zero amplitude (IPG not discharging)

"During the mandatory period of external stimulation, which is a minimum of 28 days according to Belgian health care requirements for reimbursement, each patient was trialed by application of the classical tonic stimulation (40 or 50 Hz), burst stimulation with the same electrode configuration on separate days to prevent a carryover effect, and placebo. Patients were told they would receive three stimulation designs, some of which they might feel as paresthesias and some of which they might not feel as paresthesias. After an initial tonic programming session to define which electrodes needed activation as determined by paresthesia coverage, patients were programmed, lying down, randomly for 1 week with burst mode, 1 week in tonic mode, and 1 week with placebo”.

3 treatment arms, all of 1 week duration

Eisenberg 2015

Conventional

Conventional implanted device; temporary or permanent SCS implants

Not reported

Stimulator switched on or stimulator switched off

SCS device switched off

"Temporary or permanent SCS implants for the treatment of otherwise intractable unilateral radicular leg pain, after at least 1 back surgery was inclusion criteria for trial participation.”

30 minutes

Eldabe 2020

Conventional and burst

Medtronic’s rechargeable spinal cord stimulator, RestoreSensor

1 or 2 epidural leads

Conventional stimulation was a continuous tonic stimulation at 500 Hz with a pulse width of 480 μs. Burst stimulation was "40 Hz burst of four spikes of each 1000 μs at 500 Hz per burst".

The stimulator was switched off

“Achieved stable pain relief with conventional SCS (i.e., paraesthesia inducing stimulation with frequency < 150 Hz) using the Medtronic’s rechargeable spinal cord stimulator RestoreSensor® and with either 1 or 2 epidural leads was inclusion criteria for trial participation”.

3 treatment arms, each of 2 weeks' duration

Hara 2022

Burst

Precision Novi, Boston Scientific, Inc nonrechargeable implantable pulse generator

"A 16‐contact lead (Infinion CX, Boston Scientific, Inc) was implanted for unilateral leg pain or two 8‐contact leads (Linear ST, Boston Scientific, Inc) were implanted for bilateral leg pain"

"Closely spaced, high‐frequency stimuli delivered to the spinal cord. The simulus consisted of 40 Hz of constant current with 4 spikes per burst at an amplitude corresponding to 50% to 70% of paraesthesia perception threshold."

No stimulation provided

"Epidural surgical lead insertion was performed while patients were in the prone position using local anesthetics and mild intravenous sedation to enable patient feedback and cooperation. The aim was to optimize lead placement over the dorsal columns of the spinal cord so that paresthesia occurred in the targeted spinal dermatome (ie, tonic conventional stimulation). A 16‐contact lead (Infinion CX, Boston Scientific, Inc) was implanted for unilateral leg pain or two 8‐contact leads (Linear ST, Boston Scientific, Inc) were implanted for bilateral leg pain through a small skin incision at the L1/L2 or L2/L3 vertebral levels and placed in the epidural space at the T9/T10 [vertebral] level under fluoroscopic guidance. Intraoperative electrophysiological testing and stimulation were performed during longitudinal lead navigation. The leads were anchored at the optimal localization and their positions were confirmed with x‐ray imaging. Leads were then connected to an external neurostimulator using extension cords. Programming software (Illumina 3D, Boston Scientific, Inc) was used to optimize tonic conventional stimulation and determine paresthesia thresholds during the testing period. If there was insufficient improvement in leg pain during the testing period, the leads were removed and the patients were excluded. If there was sufficient improvement in leg pain during the testing period, the patients were included in the trial and their external neurostimulator was replaced with a nonrechargeable implantable pulse generator (Precision Novi, Boston Scientific, Inc) placed subcutaneously on the upper buttock or abdomen under local anesthesia. A nonrechargeable pulse generator was chosen to avoid unblinding of patients. Immediately after implantation of the stimulator, eligible patients underwent four 3‐month periods of treatment."

12 months: 4 periods of 3 months of treatment (6 months of SCS and 6 months of placebo)

Kumar 2007

Conventional

Implantable neurostimulation system produced by Medtronic (Synergy system, Medtronic, Inc., Minneapolis, MN)

Not specified

Mean (SD) settings were an amplitude of 3.7 V (2.0), a pulse width of 350 µs (95.5) and a rate of 49 Hz (16.4). Almost half (45%) of the participants required an amplitude of 4 V or more

”Non‐SCS therapy received by both groups was reviewed and actively managed, at the discretion of the study investigator and according to local clinical practice. Non‐SCS therapy included oral medications (i.e. opioid, non‐steroidal anti‐inflammatory drug, antidepressant, anticonvulsant/antiepileptic and other analgesic therapies), nerve blocks, epidural corticosteroids, physical and psychological rehabilitative therapy, and/or chiropractic care. The protocol excluded other invasive therapy, such as spinal surgery or implantation of an intrathecal drug delivery system”.

“All patients assigned to the SCS group underwent a screening trial. Those experiencing at least 80% overlap of their pain with stimulation‐induced paresthesia and at least 50% leg pain relief received an implantable neurostimulation system”

12 months

Kapural 2022

High frequency

IPG SCS system (Senza, Nevro Corp., Redwood City, CA, USA)

Two percutaneous leads with 8 contacts each placed in the epidural space spanning vertebral levels T8 to T11

10 kHz

“All subjects will continue with their CMM, defined as the best standard of care for each individual patient, as determined by the investigator.” Options included, but were not limited to: oral medications (including analgesic medication, nonsteroidal anti‐inflammatory drugs, neuromodulating agents, antidepressants); topical analgesics, compound creams, or counter‐irritants; combined physical and psychological management; physical therapy; back rehabilitation program; spinal manipulation and spinal mobilisation traction; acupuncture/acupressure; cognitive behavioural therapy; nerve blocks; epidural steroid injections; transcutaneous electrical nerve stimulation

Stimulation at a frequency of 10 kHz and pulse width of 30 μs delivered from an external pulse generator. The stimulation target and current amplitude was adjusted until at least 50% self‐reported back pain reduction from baseline achieved, defined as trial success, or until conclusion of the trial phase

6 months

Perruchoud 2013

High frequency

Medtronic (Minneapolis, MN, USA) impulse generator, either rechargeable (RestoreADVANCED, RestoreSensor, or RestoreUltra) or battery‐powered (PrimeADVANCED)

Not reported. No more than three active contacts

5000 Hz; with pulse width adjusted to 60 ms

The stimulator was switched off

"Currently implanted with suitable SCS device" was an inclusion criteria for study participants

2‐week periods of stimulation; 8 weeks study duration; i.e. 2 weeks current stimulation, 2 weeks high‐frequency (HF) or sham, 2 weeks current stimulation, 2 weeks HF or sham

Rigoard 2019

Conventional

Medtronic neurostimulator (model 97714, n = 49; 37702, n = 39;97702, n = 27; 37714, n = 12; 97712, n = 4; 37713, n = 3; 97713, n = 3; 37712, n = 2; and 37701, n = 1)

Multicolumn surgical lead (Specify 5‐6‐5; Medtronic)

20 Hz to 1200 Hz

"All patients received optimal medical management [OMM]. As part of the confirmation of eligibility (prior to randomization), the investigator and subject will determine an individual OMM treatment plan, which should include non‐investigational pharmacologic agents (for example, tricyclic antidepressants, opioid analgesics or tramadol, antiepileptics, or lidocaine) and/or interventional therapies (for example, therapeutic injections, radiofrequency, acupuncture, functional restoration, physical therapy, and psychological interventions, such as cognitive behavioural therapy) as appropriate. The following treatments are excluded from OMM: intrathecal drug delivery, peripheral nerve stimulation (not an approved indication in the United States), back surgery at the location related to the patient’s original back pain complaint, and experimental therapies."

“The screening test may be conducted with the Specify® 5‐6‐5 surgical lead or with a percutaneous lead(s). If successful, a SCS system will be implanted. A screening test will be determined to be successful if the subject finds the feeling of paresthesia acceptable and has adequate low back pain relief with usual activity and appropriate analgesia as assessed by the physician. Physicians can consider a conducting second screening test with the Specify® 5‐6‐5 lead if a screening test with a percutaneous lead led to inadequate paresthesia coverage of low back pain and/or painful extraneous stimulation (for example, chest wall pain, pressure or sharp mid‐back pain)”.

6 months (then allowed to cross to alternative trial arm and followed to 24 months)

Schu 2014

Conventional and burst

St. Jude Medical SCS system

SCS leads located at the mid‐thoracic position (T7–T10 vertebral level)

For conventional stimulation, 500 Hz mean pulse width ± SD under 500‐Hz tonic stimulation was 370.8 ± 135.4 μs, and mean amplitude ± SD was 5.5 ± 3.6 mA. For burst spinal cord stimulation, packets of five pulses (pulse width 1 ms) at 500 Hz, delivered 40 times per second

Device was switched off

Implanted "with a St. Jude Medical SCS system at least three months previously" was an inclusion criteria for study participants.

3 treatment arms, all of 1 week duration

Sokal 2020

High‐frequency, burst, and conventional

Non‐rechargeable IPG (Precision NoviTM) and in one case, a rechargeable IPG (MontageTM) produced by Boston Scientific Co.

Either one or two linear lead 8‐ or 16‐contact (Infinion 16TM) electrodes on vertebral levels T7–T10

High‐frequency stimulation was programmed with frequency of 1 kHz, pulse width of 120 s, and amplitude = 3 Amp. Burst stimulation delivered intermittent packets using the neural targeting algorithm, which consisted of several pulses per packet with pulse width 250–500 s repeated with frequency of 40 Hz. Conventional stimulation included tonic stimulation with frequencies typically between 40 Hz and 60 Hz. The pulse width ranged between 250 s and 500 s, and the amplitude produced comfortable paraesthesia

IPG was deactivated

Participants underwent 2 weeks of trial stimulation. During the trial period, tonic low‐frequency stimulation was used to check the coverage of pain area with paraesthesia induced by an external stimulator by adjusting the optimal settings of active electrode’s contacts. After a successful 14‐day trial period, participants who achieved at least a 50% reduction in pain were qualified to the second stage of the study, which involved the placement of a permanent internal pulse generator implantation under general anaesthesia.

4 treatment arms, all of 2‐week duration

Sweet 2016

Sub‐threshold high density (HD)

Medtronic RestoreSensor implanted pulse generator (Minneapolis, MN, USA).

“[Two] epidural 8‐contact Medtronic Compact percutaneous SureScan leads (electrode contacts 3 mm long and 1.3 mm diameter, 4 mm intercontact spacing) implanted in the midline with the end of the lead at the T7‐T8 [vertebral] interspace”

Subthreshold HD stimulation (1200 Hz, 200 μs, amplitude 90% of threshold for sensory percept)

"[Same] settings but amplitude 0 V"

“[One‐week] trial of subthreshold HD stimulation, defined as 1200 Hz frequency, 200 μs pulse width, and an amplitude 90% of the threshold for detection of a sensory percept. At the end of the week, each potential participant was asked about pain relief using the subthreshold parameters. Subjects were enrolled only if they reported significant pain relief using subthreshold HD stimulation, defined as 50% reduction in pain on the visual analog scale (VAS) compared with preoperative values.”

2 treatment arms, each of 2 weeks' duration and both preceded by 2 weeks of conventional stimulation

Wolter 2012

Conventional

“With one exception (patient 6), all patients had a non‐rechargeable implantable pulse generator (IPG). In patient 6, the battery state of the IPG was checked to rule out inadvertent discharge during the trial”. Stimulators were produced by Medtronic (n = 5) or Boston Scientific (n = 1).

All participants were implanted with percutaneous‐type electrodes

25 Hz to 100 Hz

The device was switched to zero

Prior SCS for at least 3 months with significant (> 50%) pain relief was an inclusion criterion for trial participation

2 treatment arms, each of 1‐week duration

CMM: conventional medical management; IPG: internal pulse generator or implantable pulse generator; SCS: spinal cord stimulation; µs: microseconds; ms: milliseconds

In the three studies assessing SCS as an addition to medical management, the medical management options varied between studies and the non‐SCS care actually received by participants in both groups was poorly reported. In the PROCESS trial (Kumar 2007), the medical management options were guided initially by investigators but were ultimately provided according to local clinical practice. As this was a multinational study, one would expect local clinical practice for back pain to vary considerably. Medical management in both groups could have included oral medications (i.e. opioid, non‐steroidal anti‐inflammatory drug, antidepressant, anticonvulsant/antiepileptic, and other analgesic therapies), nerve blocks, epidural corticosteroids, physical and psychological rehabilitative therapy, and/or chiropractic care. In the SENZA trial (Kapural 2022), participants continued the medical management they had been receiving and received a treatment plan from investigators. Medical management in both groups could have included oral medications (including analgesic medication, non‐steroidal anti‐inflammatory drugs, neuromodulating agents, antidepressants), topical analgesics, compound creams, or counter‐irritants, combined physical and psychological management, physical therapy, back rehabilitation program, spinal manipulation and spinal mobilisation, traction, acupuncture/acupressure, cognitive behavioral therapy, nerve blocks, epidural steroid injections, or transcutaneous electrical nerve stimulation. In the PROMISE trial (Rigoard 2019), an individual treatment plan was developed by investigators for each participant but the medical management was provided outside the trial. Out‐of‐trial care in both groups could have included noninvasive treatments such as acupuncture, psychological/ behavioural therapy, and physiotherapy, or invasive treatments such as spinal injections/blocks, epidural adhesiolysis, and neurotomies. None of these three studies clearly reported on the medical management provided either by investigators or as out‐of‐trial care.

Pre‐implantation trial periods

Studies that recruited participants without an SCS device already implanted tended to include a trial run‐in period (Al‐Kaisy 2018; Hara 2022; Kumar 2007; Kapural 2022; Rigoard 2019; Sokal 2020). Sweet 2016 included a run‐in period in people with implanted stimulators and receiving conventional SCS to identify those most likely to respond to high‐frequency SCS (of the 20 people recruited, only four responded and were included in the trial). Trial run‐in periods ranged from 14 to 28 days in studies with run‐in periods. The criteria used for successful completion of the trial period varied. Achieving a 50% reduction in pain was a requirement of most studies (Al‐Kaisy 2018; Kumar 2007; Kapural 2022; Sokal 2020). Hara 2022 required a 2‐point reduction in leg pain during the two‐week run‐in period to be included in their trial. In addition, one study required participants to have at least 80% coverage of their pain area with stimulation‐induced paraesthesia (Kumar 2007). One study stated the criteria as having adequate low back pain relief with usual activity and appropriate analgesia in the context of postoperative pain (Rigoard 2019). In one study (De Ridder 2013), the experimental trial was conducted during the SCS trial period.

In the three studies using a parallel‐group design (Kumar 2007; Kapural 2022; Rigoard 2019), the SCS trial period occurred after participants were randomised to their group. The SCS trial period success rate ranged from 82.7% to 92.5%. In the PROCESS study (Kumar 2007), 55.5% of those failing the SCS trial requested to still receive an SCS implant. In the PROMISE study (Rigoard 2019), participants who failed the SCS trial did not have an SCS implanted but were still followed as part of the study and included within the intention‐to‐treat analysis.

Outcomes

We present an Outcome Reporting Bias In Trials (ORBIT) matrix for the included studies in Table 2, with outcomes measured and level of reporting for each trial.

Major outcomes

Low back pain intensity

All thirteen trials measured mean low back pain intensity using a 0‐ to 10‐point or 0‐ to 100‐point visual analogue scale (VAS) or numeric rating scale (NRS). Six trials did not clearly report measures of variance (Al‐Kaisy 2018; De Ridder 2013; Eldabe 2020; Kumar 2007; Perruchoud 2013; Sweet 2016).

Function

Seven of thirteen trials measured function outcomes: six of the seven used the Oswestry Disability Index (ODI) questionnaire (Hara 2022; Kumar 2007; Kapural 2022; Rigoard 2019; Schu 2014; Sokal 2020); and one used the Pain Disability Index (Wolter 2012). Of these seven, three did not clearly report measures of variance (Kumar 2007; Sokal 2020; Wolter 2012). One study was registered and measured ODI at baseline, but it was unclear if ODI outcomes were collected at follow‐up (Al‐Kaisy 2018). The remaining five trials did not have prospective registry records or study protocols, so it was unclear if they measured function outcomes (De Ridder 2013; Eisenberg 2015; Eldabe 2020; Perruchoud 2013; Sweet 2016).

Health‐related quality of life

Seven of thirteen trials measured health‐related quality of life (Eldabe 2020; Hara 2022; Kumar 2007; Kapural 2022; Perruchoud 2013; Rigoard 2019; Sokal 2020), but only three fully reported their results (Eldabe 2020; Hara 2022; Rigoard 2019). Six trials used the EQ‐5D instrument (Eldabe 2020; Hara 2022; Kapural 2022; Perruchoud 2013; Rigoard 2019; Sokal 2020), and one used SF‐36 (Kumar 2007). One study planned to measure health‐related quality of life but did not provide results in the trial report (Sokal 2020).

Global assessment of efficacy (≥ 50% better)

Three of thirteen trials assessed the number of people who reported a 50% or higher improvement in pain (Kumar 2007; Kapural 2022; Rigoard 2019). One trial provided insufficient data at long‐term follow‐up for inclusion in a meta‐analysis (Kumar 2007).

Withdrawals due to adverse events

Six of thirteen trials reported on withdrawals due to adverse events (Al‐Kaisy 2018; Eldabe 2020; Kumar 2007; Kapural 2022; Perruchoud 2013; Rigoard 2019), although only two of the six provided complete data suitable for meta‐analysis (Eldabe 2020; Kapural 2022).

Adverse events

Eight of thirteen trials appeared to collect data on number of adverse events (Al‐Kaisy 2018; Eldabe 2020; Hara 2022; Kapural 2022; Kumar 2007; Rigoard 2019; Schu 2014; Sokal 2020), though reporting of proportions of adverse events in each study arm was generally poor. Several studies reported adverse events only for the as‐treated participants (Al‐Kaisy 2018; Kumar 2007; Schu 2014; Sokal 2020). Only two trials fully reported the number of adverse events in each study arm (Kapural 2022; Rigoard 2019).

Serious adverse events

Only one trial clearly reported serious adverse events in each study arm (Kapural 2022).

Minor outcomes

Medication use

Three trials reported on the number of participants using opioid medicines and daily morphine milligram equivalents (Kapural 2022; Rigoard 2019; Kumar 2007).

Health care use

No trials clearly reported on health care use.

Work status

One trial reported on the number of participants who returned to work (Kumar 2007).

Excluded studies

We excluded 29 studies for the following reasons: 19 due to an ineligible comparator; four because they were not RCTs; two because they included an ineligible study population; two due to an ineligible intervention; one due to ineligible outcomes; and one because it was terminated early. See Characteristics of excluded studies for details.

Risk of bias in included studies

We provide a summary of our judgements of the risk of bias in the included studies in Figure 2. Of the thirteen included trials, five (38%) were at risk of selection bias, ten (77%) were at risk of performance and detection bias, three (23%) were at risk of attrition bias, eleven (84%) were at risk of selective reporting bias, and twelve (92%) were at risk of other potential bias (Figure 3).


Risk of bias summary: review authors' judgements about each risk of bias item for each included study

Risk of bias summary: review authors' judgements about each risk of bias item for each included study


Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies

Allocation

We rated eight studies as having a low risk of selection bias, with appropriate methods described for both the generation and concealment of the allocation sequence (Al‐Kaisy 2018; Eldabe 2020; Hara 2022; Kumar 2007; Kapural 2022; Perruchoud 2013; Rigoard 2019; Schu 2014).

In one study (Eisenberg 2015), we considered generation of the randomisation sequence as low risk. However, authors provided no details on this sequence being kept offsite or otherwise blinded to the research team. As such, we considered this study as having an unclear risk of bias for allocation concealment.

We rated three studies as having an unclear risk of sequence generation (De Ridder 2013; Sokal 2020; Wolter 2012). The De Ridder 2013 and Wolter 2012 studies stated that participants were randomly assigned to groups, but provided no detail on the randomisation process. The Sokal 2020 study described how notes were drawn for group allocation by an independent examiner, but it was unclear whether this was a random process. We also rated Sokal 2020 as having an unclear risk of bias for allocation concealment because it did not provide details about whether the independent person was blinded to the sequence of treatment allocations. We also rated the De Ridder 2013 and Wolter 2012 studies as having an unclear risk of bias for allocation concealment because they provided no details on the method. We judged the remaining study as having a high risk of selection bias due to its very small sample and highly enriched design (only participants responding to high‐frequency SCS were included) (Sweet 2016).

Blinding

We judged three studies to be at low risk of performance and detection bias (Al‐Kaisy 2018; Hara 2022; Perruchoud 2013). All three studies described clear methods to ensure blinding to group allocation, and the investigators documented patient responses to show blinding was successful.

We considered all three parallel‐group studies to have a high risk of performance and detection bias due to the inability to blind participants or investigators to group allocation (Kumar 2007; Kapural 2022; Rigoard 2019). The process of implanting and managing the SCS meant that group allocation could not be concealed.

Of the other seven cross‐over trials, we rated four as having an unclear risk of performance and detection bias (Eldabe 2020; Schu 2014; Sweet 2016; Wolter 2012), and three as high risk (De Ridder 2013; Eisenberg 2015; Sokal 2020). In the studies rated as unclear risk, efforts were made to ensure blinding to treatment allocation by programming stimulation to sub‐sensory amplitude, but no detail was provided to confirm participants were not able to distinguish between trial arms to confirm these efforts were successful. Conversely, in the studies we rated as having a high risk of bias, participants were reported to experience paraesthesia during at least one of the active stimulation trial arms, allowing identification of stimulation phases. As a result, participants would be able to identify receipt of active treatment during this phase of the trial.

Incomplete outcome data

We judged ten studies to have a low risk of attrition bias (De Ridder 2013; Eisenberg 2015; Eldabe 2020; Hara 2022; Kumar 2007; Perruchoud 2013; Schu 2014; Sokal 2020; Sweet 2016; Wolter 2012). In six very small studies with immediate‐term follow‐up, all randomised participants completed all aspects of the trial and were included in the analysis (De Ridder 2013; Eisenberg 2015; Schu 2014; Sokal 2020; Sweet 2016; Wolter 2012). In the PROCESS trial (Kumar 2007), 88% of participants recruited were available for the 12‐month analysis and 87% of participants were available for the 24‐month analysis. Furthermore, the rate of attrition was comparable between groups. Eldabe 2020 lost three participants from an initial sample of 19 (16%), all of whom were reported as withdrawing from the study after different treatment exposures. Similarly, Perruchoud 2013 reported the loss of five participants from 38 randomised (13%), which was attributed to SCS lead breakage, battery exhaustion during the second treatment phase, pulse generator flipping, and the withdrawal of consent after randomisation.

We judged the remaining three trials to have a high risk of attrition bias (Al‐Kaisy 2018; Kapural 2022; Rigoard 2019). In Al‐Kaisy 2018, only 24 (80%) of the 30 participants randomised contributed to the analysis. Reasons for exclusion were given as early discontinuations, deviations associated with randomisation and programming affecting the ability to evaluate participant data, and lack of device use. However, there was insufficient detail to consider this study as having a low risk of bias in this domain.

We judged the PROMISE trial as high risk due to a marked difference in attrition between groups and limited justification provided for participant exclusion (Rigoard 2019). At the six‐month analysis, only one participant had discontinued in the study from the 'no intervention' group but 13 had discontinued from the SCS group. Furthermore, over the duration of the study, 21 participants (12 from the 'no intervention' group, 9 from the SCS group) were reported as being discontinued due to “withdrawal by the investigator”, with no further information provided to justify this exclusion.

At the six‐month assessment of the SENZA trial (Kapural 2022), one participant from the medical management group had been lost since randomisation but 17 had been lost from the SCS group. Although six of these participants were excluded due to an unsuccessful SCS trial, the other 11 were lost due to a mix of having withdrawn consent (n = 4), adverse events (n = 2), physician decision (n = 3), or were just reported as “lost to follow up” (n = 2). The lack of explanation of physician decisions to withdraw participants and the loss to follow‐up meant that we considered this trial as having a high risk of attrition bias.

Selective reporting

We considered only two studies to have a low risk of selective reporting bias (De Ridder 2013; Hara 2022). Of the remaining 11 studies, we rated three as having an unclear risk of selective reporting (Al‐Kaisy 2018; Kapural 2022; Rigoard 2019). In two studies (Al‐Kaisy 2018; Rigoard 2019), the trial registration did not fully match the information provided in the study report. In Al‐Kaisy 2018, the published report included outcomes (e.g. leg pain, adverse events) which had not been described in the trial registry. Conversely, in the Rigoard 2019 parallel‐group study, data on several outcomes were presented ‘as treated’ only, despite some participants switching from the group to which they were originally randomised after the six‐month follow‐up assessment. We also rated the SENZA trial as having an unclear risk of bias because it presented data only for the 'as treated' group (Kapural 2022).

We judged eight studies in total as having a high risk of reporting bias due to: not providing any details about trial registration (Eisenberg 2015; Eldabe 2020; Perruchoud 2013; Schu 2014; Wolter 2012); discrepancies between the trial registration and the study report (Sokal 2020); a lack of clarity in data provided (Perruchoud 2013); or retrospective publication of the trial protocol or registry (Kumar 2007; Sweet 2016). In Sokal 2020, the trial registry described use of the EuroQol group ‐ 5 Dimensions (EQ‐5D) for the assessment of quality of life; however, these data were not presented in the study report. For the Perruchoud 2013 study, in addition to providing no information on study registration, the authors also reported medication use and side‐effects as part of the study outcomes but provided no timings for these findings. We considered retrospective publication of the protocol of the PROCESS trial as representing a high risk of reporting bias: the study reported recruitment was completed in 2003 but the protocol was not published until 2005 (Kumar 2007).

Other potential sources of bias

We judged 12 of the 13 trials to be at high risk of 'other' sources of bias. Eight studies using a cross‐over design did not describe the methods they used to account for the carryover and period effects between the treatment phases of the study (De Ridder 2013; Eisenberg 2015; Eldabe 2020; Perruchoud 2013; Schu 2014; Sokal 2020; Sweet 2016; Wolter 2012). Hara 2022 reported accounting for carryover effects in their analysis and period effects in their study design (they implemented long (three month) intervention periods), but it was unclear if they formally tested for period effects. Thus, we rated this study as having an unclear risk of bias on this item. The only study that clearly reported accounting for both carryover and period effects in their design and analysis was the SCS‐Frequency trial (Al‐Kaisy 2018), which we assessed as having a low risk for other sources of bias. They collected outcome measures over the last three days of the final week from each of the three‐week cross‐over assignments to minimise the cross‐over effects from the previous phases, accounted for the paired nature of their data, and adjusted their analysis for multiple comparisons.

In the parallel‐group trials (Kumar 2007; Kapural 2022; Rigoard 2019), we considered the lack of placebo control to be an important source of bias, leading to a judgement of high risk of bias. Additionally, in all three trials, participants were given the option to switch between SCS and medical management after six months, which would bias any effects observed beyond six months.

Effects of interventions

See: Summary of findings 1 Spinal cord stimulation versus placebo for low back pain in adults

See summary of findings Table 1 for the main comparison of SCS versus placebo.

Comparison 1: SCS versus placebo

No trials assessed SCS versus placebo at long‐term follow‐up. Only the Hara 2022 study assessed the benefits of SCS versus placebo using a treatment period of longer than three weeks. We judged eight of the 10 placebo‐controlled trials to be sufficiently similar to warrant pooling of data in a meta‐analysis of immediate‐term outcomes (Al‐Kaisy 2018; De Ridder 2013; Eldabe 2020; Perruchoud 2013; Schu 2014; Sokal 2020; Sweet 2016; Wolter 2012). The one trial which we excluded from that meta‐analysis of immediate‐term outcomes – Eisenberg 2015, with 18 participants – measured outcomes on the same day of the experiment, a substantially shorter gap than the other studies. The longest duration of treatment in any placebo‐controlled trial of SCS for low back pain was six months (i.e. medium‐term follow‐up) (Hara 2022).

Benefits
Pain intensity

Low back pain at immediate‐term follow‐up

Based on data from eight trials (Al‐Kaisy 2018; De Ridder 2013; Eldabe 2020; Perruchoud 2013; Schu 2014; Sokal 2020; Sweet 2016; Wolter 2012), it is uncertain whether SCS improves low back pain intensity compared with placebo at immediate‐term follow‐up, because the certainty of the evidence was very low. At one month follow‐up, mean back pain (0 to 100; higher is worse) was 13.8 points better with SCS compared to placebo (95% CI 20.6 points better to 7.0 points better; I2 = 80%; 8 studies, 139 participants; very low‐certainty evidence; Analysis 1.1; Figure 4). Our sensitivity analysis found this effect was robust to removal of trials that were at high or unclear risk of selection bias (De Ridder 2013; Sokal 2020; Sweet 2016; Wolter 2012) (MD 10.0 points better, 95% CI 18.4 points better to 1.6 points better; I2 =76%; 4 studies, 96 participants), but not to removal of trials that were at high or unclear risk of detection bias (i.e. all but Al‐Kaisy 2018 and Perruchoud 2013). In trials at low risk of detection bias, there was no benefit with SCS in the immediate term (MD 3.00 points better, 95% CI 9.3 points better to 3.2 points worse; I2 =0%; 2 studies, 62 participants).


Comparison 1: spinal cord stimulation versus placebo. Outcome 1.1: low back pain intensity (0‐100) at immediate‐term follow‐up (< 1 month)

Comparison 1: spinal cord stimulation versus placebo. Outcome 1.1: low back pain intensity (0‐100) at immediate‐term follow‐up (< 1 month)

It is uncertain whether different types of SCS differed in efficacy. At one month, mean back pain was 11.4 points better with high‐frequency SCS (95% CI 23.7 points better to 0.8 points worse; I2 = 84%; 4 studies, 79 participants; very low‐certainty evidence; Analysis 1.1.1) compared with placebo. Conventional SCS may slightly improve low back pain intensity in the immediate term compared with placebo. At one month, mean back pain was 16.5 points better with conventional SCS (95% CI 23.6 points better to 9.5 points better; I2 = 46%; 6 studies, 82 participants; low‐certainty evidence; Analysis 1.1.2) compared with placebo. It is uncertain whether burst SCS improves low back pain intensity in the immediate term compared with placebo because the certainty of the evidence was very low. At one month, mean back pain was 13.5 points better with burst SCS (95% CI 32.6 points better to 5.6 points worse; I2 = 88%; 4 studies, 72 participants; very low‐certainty evidence; Analysis 1.1.3) compared with placebo.

Low back pain at medium‐term follow‐up

At six months, one trial provided moderate‐certainty evidence that SCS was probably not superior to placebo in reducing low back pain intensity (MD 4.00 points better, 95% CI 8.9 points better to 0.19 points worse; Analysis 1.2) (Hara 2022).

Leg pain at immediate‐term follow‐up

Two trials assessed benefits on leg pain intensity in the immediate term (Al‐Kaisy 2018; De Ridder 2013). It is uncertain whether SCS improves leg pain intensity in the immediate term compared with placebo. At one month, mean leg pain (0 to 100; higher is worse) was 10.0 points better with SCS (95% CI 20.3 points better to 0.3 points worse; I2 = 14%; 2 studies, 39 participants; very low‐certainty evidence; Analysis 1.3; Figure 5) compared with placebo. Our sensitivity analysis found this effect was not robust to removal of one study at unclear risk of selection bias and high risk of detection bias (De Ridder 2013). After omitting De Ridder 2013, the estimated effect on leg pain intensity in the immediate term approached zero (MD 3.8 points better, 95% CI 15.6 points better to 8.0 points worse; Analysis 1.3.1).


Comparison 1: spinal cord stimulation versus placebo. Outcome 1.3: leg pain intensity (0‐100) at immediate‐term follow‐up (< 1 month)

Comparison 1: spinal cord stimulation versus placebo. Outcome 1.3: leg pain intensity (0‐100) at immediate‐term follow‐up (< 1 month)

Leg pain at medium‐term follow‐up

One trial assessed this outcome in the medium term (Hara 2022). At six months, one trial provided moderate‐certainty evidence that SCS was probably not superior to placebo in reducing leg pain intensity (MD 2 points better, 95% CI 6.47 points better to 2.47 points worse; Analysis 1.4).

Function

Two placebo‐controlled trials reported on mean function (0 to 100; higher is better) in a way we could use in our analysis (Schu 2014; Hara 2022). It is uncertain whether SCS improves function compared with placebo in the immediate term, because the certainty of the evidence was very low. At one month, mean function was 15.1 points better with SCS compared with placebo (95% CI 4.5 points better to 25.7 points better; 1 study, 20 participants; very low‐certainty evidence; Analysis 1.5). Sweet 2016 provided function scores from the SF‐36 questionnaire in their trial of four participants, but we were unable to pool these results with other studies because they did not account for the paired nature of the data. In that study, there was no statistically significant effect of SCS compared to placebo on function at immediate‐term follow‐up. At six months, one trial provided moderate‐certainty evidence that SCS was probably not superior to placebo in improving function (Analysis 1.6) (Hara 2022).

Health‐related quality of life

Two placebo‐controlled trials, both at high risk of bias for selective reporting and 'other' bias, measured health‐related quality of life in the immediate term (Eldabe 2020; Perruchoud 2013). It is uncertain whether SCS improves health‐related quality of life compared with placebo, because the certainty of the evidence was very low. Both studies measuring health‐related quality of life suggested no benefit, though we were unable to pool the results of those studies. Perruchoud 2013 estimated there was no effect of high‐frequency SCS compared with placebo on health‐related quality of life in the immediate term, measured using the EQ‐5D (index scored from 0 to 1, 1 indicates full health; mean difference, adjusted for period effects, was 0.02; 95% CI –0.10 to 0.13; Analysis 1.7). At immediate‐term follow‐up, Eldabe 2020 reported the median EQ‐5D index scores to be 0.656 for placebo (interquartile range (IQR) 0.516 to 0.691), 0.620 for conventional SCS (IQR 0.516 to 0.691) and 0.516 for burst SCS (IQR 0.002 to 0.705). Sweet 2016 evaluated SF‐36 (role emotional) and found no statistically significant effect of SCS compared to placebo at immediate‐term follow‐up. At six months, one trial provided moderate‐certainty evidence that SCS was probably not superior to placebo in improving health‐related quality of life (MD 0.04, 95% CI ‐0.08 to 0.16; Analysis 1.8) (Hara 2022).

Global assessment of efficacy (≥ 50% better)

None of the placebo‐controlled trials reported on this measure of global assessment.

Harms
Withdrawals due to adverse events

Only one placebo‐controlled trial reported on withdrawals due to adverse events, by stimulation condition, at any time point (Eldabe 2020). It is uncertain whether SCS increases withdrawals due to adverse events compared with placebo at any time point, because the certainty of the evidence was very low. In their cross‐over trial of 19 participants, Eldabe 2020 reported two withdrawals during the placebo SCS phase, one withdrawal during the conventional SCS phase, and zero withdrawals during the burst SCS phase.

Adverse events

None of the placebo‐controlled trials clearly reported on the number of participants with any adverse event in each study arm. The certainty of the evidence for adverse events with SCS versus placebo at six weeks was very low (one trial, 19 participants) (Eldabe 2020). Eldabe 2020 provided a count of total adverse events associated with conventional SCS, burst SCS, and sham SCS. There were 15 adverse events during the two‐week conventional SCS period; 11 adverse events during the two‐week burst SCS period, and 12 adverse events during the two‐week sham SCS period. The most common adverse event was increased pain: 35% had increased pain with conventional SCS, 24% with burst SCS and 24% with sham SCS. Hara 2022 reported on adverse events after 12 months of placebo and burst SCS and found nine of 50 participants (18%) experienced adverse events, including infection.

Serious adverse events

None of the placebo‐controlled trials reported on serious adverse events by stimulation condition. Hara 2022 found four participants (8%) required surgical revision over 12 months. Two other studies, where participants received a new SCS implant, reported on the number of people requiring surgical revision in the short term. Al‐Kaisy 2018 found one of 24 participants (4.1%) required surgical revision at 12 weeks and Sokal 2020 found one of 18 participants (5.5%) required surgical revision at eight weeks. Serious adverse events included unintentional durotomy during lead placement, revision of leads, infection requiring surgery, infection requiring antibiotics, pulse generator replacement, and micturition problems.

Minor outcomes

None of the placebo‐controlled trials reported on medication use, health care use, or work status.

Comparison 2: SCS plus medical management versus medical management alone

Benefits
Pain intensity

No trials of SCS plus medical management versus medical management alone reported on mean low back pain intensity in both groups at long‐term follow‐up.

Low back pain at short‐term follow‐up

At short‐term follow‐up, one trial found the addition of SCS to medical management may slightly improve back pain intensity (Kumar 2007). At three months, mean back pain was 8.7 points better with the addition of SCS (95% CI 19.0 points better to 1.6 points worse; 1 study, 98 participants; low‐certainty evidence; Analysis 2.1).

Low back pain at medium‐term follow‐up

Three trials reported on mean low back pain intensity at medium‐term follow‐up (≥ 3 months to < 12 months) (Kapural 2022; Kumar 2007; Rigoard 2019). It is uncertain whether the addition of SCS to medical management reduces back pain intensity, because the certainty of the evidence was very low. In the medium term, mean pain was 26.0 points better with the addition of SCS, though the estimate was uncertain (95% CI 56.2 points better to 4.2 points worse; I2 = 98%; 3 studies, 430 participants; very low‐certainty evidence; Analysis 2.2; Figure 6). One trial, which added high‐frequency SCS to medical management, reported a very large effect size (Kapural 2022), which explained most of the heterogeneity. When we excluded this trial from the analysis, the estimated benefit with SCS at medium‐term follow‐up was 11.8 points on a 100‐point scale (95% CI 16.7 points better to 6.8 points better; I2 = 0%; 2 studies, 290 participants; Analysis 2.2.2)


Comparison 2: spinal cord stimulation plus medical management versus medical management alone. Outcome 2.2: low back pain intensity at medium‐term follow‐up (≥ 3 months to < 12 months)

Comparison 2: spinal cord stimulation plus medical management versus medical management alone. Outcome 2.2: low back pain intensity at medium‐term follow‐up (≥ 3 months to < 12 months)

Leg pain at short‐term follow‐up

One trial reported on benefits for leg pain intensity in the short term (Kumar 2007). The addition of SCS to medical management may improve leg pain intensity in the short term. At three months, mean leg pain intensity was 32.3 points better with the addition of conventional SCS (95% CI 42.3 points better to 22.3 points better; 1 study, 98 participants; low‐certainty evidence; Analysis 2.3).

Leg pain at medium‐term follow‐up

Two trials reported on benefits for leg pain intensity in the medium term (Kumar 2007; Rigoard 2019). It is uncertain whether adding SCS to medical management improves leg pain intensity in the medium term, because the certainty of the evidence was very low. In the medium term, mean leg pain intensity was 18.8 points better with the addition of SCS (95% CI 33.2 points better to 4.5 points better; I2 = 82%; 2 studies, 290 participants; very low‐certainty evidence; Analysis 2.4; Figure 7).


Comparison 2: spinal cord stimulation plus medical management versus medical management alone. Outcome 2.4: leg pain intensity at medium‐term follow‐up (≥ 3 months to < 12 months)

Comparison 2: spinal cord stimulation plus medical management versus medical management alone. Outcome 2.4: leg pain intensity at medium‐term follow‐up (≥ 3 months to < 12 months)

Function

No trials of SCS plus medical management versus medical management alone reported on function in both groups at long‐term follow‐up.

Three trials reported on mean function at medium‐term follow‐up, and one trial at short‐term follow‐up. Adding SCS to medical management may slightly improve function in the short term. At three months, mean function, measured on a 100‐point scale, was 12.6 points better with the addition of SCS (95% CI 20.1 points better to 5.2 points better; 1 study, 94 participants; low‐certainty evidence; Analysis 2.5). In the medium term, mean function was 16.2 points better with the addition of SCS (95% CI 19.4 points better to 13.0 points better; I2 = 95%; 3 studies, 430 participants; low‐certainty evidence; Analysis 2.6; Figure 8). As with the pain outcomes, most of the heterogeneity was due to one trial reporting a large effect size (Kapural 2022). When we excluded this trial from the analysis, the estimated benefit of SCS on function in the medium term was 7.7 points on a 100‐point scale (95% CI 11.8 points better to 3.6 points better; I2 = 15%; 2 studies, 290 participants; Analysis 2.6.2).


Comparison 2: spinal cord stimulation plus medical management versus medical management alone. Outcome 2.6: function at medium‐term follow‐up (≥ 3 months to < 12 months)

Comparison 2: spinal cord stimulation plus medical management versus medical management alone. Outcome 2.6: function at medium‐term follow‐up (≥ 3 months to < 12 months)

Health‐related quality of life

No trials of SCS plus medical management versus medical management alone reported on health‐related quality of life in both groups at long‐term follow‐up.

Two trials of conventional SCS reported on health‐related quality of life at medium‐term follow‐up. It is uncertain whether adding SCS to medical management improves health‐related quality of life in the medium term, because the certainty of the evidence was very low. In the medium term, mean health‐related quality of life, measured on a 100‐point scale, was 7.6 points better with SCS (95% CI 15.8 points better to 0.6 points worse; I2 = 53%; 2 studies, 289 participants; very low‐certainty evidence; Analysis 2.7; Figure 9).


Comparison 2: spinal cord stimulation plus medical management versus medical management alone. Outcome 2.7: health‐related quality of life at medium‐term follow‐up (≥ 3 months to < 12 months)

Comparison 2: spinal cord stimulation plus medical management versus medical management alone. Outcome 2.7: health‐related quality of life at medium‐term follow‐up (≥ 3 months to < 12 months)

Global assessment of efficacy (≥ 50% better)

One trial of SCS plus medical management versus medical management alone reported on global assessment of efficacy (≥ 50% improvement) in both groups at long‐term follow‐up (Kumar 2007). At their 24‐month follow‐up, Kumar 2007 estimated that 17 of 52 participants in the SCS group achieved 50% or better improvement compared with eight of 48 participants in the medical management group (risk ratio (RR) 1.96, 95% CI 0.93 to 4.12; very low‐certainty evidence; Analysis 2.9).

Three trials of SCS plus medical management versus medical management alone reported on the number of participants who perceived a 50% or better improvement in pain at medium‐term follow‐up. It is uncertain whether the addition of SCS increases the number of people reporting a 50% or better improvement in the medium term, because the certainty of the evidence was very low. In the medium term, participants receiving SCS were 7.4 times as likely to report a 50% or better improvement in pain with SCS compared with participants in the control group (95% CI 23.4 times more likely to 2.3 times more likely; I2 = 70%; 3 studies, 430 participants; very low‐certainty evidence; Analysis 2.8; Figure 10). Most of the heterogeneity could be explained by one trial reporting a very large effect size (Kapural 2022). When we excluded this trial from the analysis, the estimated risk ratio for having a 50% or better improvement in the medium term was 4.2 (95% CI 2.1 times more likely to report being a 50% or better improvement to 8.4 times more likely; I2 = 0%; 2 studies, 290 participants; Analysis 2.8.2).


Comparison 2: spinal cord stimulation plus medical management versus medical management alone. Outcome 2.8: global assessment of efficacy at medium‐term follow‐up (≥ 3 months to < 12 months)

Comparison 2: spinal cord stimulation plus medical management versus medical management alone. Outcome 2.8: global assessment of efficacy at medium‐term follow‐up (≥ 3 months to < 12 months)

Harms
Withdrawals due to adverse events

One trial reported on withdrawals due to adverse events (Kapural 2022). It is uncertain whether the addition of SCS increases the risk of withdrawals due to adverse events, because the certainty of the evidence was very low. In the medium term, two of 83 participants allocated to high‐frequency SCS withdrew due to adverse events compared with zero of 76 participants in the control group.

Adverse events

Two trials which added SCS to medical management reported on the proportion of participants who experienced at least one adverse event in each group (Kapural 2022; Rigoard 2019). It is uncertain whether the addition of SCS increases the risk of experiencing an adverse event because the certainty of the evidence was very low. In the medium term, 65 of 157 (41.4%) participants randomised to SCS plus medical management experienced an adverse event compared with 49 of 179 (27.4%) participants randomised to medical management alone (RR 2.32, 95% CI 0.39 to 13.79; I2 = 90%; 2 studies, 336 participants; very low‐certainty evidence; Analysis 2.11).

Kumar 2007 reported on adverse events at 12 and 24 months but did not include the proportion of participants who experienced at least one adverse event in each group. Adverse events in those receiving SCS at 12 months included: lead migration (eight of 84 participants, 10%), lead/extension fracture/torqued contacts (two of 84 participants, 2%); IPG migration (one of 84 participants, 1%); loss of therapeutic effect/unpleasant paraesthesia (six of 84 participants, 7%); technique‐related events such as incorrectly implanted electrode causing shocks and dural tears (four of 84 participants, 5%); infections (seven of 84 participants, 8%); pain at IPG/incision site (five of 84 participants, 6%); neurostimulator pocket fluid collection (four of 84 participants, 5%).

Serious adverse events

One trial reported on serious adverse events in both treatment arms (Kapural 2022). It is uncertain whether SCS increases the risk of serious adverse events compared with no SCS, because the certainty of the evidence was very low. In the medium term, six of 65 participants in a high‐frequency SCS group who were followed up experienced a serious adverse event compared with four of 75 in the control group (RR 1.73, 95% CI 0.51 to 5.87; one study, 140 participants; I2 = 0%; Analysis 2.12). Serious adverse events in the SCS arm included: osteomyelitis, severe lethargy, surgical revision/explant due to infection, and surgical revision/explant due to delayed wound healing. It was unclear how many participants receiving SCS required surgical revision due to device issues or adverse events.

Kumar 2007 reported on the number of participants requiring surgery due to an adverse event in those who received SCS at 12 months, but did not include proportions in each group that would allow estimation of risk. At 12 months, 20 of the 84 participants receiving SCS (24%) experienced a serious adverse event that required surgery to resolve. Of the 42 participants receiving SCS who were followed up at 24 months, 19 (45%) had experienced a total of 34 adverse events, and 13 (31%) had required surgical revision.

Rigoard 2019 found that of the 102 participants receiving SCS who were followed up at six months, 12 (12%) had required surgical revision.

Minor outcomes

Three trials of SCS plus medical management versus medical management alone reported on medication use at medium‐term follow‐up (Kapural 2022; Kumar 2007; Rigoard 2019). The addition of SCS to medical management may slightly reduce the proportion of participants taking opioid medicines in the medium term. At medium‐term follow‐up, the number of participants taking opioid medicines was 15% lower with SCS compared with no SCS (95% CI 27% lower to 0% lower; I2 = 0%; 2 studies, 290 participants; low‐certainty evidence; Analysis 2.13). The addition of SCS to medical management may slightly reduce daily morphine equivalents (MME) in the medium term. In the medium term, daily MMEs were 9.4 points lower with SCS compared with no SCS (95% CI 19.9 points lower to 1.2 points higher; I2 = 0%; 3 studies, 430 participants; low‐certainty evidence; Analysis 2.14).

One trial reported on the number returning to work (Kumar 2007). The addition of SCS to medical management may slightly increase the number of people returning to work. In the medium term, four of 52 participants in the SCS group had returned to work at medium‐term follow‐up compared with one of 48 participants in the control group (RR 3.7, 95% CI 0.4 to 31.9; 1 study, 100 participants; low‐certainty evidence; Analysis 2.15).

None of the trials clearly reported on health care use in the study groups.

Discussion

Summary of main results

SCS versus placebo

There is no evidence on the benefits or harms of SCS compared with placebo in the long term. Most trials have only assessed low back pain outcomes in the immediate term (8 trials, 139 participants). Based on these trials, it is uncertain whether SCS reduces back pain intensity compared with placebo because the certainty of the evidence is very low. Only one trial investigated the efficacy of SCS beyond three weeks of treatment. There is moderate‐certainty evidence that, at six months, SCS is not superior to placebo for pain, function, and health‐related quality of life outcomes.

Due to poor reporting in the included studies, we are uncertain to what extent SCS increases the risk of harms compared with placebo.

The uncertainty of the evidence was mostly due to study limitations, very small studies with imprecise estimates of effect, and inconsistency in the effects reported. None of the studies we located had a low risk of bias across all domains. It is also uncertain whether different types of SCS (i.e. burst, high‐frequency or conventional) differ in efficacy. An exploratory subgroup analysis of trials comparing high‐frequency SCS to placebo in the immediate term suggested no benefit. Analysis 1.1 included one study of four participants reporting a very large effect size (Sweet 2016), as well as the only two studies we considered to have achieved adequate blinding (Al‐Kaisy 2018; Perruchoud 2013). When we removed Sweet 2016 from that analysis, the I2 statistic dropped from 91% to 0% and the 95% confidence interval suggested no clinical benefit of high‐frequency SCS compared with placebo at less than one month of follow‐up.

SCS plus medical management versus medical management alone

We are uncertain whether the addition of SCS to medical management is beneficial for back pain intensity in the medium term because the certainty of the evidence was very low. Similarly, we are uncertain whether adding SCS to medical management improves function, leg pain intensity, global assessment of improvement, health‐related quality of life, return to work, or opioid medicine use, because the certainty of the evidence was very low.

Due to poor reporting in the included studies, we are uncertain to what extent adding SCS to medical management increases the risk of harms. The proportion of people receiving an SCS implant who were followed for up to two years and required surgical revision due to adverse events ranged from 11.7% to 30.9%.

Although the three parallel‐group trials that added SCS to medical management appeared to show clinical benefits for some outcomes, one study had results that differed widely from the other studies. When we excluded this study from analysis, the I2 statistic reduced to 0% and effect sizes were more modest (e.g. for low back pain intensity in the medium term, the effect was 12 points on a 100‐point scale, based on the two published trials). The two trials contributing to this estimate had critical study limitations. These included lack of blinding, attrition bias, and an enrichment‐type design where, after randomisation, the trialists excluded participants who did not respond to SCS, but did not take the same approach to those who did not respond in the control arm. This design feature essentially disrupts the benefits of randomisation, and the 12‐point benefit for pain in the medium term may therefore be an overestimate. Another challenge to the interpretation of these trials is the reporting of the medical management provided to both groups. From reading the trial reports, it is not possible to know precisely what medical management was provided or whether this care was consistently provided across the study groups. Any differences between the groups could be explained by differences in the medical management provided rather than the addition of SCS. We therefore suggest caution when interpreting the estimated benefits in trials of SCS compared with "conventional" or "optimal" medical management, as they were based on very low‐certainty evidence.

Overall completeness and applicability of evidence

Our findings likely apply to the typical person with low back pain with or without leg pain who is being considered for a new SCS intervention or for changes to parameters of a previously implanted stimulator. Studies included participants from 12 countries and mean age ranged from 48 years to 59 years. Mean back pain intensity at baseline was above 50 points on a 100‐point scale in six of 13 trials (range of mean back pain intensity at baseline: 36 points to 84 points on a 100‐point scale). Duration of low back pain at baseline was on average more than six months in all six trials reporting these data (range of mean pain duration at baseline: 5.1 years to 12.3 years).

Although some studies reported that both study arms received "optimal medical management", in no studies was this controlled as part of the trial or audited to ensure it was applied equally across groups. This means that the true benefit of adding SCS to optimal care consistent with clinical guidelines for low back pain remains unknown.

We found no evidence that newer approaches to SCS – for example, using burst or high‐frequency stimulation patterns – were superior to conventional SCS interventions.

Due to the small number of studies and participants, we were unable to determine if the estimated benefits and harms of SCS differ in subgroups of people with low back pain (e.g. people classified as having 'refractory neuropathic pain' or 'failed back surgery syndrome').

Certainty of the evidence

SCS versus placebo

We located no evidence for our primary comparison of SCS versus placebo at long‐term follow‐up.

For immediate‐term pain and function outcomes, the evidence was of very low certainty. We downgraded the evidence due to risk of bias (primarily from insufficient blinding, and potential for period and cross‐over effects), imprecision, inconsistency, and indirectness (eight out of ten studies did not assess medium‐ or long‐term efficacy). Given the small size of the effects observed in trials that we judged to be at high risk of bias, we consider it unlikely that future trials with a low risk of bias will show larger effects. Indeed, the only trial we rated as having an overall low risk of bias suggested no benefit of SCS. Hara 2022 provided moderate‐certainty evidence for medium‐term pain and function outcomes. We downgraded the evidence for medium‐term outcomes due to potential indirectness (we could not be certain the results of Hara 2022 could be applied to all types of SCS).

There was too little data on health‐related quality of life, global assessment of efficacy, and our minor outcomes (healthcare use, medication use, work status) to make any conclusions about benefits.

For harms in trials of SCS versus placebo, there were either no data (i.e. number of adverse events; number of serious adverse events, by treatment condition) or the certainty of the evidence was very low and based on only one very small study (i.e. withdrawals due to adverse events).

SCS plus medical management versus medical management alone

For pain and function outcomes in trials that added SCS to medical management, the certainty of the evidence ranged from low to very low. We downgraded the evidence due to bias (primarily due to lack of blinding and attrition bias), imprecision, and inconsistency.

There were sparse data on health‐related quality of life, global assessment of efficacy, and our minor outcomes (healthcare use, medication use, work status) from trial that added SCS to medical management, and the certainty of the evidence ranged from low to very low.

Most of the information on the incidence of harms was from one trial at high risk of bias. We are therefore very uncertain about the risk of harms when SCS is added to medical management.

Potential biases in the review process

We conducted a comprehensive search of major databases, clinical trials registries, and consulted with experts to try to ensure we identified all relevant trials. Two review authors independently performed key steps in the review process, including: assessing trials for inclusion, extracting data, conducting risk of bias assessments, and grading the certainty of the evidence. In all cases, a third review author adjudicated if there were discrepancies in judgements.

We identified 13 ongoing studies (ACTRN12620000720910; Ahmadi 2021; Al‐Kaisy 2020; ISRCTN10663814; ISRCTN33292457; NCT03419312; NCT03462147; NCT03718325; NCT03858790; NCT04479787; NCT04676022; NCT04732325; Reiter 2019). When published, the results from these trials may change the estimates from our analyses. However, given the small, inconsistent, immediate‐term‐only effects we observed in trials with biases that would tend to inflate effects (e.g. detection bias), we consider it unlikely that future well‐designed, placebo‐controlled trials will result in large, clinically important effects. Several ongoing trials are testing SCS versus medical management; it is unclear from their registry records whether these trials will overcome some of the important biases we identified in the parallel‐groups trials included in this review.

Agreements and disagreements with other studies or reviews

Two recent systematic reviews (Duarte 2020; O'Connell 2021), one of which was a Cochrane Review (O'Connell 2021), have examined the effects of SCS versus placebo in people with chronic pain. Neither review estimated effects in populations with low back pain. Duarte 2020 concluded that SCS leads to reduced pain intensity when compared to placebo. However, they did not grade the certainty of evidence. Our review, along with that of O'Connell 2021, suggests that the certainty of the evidence for the efficacy of SCS in the immediate term is very low and the effect size uncertain. In the medium term, a recent high‐quality trial not included in either of those reviews suggests that the true effect size of SCS over placebo is probably not clinically important (Hara 2022). Duarte 2020 did note that success of blinding probably influenced treatment effects observed in the placebo‐controlled trials. We also note that trials with adequate blinding in our analysis tended to produce lower estimates for benefits on back pain.

O'Connell 2021 reported low‐ to very low‐certainty evidence that, in people with chronic pain, SCS could provide clinically important benefits for pain intensity when added to conventional medical management or physical therapy. Our analysis of effects of adding SCS to medical management on low back pain intensity in the medium term was also based on very low‐certainty evidence and had 95% confidence intervals that included both a large benefit and no benefit at all. Together the reviews suggest that we are still very uncertain about the magnitude of clinical benefits of SCS as an addition to medical management. We would also point out that the reported large effects of SCS in some isolated studies were essentially observed in comparison to no treatment, and in trials at high risk of performance, detection, attrition and other biases, including uneven application of co‐interventions. We could not locate any study where trialists controlled and reported on the medical management provided to the study groups, to make it possible to estimate the benefit of adding SCS to conventional or "optimal" non‐SCS care. Therefore, the clinical benefit of adding SCS to optimal care for low back pain remains unknown.

Although we located almost no evidence on risk of adverse events with SCS versus placebo or no intervention, other studies have provided estimates of potential harms. In both ours and the O'Connell 2021 review, there was very low‐certainty evidence that the incidence of adverse events (e.g. infection) and serious adverse events (e.g. re‐operation) was higher with SCS than with no intervention, though the estimates were imprecise. A recent analysis of adverse events reported to the Australian Therapeutic Goods Administration found there were 520 adverse events reported between 2012 and 2019 of which 79% were "severe" and 13% were "life‐threatening" (Jones 2022). Future trials should report on the incidence of adverse events and serious adverse events in all study arms and at long‐term follow‐up to determine the risk of harms with SCS.

PRISMA study flow diagram

Figures and Tables -
Figure 1

PRISMA study flow diagram

Risk of bias summary: review authors' judgements about each risk of bias item for each included study

Figures and Tables -
Figure 2

Risk of bias summary: review authors' judgements about each risk of bias item for each included study

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies

Figures and Tables -
Figure 3

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies

Comparison 1: spinal cord stimulation versus placebo. Outcome 1.1: low back pain intensity (0‐100) at immediate‐term follow‐up (< 1 month)

Figures and Tables -
Figure 4

Comparison 1: spinal cord stimulation versus placebo. Outcome 1.1: low back pain intensity (0‐100) at immediate‐term follow‐up (< 1 month)

Comparison 1: spinal cord stimulation versus placebo. Outcome 1.3: leg pain intensity (0‐100) at immediate‐term follow‐up (< 1 month)

Figures and Tables -
Figure 5

Comparison 1: spinal cord stimulation versus placebo. Outcome 1.3: leg pain intensity (0‐100) at immediate‐term follow‐up (< 1 month)

Comparison 2: spinal cord stimulation plus medical management versus medical management alone. Outcome 2.2: low back pain intensity at medium‐term follow‐up (≥ 3 months to < 12 months)

Figures and Tables -
Figure 6

Comparison 2: spinal cord stimulation plus medical management versus medical management alone. Outcome 2.2: low back pain intensity at medium‐term follow‐up (≥ 3 months to < 12 months)

Comparison 2: spinal cord stimulation plus medical management versus medical management alone. Outcome 2.4: leg pain intensity at medium‐term follow‐up (≥ 3 months to < 12 months)

Figures and Tables -
Figure 7

Comparison 2: spinal cord stimulation plus medical management versus medical management alone. Outcome 2.4: leg pain intensity at medium‐term follow‐up (≥ 3 months to < 12 months)

Comparison 2: spinal cord stimulation plus medical management versus medical management alone. Outcome 2.6: function at medium‐term follow‐up (≥ 3 months to < 12 months)

Figures and Tables -
Figure 8

Comparison 2: spinal cord stimulation plus medical management versus medical management alone. Outcome 2.6: function at medium‐term follow‐up (≥ 3 months to < 12 months)

Comparison 2: spinal cord stimulation plus medical management versus medical management alone. Outcome 2.7: health‐related quality of life at medium‐term follow‐up (≥ 3 months to < 12 months)

Figures and Tables -
Figure 9

Comparison 2: spinal cord stimulation plus medical management versus medical management alone. Outcome 2.7: health‐related quality of life at medium‐term follow‐up (≥ 3 months to < 12 months)

Comparison 2: spinal cord stimulation plus medical management versus medical management alone. Outcome 2.8: global assessment of efficacy at medium‐term follow‐up (≥ 3 months to < 12 months)

Figures and Tables -
Figure 10

Comparison 2: spinal cord stimulation plus medical management versus medical management alone. Outcome 2.8: global assessment of efficacy at medium‐term follow‐up (≥ 3 months to < 12 months)

Comparison 1: Spinal cord stimulation (SCS) versus placebo, Outcome 1: Low back pain intensity (0‐100) at immediate‐term follow‐up (< 1 month)

Figures and Tables -
Analysis 1.1

Comparison 1: Spinal cord stimulation (SCS) versus placebo, Outcome 1: Low back pain intensity (0‐100) at immediate‐term follow‐up (< 1 month)

Comparison 1: Spinal cord stimulation (SCS) versus placebo, Outcome 2: Low back pain intensity (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Figures and Tables -
Analysis 1.2

Comparison 1: Spinal cord stimulation (SCS) versus placebo, Outcome 2: Low back pain intensity (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Comparison 1: Spinal cord stimulation (SCS) versus placebo, Outcome 3: Leg pain intensity (0‐100) at immediate‐term follow‐up (< 1 month)

Figures and Tables -
Analysis 1.3

Comparison 1: Spinal cord stimulation (SCS) versus placebo, Outcome 3: Leg pain intensity (0‐100) at immediate‐term follow‐up (< 1 month)

Comparison 1: Spinal cord stimulation (SCS) versus placebo, Outcome 4: Leg pain intensity (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Figures and Tables -
Analysis 1.4

Comparison 1: Spinal cord stimulation (SCS) versus placebo, Outcome 4: Leg pain intensity (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Comparison 1: Spinal cord stimulation (SCS) versus placebo, Outcome 5: Function (0‐100) at immediate‐term follow‐up (< 1 month)

Figures and Tables -
Analysis 1.5

Comparison 1: Spinal cord stimulation (SCS) versus placebo, Outcome 5: Function (0‐100) at immediate‐term follow‐up (< 1 month)

Comparison 1: Spinal cord stimulation (SCS) versus placebo, Outcome 6: Function (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Figures and Tables -
Analysis 1.6

Comparison 1: Spinal cord stimulation (SCS) versus placebo, Outcome 6: Function (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Comparison 1: Spinal cord stimulation (SCS) versus placebo, Outcome 7: Health‐related quality of life (0‐1 index) at immediate‐term follow‐up (< 1 month)

Figures and Tables -
Analysis 1.7

Comparison 1: Spinal cord stimulation (SCS) versus placebo, Outcome 7: Health‐related quality of life (0‐1 index) at immediate‐term follow‐up (< 1 month)

Comparison 1: Spinal cord stimulation (SCS) versus placebo, Outcome 8: Health‐related quality of life (0‐1 index) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Figures and Tables -
Analysis 1.8

Comparison 1: Spinal cord stimulation (SCS) versus placebo, Outcome 8: Health‐related quality of life (0‐1 index) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 1: Low back pain intensity (0‐100) at short‐term follow‐up (≥ 1 mo to < 3 mo)

Figures and Tables -
Analysis 2.1

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 1: Low back pain intensity (0‐100) at short‐term follow‐up (≥ 1 mo to < 3 mo)

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 2: Low back pain intensity (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Figures and Tables -
Analysis 2.2

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 2: Low back pain intensity (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 3: Leg pain intensity (0‐100) at short‐term follow‐up (≥ 1 mo to < 3 mo)

Figures and Tables -
Analysis 2.3

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 3: Leg pain intensity (0‐100) at short‐term follow‐up (≥ 1 mo to < 3 mo)

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 4: Leg pain intensity (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Figures and Tables -
Analysis 2.4

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 4: Leg pain intensity (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 5: Function (0‐100) at short‐term follow‐up (≥ 1 mo to < 3 mo)

Figures and Tables -
Analysis 2.5

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 5: Function (0‐100) at short‐term follow‐up (≥ 1 mo to < 3 mo)

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 6: Function (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Figures and Tables -
Analysis 2.6

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 6: Function (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 7: Health‐related quality of life (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Figures and Tables -
Analysis 2.7

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 7: Health‐related quality of life (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 8: Global assessment of efficacy at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Figures and Tables -
Analysis 2.8

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 8: Global assessment of efficacy at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 9: Global assessment of efficacy at long‐term follow‐up (≥ 12 mo)

Figures and Tables -
Analysis 2.9

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 9: Global assessment of efficacy at long‐term follow‐up (≥ 12 mo)

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 10: Withdrawals due to adverse events at longest follow‐up

Figures and Tables -
Analysis 2.10

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 10: Withdrawals due to adverse events at longest follow‐up

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 11: Proportion with any adverse event at longest follow‐up

Figures and Tables -
Analysis 2.11

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 11: Proportion with any adverse event at longest follow‐up

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 12: Proportion with serious adverse event at longest follow‐up

Figures and Tables -
Analysis 2.12

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 12: Proportion with serious adverse event at longest follow‐up

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 13: Medication use 1 (number (%) taking opioid medicines) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Figures and Tables -
Analysis 2.13

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 13: Medication use 1 (number (%) taking opioid medicines) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 14: Medication use 2 (daily MME) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Figures and Tables -
Analysis 2.14

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 14: Medication use 2 (daily MME) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 15: Work status 1 (number returned to work) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Figures and Tables -
Analysis 2.15

Comparison 2: Spinal cord stimulation (SCS) plus medical management versus medical management alone, Outcome 15: Work status 1 (number returned to work) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

Summary of findings 1. Spinal cord stimulation versus placebo for low back pain in adults

Spinal cord stimulation (SCS) versus placebo for low back pain in adults

Patient or population: adults with low back pain

Setting: outpatient

Intervention: conventional, burst, or high‐frequency SCS

Comparison: placebo

Outcomes

Anticipated absolute effects (95% CI)

Relative effect

No. of participants (studies)

Certainty of the evidence (GRADE)

Comments

Risk with placebo

Risk with SCS

Pain intensity

VAS, translated to a 0‐ to 100‐point scale, where 0 is no pain

Medium‐term follow‐up (≥ 3 months to < 12 months)

Mean back pain during placebo period was 61 points

Mean back pain was 4 points better (8.2 points better to 0.2 points worse)

50 participants (1 study)

Moderatea

SCS probably does not improve back or leg pain in the medium term. Data are based on a single trial of burst SCS at low risk of bias. The CIs excluded clinically important benefits.

Eight of 10 available placebo‐controlled trials measured low back pain outcomes in the immediate‐term only. Based on those trials, it was uncertain whether SCS improves low back pain more than placebo in the immediate term (8 studies, 139 participants; very low‐certainty evidence).

Two trials measured leg pain in the immediate term. Based on those two trials, it was uncertain whether SCS improves leg pain more than placebo in the immediate term (2 studies, 39 participants; very low‐certainty evidence).

Function

Roland‐Morris Disability Questionnaire &

Oswestry Disability Index translated to a 0‐ to 100‐point scale, where 0 is no disability or best function

Medium‐term follow‐up (≥ 3 months to < 12 months)

Mean disability during placebo period was 35.4 points

Mean disability was 1.3 points better (3.9 points better to 1.3 points worse)

50 participants (1 study)

Moderatea

SCS probably does not improve function in the medium term. Data are based on a single trial of burst SCS at low risk of bias. The CIs excluded clinically important benefits.

One other study measured function in the immediate‐term only. Based on that trial, it was uncertain whether SCS improves function more than placebo in the immediate term (1 study, 20 participants; very low‐certainty evidence).

Health‐related quality of life

EQ‐5D, index from 0 to 1 where 0 is worst quality of life

Medium‐term follow‐up (≥ 3 months to < 12 months)

Mean quality of life during placebo period was 0.44 points out of 1

Mean quality of life was 0.04 points better (0.16 points better to 0.08 points worse)

50 participants (1 study)

Moderatea

SCS probably provides little to no benefit for health‐related quality of life in the medium term. Data are based on a single trial of burst SCS at low risk of bias. The CIs excluded clinically important benefits.

Two other trials measured health‐related quality of life in the immediate‐term only. Both suggested no benefit, though we were unable to pool the results of those studies (2 studies, 52 participants; very low‐certainty evidence).

Global assessment of efficacy

≥ 50% improvement in pain

Medium‐term follow‐up (≥ 3 months to < 12 months)

Not estimable

Not estimable

(0 studies)

No data available

Withdrawals due to adverse events

Follow‐up: longest measuredb

Not estimable

Not estimable

(0 studies)

Very lowe

Poorly reported in included studies. We are uncertain whether SCS results in more people withdrawing due to adverse events.

One small cross‐over RCT with 6‐week follow‐up reported 2 withdrawals with placebo versus 1 withdrawal with SCS (1 study, 19 participants; very low‐certainty evidence).

Adverse eventsc

Follow‐up: longest measuredb

Not estimable

Not estimable

(0 studies)

Very lowe

Poorly reported in included studies. One cross‐over study at low risk of bias found 9 out of 50 (18%) people who received SCS experienced an adverse event over a 12‐month period, but did not specify whether events occurred during the placebo or active SCS period.

Serious adverse eventsd

Follow‐up: longest measuredb

Not estimable

Not estimable

(0 studies)

Very lowe

Poorly reported in included studies. Although the incidence was uncertain, serious adverse events included infections, neurological damage, and lead migration requiring repeated surgery. One placebo‐controlled study at low risk of bias found 4 out of 50 (8%) people who received SCS required surgical revision within 12 months.

In the six trials in this review that followed people receiving a new SCS implant, surgical revision rates in the SCS group due to adverse events ranged from 4.1% at 8 weeks to 30.9% at 24 months.

CI: confidence interval; VAS: visual analogue scale

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.

aDowngraded one level for indirectness due to possible differences between the burst SCS regimen provided in the trial and other SCS regimens provided internationally.
bLong‐term efficacy and safety were not estimable as no data were reported.
cAdverse events included increased pain, infection, unpleasant paraesthesia, incorrectly implanted electrode causing shocks, pain at internal pulse generator/incision site, neurostimulator pocket fluid collection.
dSerious adverse events included unintentional dural tears during lead placement, revision of leads, infection requiring surgery, pulse generator replacement, and micturition problems requiring explant or revision surgery.
eDowngraded one level for risk of bias, one level for imprecision, and one level for indirectness.

Figures and Tables -
Summary of findings 1. Spinal cord stimulation versus placebo for low back pain in adults
Table 1. Summary data extracted from cross‐over trials and methods used to estimate mean difference and standard error

Analysis

Study

Mean (intervention)

SD (intervention)

N (intervention)

Mean (placebo)

SD (placebo)

N (Placebo)

Effect

size (mean

difference)

SE

Effect size adjusted for cross‐over design?

Effect size adjusted

for multiple

comparisons

to placebo group?

Notes

1.1 SCS versus placebo SCS, Outcome 1: Low back pain intensity (0‐100) at immediate‐term follow‐up (< 1 month)

1.1.1

Al‐Kaisy 2018 (High‐frequency SCS)(1)

45.1

18.7

24

48.3

24.5

8

‐3.2

9.47

No

Yes

Mean difference and SE estimated from 2 sample t‐test with unequal variance adjusted for multiplicity by dividing N of placebo period by 3.

Mean and SD were rescaled (x10)

1.1.1

Al‐Kaisy 2018 (High‐frequency SCS)(2)

45.7

20.7

24

48.3

24.5

8

‐2.6

9.66

No

Yes

Mean difference and SE estimated from 2 sample t‐test with unequal variance adjusted for multiplicity by dividing N of placebo period by 3. Mean and SD were rescaled (x10)

1.1.1

Al‐Kaisy 2018 (High‐frequency SCS)(3)

32.2

19.8

24

48.3

24.5

8

‐16.1

9.56

No

Yes

Mean difference and SE estimated from 2 sample t‐test with unequal variance adjusted for multiplicity by dividing N of placebo period by 3. Mean and SD were rescaled (x10)

1.1.1

Perruchoud

2013 (High‐frequency SCS)

43.5

19.2

33

42.6

21.4

33

‐0.9

3.93

Yes. Means were from within subjects model

N/A

Effect size = ‐0.09 (95% CI ‐0.68 to 0.86) reported in paper. SE calculated from CI. Effect size and CI were rescaled (x10). Adjustment for period effects not required

1.1.1

Sokal 2020 (High‐frequency SCS)

51.7

14

18

54.2

12.2

18

‐1.7

2.20

Yes, using regression

weights ( β = ‐0.17) and SD of individual

regression

weight (τ = 0.68) provided by authors in Table A1

No

Effect size = ‐0.17, SE = 0.22 reported in paper Table A1. Effect size and SE were rescaled (x10). Effect size estimates are adjusted for cross‐over. Unclear if multiplicity was accounted for.

1.1.1

Sweet 2016 (High‐frequency SCS)

22.9

4.1

4

63.1

12.2

2

‐50.1

6.44

Adjusted for cross‐over, period and sequence effects

Yes

Patient level scores were digitally extracted from Figure 3. To estimate effect size, a mixed‐effects model was fitted accounting for cross‐over, period and sequence effects

1.1.2

De Ridder 2013 (Conventional SCS)

51.5

15

59.5

15

‐7.8

12.30

No

No

Mean estimates digitally extracted from Figure 3. Mean difference was calculated, and SE was assumed equal to burst SCS estimate from De Ridder. Results were rescaled (x10).

1.1.2

Eldabe 2020 (Conventional SCS)

51.0

19

38.0

19

‐12.8

3.9

Yes

No

Means extracted from report. Mean percentage reduction and confidence interval were reported. These were converted to absolute values and rescaled (x10).

1.1.2

Schu 2014 (Conventional SCS)

71

19

20

83

11

10

‐12.0

5.49

No

Yes

Mean difference and SE estimated from 2 sample t‐test with unequal variance adjusted for multiplicity by dividing N of placebo period by 2.

1.1.2

Sokal 2020 (Conventional SCS)

41.8

17.6

18

54.2

12.2

6

‐9.9

5.6

Yes

Yes

Effect size = ‐0.99, SE = 0.56 reported in paper Table A1. Effect size and SE were rescaled (x10)

1.1.2

Sweet 2016 (Conventional SCS)

53.2

6.3

4

63.1

12.2

2

‐31.2

7.2

Yes

No

Patient level scores were digitally extracted from Figure 3. To estimate effect size, a mixed‐effects model was fitted, accounting for cross‐over, period and sequence effects

1.1.2

Wolter 2012 (Conventional SCS)

56.8

22.4

6

63.7

20

6

‐28.5

6.4

Yes

N/A

Patient level scores were reported in Table 3. A paired 2 sample t‐test was performed (accounts for carryover).

Estimates were rescaled (x10)

1.1.3

DeRidder 2013 (Burst SCS)

35.5

15

59.5

15

‐24.1

12.3

No

No

Mean estimates digitally extracted from Figure 3. Difference between burst SCS and placebo was reported statistically significant at 0.05 threshold. Mean difference was calculated, and conservatively assuming P = 0.05 allowed calculation of the standard error for back pain.

Results were rescaled (x10)

1.1.3

Eldabe 2020 (Burst SCS)

54

19

51

19

2.55

5.2

Yes

No

Means extracted from report. Mean percentage reduction and confidence interval were reported. These were converted to absolute values and rescaled (x10).

1.1.3

Schu 2014 (Burst SCS)

47

25

20

83

11

10

‐36

6.58

Yes

Yes

Mean difference and SE estimated from 2 sample t‐test with unequal variance adjusted for multiplicity by dividing N of placebo period by 2.

1.1.3

Sokal 2020 (Burst SCS)

52.7

13.3

18

54.2

12.2

6

‐0.3

3.7

Yes

Yes

Effect size = ‐0.03, SE = 0.37 reported in paper Table A1. Effect size and SE were rescaled (x10). Effect size estimates are adjusted for cross‐over. Unclear if multiplicity was accounted for.

1.4 SCS versus placebo SCS, Outcome 1: Low back pain intensity (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

1.4.1

Hara 2022 (Burst SCS)

57

50

61

50

‐4.0

2.14

Yes

N/A

Mean difference and confidence intervals were reported in Table 2. SE was calculated from confidence interval. Results were rescaled (x10)

1.2 SCS versus placebo SCS, Outcome 2: Function (0‐100) at immediate‐term follow‐up (< 1 month)

1.2.2

Schu 2014 (Conventional SCS)

49.2

14.6

20

59

20.6

10

‐9.8

7.29

Yes

Yes

Mean difference and SE estimated from 2 sample t‐test with unequal variance adjusted for multiplicity by dividing N of placebo by 2.

Results were rescaled (x2).

1.2.3

Schu 2014 (Burst SCS)

38.4

16

20

59

20.6

10

‐20.6

7.43

Yes

Yes

Mean difference and SE estimated from 2 sample t‐test with unequal variance adjusted for multiplicity by dividing N of placebo by 2.

Results were rescaled (x2).

1.5 SCS versus placebo SCS, Outcome 2: Function (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

1.5.1

Hara 2022 (Burst SCS)

34.0

50

35.4

50

‐1.3

1.33

Yes

N/A

Mean difference and confidence intervals are reported in Table 2. SE was calculated from confidence interval.

1.6 SCS versus placebo SCS, Outcome 3: Health‐related quality of life (0‐1) at immediate‐term follow‐up (<1 month)

1.6.1

Perruchoud 2013 (High‐frequency SCS)

0.48

33

0.46

33

0.017

0.0602

Yes

N/A

Means were from within subjects model. Effect size = 0.017 (95% CI ‐0.101 to 0.135) extracted from report. SE calculated from CI.

1.8 SCS versus placebo SCS, Outcome 3: Health‐related quality of life (0‐1) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

1.8.1

Hara 2022 (Burst SCS)

0.48

50

0.44

50

0.04

0.0632

Yes

N/A

Mean difference and confidence intervals were reported in Table 2. SE was calculated from confidence interval.

1.3 SCS versus placebo SCS, Outcome 4: Leg pain intensity (0‐100) at immediate‐term follow‐up (<1 month)

1.3.1

Al‐Kaisy 2018 (High‐frequency SCS)(1)

18.1

25.5

24

25.1

25.5

8

‐1.4

10.41

No

Yes

Mean difference and SE estimated from 2 sample t‐test with unequal variance adjusted for multiplicity by dividing N of placebo group by 3.

Mean and SD were rescaled (x10). Follow‐up SD values not reported; taken from baseline

1.3.1

Al‐Kaisy 2018 (High‐frequency SCS)(2)

23.7

25.5

24

25.1

25.5

8

‐3.1

10.41

No

Yes

Mean difference and SE estimated from 2 sample t‐test with unequal variance adjusted for multiplicity by dividing N of placebo group by 3.

Mean and SD were rescaled (x10). Follow‐up SD values not reported; taken from baseline

1.3.1

Al‐Kaisy 2018 (High‐frequency SCS)(3)

22

25.5

24

25.1

25.5

8

‐7.0

10.41

No

Yes

Mean difference and SE estimated from 2 sample t‐test with unequal variance adjusted for multiplicity by dividing N of placebo group by 3.

Mean and SD were rescaled (x10). Follow‐up SD values not reported; taken from baseline

1.3.2

DeRidder 2013 (Conventional SCS)

36

15

66

15

‐30.1

15.3

No

No

Mean estimates digitally extracted from Figure 3 in report. Mean difference was calculated, and SE was assumed equal to burst SCS leg pain estimate from De Ridder.

Results were rescaled (x10).

1.3.3

DeRidder 2013 (Burst SCS)

36

15

66

15

‐30.1

15.3

No

No

Mean estimates digitally extracted from Figure 3. Difference between burst and placebo was reported statistically significant at 0.05 threshold. Mean difference was calculated, and conservatively assuming P = 0.05 allowed calculation of the standard error for leg pain.

Results were rescaled (x10).

1.7 SCS versus placebo SCS, Outcome 4: Leg pain intensity (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo)

1.7.1

Hara 2022 (Burst SCS)

59

50

61

50

‐2.0

2.28

Yes

N/A

Mean difference and confidence intervals were reported in Table 2. SE was calculated from confidence interval. Results were rescaled (x10).

Figures and Tables -
Table 1. Summary data extracted from cross‐over trials and methods used to estimate mean difference and standard error
Table 2. Outcome Reporting Bias In Trials (ORBIT) matrix

Study ID

Low back pain intensity

Function

Health‐related quality of life

Global assessment (≥ 50% better)

Withdrawals due to adverse events

% with adverse events

% with serious adverse events

Al‐Kaisy 2018

Partial

?

?

?

Partial

Partial

Partial

De Ridder 2013

Partial

?

?

?

?

?

?

Eisenberg 2015

Full

?

?

?

?

?

?

Eldabe 2020

Partial

?

Full

?

Full

Partial

?

Hara 2022

Full

Full

Full

Not measured

Full

Partial

Partial

Kumar 2007

Partial

Partial

Partial

Partial

Partial

Partial

Measured

Kapural 2022

Full

Full

Partial

Full

Full

Full

Full

Perruchoud 2013

Partial

?

Partial

?

Partial

?

?

Rigoard 2019

Full

Full

Full

Full

Partial

Full

Partial

Schu 2014

Full

Full

?

?

?

Partial

Partial

Sokal 2020

Full

Partial

Measured

?

?

Measured

?

Sweet 2016

Full

Partial

Partial

?

?

?

?

Wolter 2012

Full

Partial

?

?

?

?

?

'Full': sufficient data for inclusion in a meta‐analysis were reported (e.g. mean, standard deviation, sample size per group for continuous outcomes).
'Partial': insufficient data for inclusion in a meta‐analysis were reported (e.g. means only, with no measures of variance).
'Measured': outcome was measured but no outcome data were reported.
'Not measured': outcome was not measured by trialists.
'?': unclear whether the outcome was measured or not (as a trial protocol or prospective study registry entry was unavailable).

Figures and Tables -
Table 2. Outcome Reporting Bias In Trials (ORBIT) matrix
Table 3. Characteristics of SCS interventions in included studies

Study ID

Type of stimulation given

Device details

Electrode type/number

Stimulation parameters

Comparator

Details of pre‐implantation trial period

Duration of stimulation

Al‐Kaisy 2018

3 high‐frequency stimulation waveforms

Rechargeable implanted pulse generator produced by Medtronic (Minneapolis, MN, USA).

Dual octapolar leads (Octad, Medtronic, Minneapolis, MN, USA).

High‐frequency stimulation 1 included 5882 Hz for 30 μs; high‐frequency stimulation 2 included 3030 Hz for 30 μs; high‐frequency stimulation 3 included 1200 Hz for 180 μs

Placebo stimulation with the generator turned on and discharging, but without electricity transmitted to the lead

"All the recruited subjects received a trial of HF10 therapy for 7–14 days to assess efficacy and tolerability to the treatment. For every subject we initially activated a single bipole corresponding to the vertebral area of T9–T10, titrating up the HF10 SCS amplitude (1–5 mA range) during the first two to three days of the trial. If significant relief was not obtained (50%, but usually >70%), we activated a new bipole below the tested one for the following two to three days and, if again not successful, we moved to a new bipole higher than the one initially tested. At the end of the trial period, only those subjects reporting at least 50% or greater back pain VAS reduction from baseline were permanently implanted”

4 treatment arms, all of 3 weeks' duration

De Ridder 2013

Burst and conventional

Nonsterile EON IPG System (St. Jude Medical)

Externalised extension wires, Lamitrode tripole, 88, penta or 44

“Burst stimulation consists of intermittent packets of closely spaced, high‐frequency stimuli, for instance, 40‐Hz burst mode with five spikes at 500 Hz per burst, with a pulse width of 1 ms and 1 ms interspike intervals delivered in constant current mode. The cumulative charge of the five 1 ms spikes is balanced during 5 ms after the spikes.”

Conventional stimulation included tonic stimulation of 40 Hz or 50 Hz

Zero amplitude (IPG not discharging)

"During the mandatory period of external stimulation, which is a minimum of 28 days according to Belgian health care requirements for reimbursement, each patient was trialed by application of the classical tonic stimulation (40 or 50 Hz), burst stimulation with the same electrode configuration on separate days to prevent a carryover effect, and placebo. Patients were told they would receive three stimulation designs, some of which they might feel as paresthesias and some of which they might not feel as paresthesias. After an initial tonic programming session to define which electrodes needed activation as determined by paresthesia coverage, patients were programmed, lying down, randomly for 1 week with burst mode, 1 week in tonic mode, and 1 week with placebo”.

3 treatment arms, all of 1 week duration

Eisenberg 2015

Conventional

Conventional implanted device; temporary or permanent SCS implants

Not reported

Stimulator switched on or stimulator switched off

SCS device switched off

"Temporary or permanent SCS implants for the treatment of otherwise intractable unilateral radicular leg pain, after at least 1 back surgery was inclusion criteria for trial participation.”

30 minutes

Eldabe 2020

Conventional and burst

Medtronic’s rechargeable spinal cord stimulator, RestoreSensor

1 or 2 epidural leads

Conventional stimulation was a continuous tonic stimulation at 500 Hz with a pulse width of 480 μs. Burst stimulation was "40 Hz burst of four spikes of each 1000 μs at 500 Hz per burst".

The stimulator was switched off

“Achieved stable pain relief with conventional SCS (i.e., paraesthesia inducing stimulation with frequency < 150 Hz) using the Medtronic’s rechargeable spinal cord stimulator RestoreSensor® and with either 1 or 2 epidural leads was inclusion criteria for trial participation”.

3 treatment arms, each of 2 weeks' duration

Hara 2022

Burst

Precision Novi, Boston Scientific, Inc nonrechargeable implantable pulse generator

"A 16‐contact lead (Infinion CX, Boston Scientific, Inc) was implanted for unilateral leg pain or two 8‐contact leads (Linear ST, Boston Scientific, Inc) were implanted for bilateral leg pain"

"Closely spaced, high‐frequency stimuli delivered to the spinal cord. The simulus consisted of 40 Hz of constant current with 4 spikes per burst at an amplitude corresponding to 50% to 70% of paraesthesia perception threshold."

No stimulation provided

"Epidural surgical lead insertion was performed while patients were in the prone position using local anesthetics and mild intravenous sedation to enable patient feedback and cooperation. The aim was to optimize lead placement over the dorsal columns of the spinal cord so that paresthesia occurred in the targeted spinal dermatome (ie, tonic conventional stimulation). A 16‐contact lead (Infinion CX, Boston Scientific, Inc) was implanted for unilateral leg pain or two 8‐contact leads (Linear ST, Boston Scientific, Inc) were implanted for bilateral leg pain through a small skin incision at the L1/L2 or L2/L3 vertebral levels and placed in the epidural space at the T9/T10 [vertebral] level under fluoroscopic guidance. Intraoperative electrophysiological testing and stimulation were performed during longitudinal lead navigation. The leads were anchored at the optimal localization and their positions were confirmed with x‐ray imaging. Leads were then connected to an external neurostimulator using extension cords. Programming software (Illumina 3D, Boston Scientific, Inc) was used to optimize tonic conventional stimulation and determine paresthesia thresholds during the testing period. If there was insufficient improvement in leg pain during the testing period, the leads were removed and the patients were excluded. If there was sufficient improvement in leg pain during the testing period, the patients were included in the trial and their external neurostimulator was replaced with a nonrechargeable implantable pulse generator (Precision Novi, Boston Scientific, Inc) placed subcutaneously on the upper buttock or abdomen under local anesthesia. A nonrechargeable pulse generator was chosen to avoid unblinding of patients. Immediately after implantation of the stimulator, eligible patients underwent four 3‐month periods of treatment."

12 months: 4 periods of 3 months of treatment (6 months of SCS and 6 months of placebo)

Kumar 2007

Conventional

Implantable neurostimulation system produced by Medtronic (Synergy system, Medtronic, Inc., Minneapolis, MN)

Not specified

Mean (SD) settings were an amplitude of 3.7 V (2.0), a pulse width of 350 µs (95.5) and a rate of 49 Hz (16.4). Almost half (45%) of the participants required an amplitude of 4 V or more

”Non‐SCS therapy received by both groups was reviewed and actively managed, at the discretion of the study investigator and according to local clinical practice. Non‐SCS therapy included oral medications (i.e. opioid, non‐steroidal anti‐inflammatory drug, antidepressant, anticonvulsant/antiepileptic and other analgesic therapies), nerve blocks, epidural corticosteroids, physical and psychological rehabilitative therapy, and/or chiropractic care. The protocol excluded other invasive therapy, such as spinal surgery or implantation of an intrathecal drug delivery system”.

“All patients assigned to the SCS group underwent a screening trial. Those experiencing at least 80% overlap of their pain with stimulation‐induced paresthesia and at least 50% leg pain relief received an implantable neurostimulation system”

12 months

Kapural 2022

High frequency

IPG SCS system (Senza, Nevro Corp., Redwood City, CA, USA)

Two percutaneous leads with 8 contacts each placed in the epidural space spanning vertebral levels T8 to T11

10 kHz

“All subjects will continue with their CMM, defined as the best standard of care for each individual patient, as determined by the investigator.” Options included, but were not limited to: oral medications (including analgesic medication, nonsteroidal anti‐inflammatory drugs, neuromodulating agents, antidepressants); topical analgesics, compound creams, or counter‐irritants; combined physical and psychological management; physical therapy; back rehabilitation program; spinal manipulation and spinal mobilisation traction; acupuncture/acupressure; cognitive behavioural therapy; nerve blocks; epidural steroid injections; transcutaneous electrical nerve stimulation

Stimulation at a frequency of 10 kHz and pulse width of 30 μs delivered from an external pulse generator. The stimulation target and current amplitude was adjusted until at least 50% self‐reported back pain reduction from baseline achieved, defined as trial success, or until conclusion of the trial phase

6 months

Perruchoud 2013

High frequency

Medtronic (Minneapolis, MN, USA) impulse generator, either rechargeable (RestoreADVANCED, RestoreSensor, or RestoreUltra) or battery‐powered (PrimeADVANCED)

Not reported. No more than three active contacts

5000 Hz; with pulse width adjusted to 60 ms

The stimulator was switched off

"Currently implanted with suitable SCS device" was an inclusion criteria for study participants

2‐week periods of stimulation; 8 weeks study duration; i.e. 2 weeks current stimulation, 2 weeks high‐frequency (HF) or sham, 2 weeks current stimulation, 2 weeks HF or sham

Rigoard 2019

Conventional

Medtronic neurostimulator (model 97714, n = 49; 37702, n = 39;97702, n = 27; 37714, n = 12; 97712, n = 4; 37713, n = 3; 97713, n = 3; 37712, n = 2; and 37701, n = 1)

Multicolumn surgical lead (Specify 5‐6‐5; Medtronic)

20 Hz to 1200 Hz

"All patients received optimal medical management [OMM]. As part of the confirmation of eligibility (prior to randomization), the investigator and subject will determine an individual OMM treatment plan, which should include non‐investigational pharmacologic agents (for example, tricyclic antidepressants, opioid analgesics or tramadol, antiepileptics, or lidocaine) and/or interventional therapies (for example, therapeutic injections, radiofrequency, acupuncture, functional restoration, physical therapy, and psychological interventions, such as cognitive behavioural therapy) as appropriate. The following treatments are excluded from OMM: intrathecal drug delivery, peripheral nerve stimulation (not an approved indication in the United States), back surgery at the location related to the patient’s original back pain complaint, and experimental therapies."

“The screening test may be conducted with the Specify® 5‐6‐5 surgical lead or with a percutaneous lead(s). If successful, a SCS system will be implanted. A screening test will be determined to be successful if the subject finds the feeling of paresthesia acceptable and has adequate low back pain relief with usual activity and appropriate analgesia as assessed by the physician. Physicians can consider a conducting second screening test with the Specify® 5‐6‐5 lead if a screening test with a percutaneous lead led to inadequate paresthesia coverage of low back pain and/or painful extraneous stimulation (for example, chest wall pain, pressure or sharp mid‐back pain)”.

6 months (then allowed to cross to alternative trial arm and followed to 24 months)

Schu 2014

Conventional and burst

St. Jude Medical SCS system

SCS leads located at the mid‐thoracic position (T7–T10 vertebral level)

For conventional stimulation, 500 Hz mean pulse width ± SD under 500‐Hz tonic stimulation was 370.8 ± 135.4 μs, and mean amplitude ± SD was 5.5 ± 3.6 mA. For burst spinal cord stimulation, packets of five pulses (pulse width 1 ms) at 500 Hz, delivered 40 times per second

Device was switched off

Implanted "with a St. Jude Medical SCS system at least three months previously" was an inclusion criteria for study participants.

3 treatment arms, all of 1 week duration

Sokal 2020

High‐frequency, burst, and conventional

Non‐rechargeable IPG (Precision NoviTM) and in one case, a rechargeable IPG (MontageTM) produced by Boston Scientific Co.

Either one or two linear lead 8‐ or 16‐contact (Infinion 16TM) electrodes on vertebral levels T7–T10

High‐frequency stimulation was programmed with frequency of 1 kHz, pulse width of 120 s, and amplitude = 3 Amp. Burst stimulation delivered intermittent packets using the neural targeting algorithm, which consisted of several pulses per packet with pulse width 250–500 s repeated with frequency of 40 Hz. Conventional stimulation included tonic stimulation with frequencies typically between 40 Hz and 60 Hz. The pulse width ranged between 250 s and 500 s, and the amplitude produced comfortable paraesthesia

IPG was deactivated

Participants underwent 2 weeks of trial stimulation. During the trial period, tonic low‐frequency stimulation was used to check the coverage of pain area with paraesthesia induced by an external stimulator by adjusting the optimal settings of active electrode’s contacts. After a successful 14‐day trial period, participants who achieved at least a 50% reduction in pain were qualified to the second stage of the study, which involved the placement of a permanent internal pulse generator implantation under general anaesthesia.

4 treatment arms, all of 2‐week duration

Sweet 2016

Sub‐threshold high density (HD)

Medtronic RestoreSensor implanted pulse generator (Minneapolis, MN, USA).

“[Two] epidural 8‐contact Medtronic Compact percutaneous SureScan leads (electrode contacts 3 mm long and 1.3 mm diameter, 4 mm intercontact spacing) implanted in the midline with the end of the lead at the T7‐T8 [vertebral] interspace”

Subthreshold HD stimulation (1200 Hz, 200 μs, amplitude 90% of threshold for sensory percept)

"[Same] settings but amplitude 0 V"

“[One‐week] trial of subthreshold HD stimulation, defined as 1200 Hz frequency, 200 μs pulse width, and an amplitude 90% of the threshold for detection of a sensory percept. At the end of the week, each potential participant was asked about pain relief using the subthreshold parameters. Subjects were enrolled only if they reported significant pain relief using subthreshold HD stimulation, defined as 50% reduction in pain on the visual analog scale (VAS) compared with preoperative values.”

2 treatment arms, each of 2 weeks' duration and both preceded by 2 weeks of conventional stimulation

Wolter 2012

Conventional

“With one exception (patient 6), all patients had a non‐rechargeable implantable pulse generator (IPG). In patient 6, the battery state of the IPG was checked to rule out inadvertent discharge during the trial”. Stimulators were produced by Medtronic (n = 5) or Boston Scientific (n = 1).

All participants were implanted with percutaneous‐type electrodes

25 Hz to 100 Hz

The device was switched to zero

Prior SCS for at least 3 months with significant (> 50%) pain relief was an inclusion criterion for trial participation

2 treatment arms, each of 1‐week duration

CMM: conventional medical management; IPG: internal pulse generator or implantable pulse generator; SCS: spinal cord stimulation; µs: microseconds; ms: milliseconds

Figures and Tables -
Table 3. Characteristics of SCS interventions in included studies
Comparison 1. Spinal cord stimulation (SCS) versus placebo

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1.1 Low back pain intensity (0‐100) at immediate‐term follow‐up (< 1 month) Show forest plot

8

Mean Difference (IV, Random, 95% CI)

‐13.79 [‐20.62, ‐6.96]

1.1.1 High‐frequency SCS

4

Mean Difference (IV, Random, 95% CI)

‐11.44 [‐23.72, 0.84]

1.1.2 Conventional SCS

6

Mean Difference (IV, Random, 95% CI)

‐16.57 [‐23.63, ‐9.52]

1.1.3 Burst SCS

4

Mean Difference (IV, Random, 95% CI)

‐13.53 [‐32.61, 5.56]

1.2 Low back pain intensity (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo) Show forest plot

1

Mean Difference (IV, Fixed, 95% CI)

Subtotals only

1.2.1 Burst SCS

1

Mean Difference (IV, Fixed, 95% CI)

‐4.00 [‐8.19, 0.19]

1.3 Leg pain intensity (0‐100) at immediate‐term follow‐up (< 1 month) Show forest plot

2

Mean Difference (IV, Fixed, 95% CI)

‐10.03 [‐20.33, 0.27]

1.3.1 High‐frequency SCS

1

Mean Difference (IV, Fixed, 95% CI)

‐3.83 [‐15.61, 7.95]

1.3.2 Conventional SCS

1

Mean Difference (IV, Fixed, 95% CI)

‐30.10 [‐60.09, ‐0.11]

1.3.3 Burst SCS

1

Mean Difference (IV, Fixed, 95% CI)

‐30.10 [‐60.09, ‐0.11]

1.4 Leg pain intensity (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo) Show forest plot

1

Mean Difference (IV, Fixed, 95% CI)

Subtotals only

1.4.1 Burst SCS

1

Mean Difference (IV, Fixed, 95% CI)

‐2.00 [‐6.47, 2.47]

1.5 Function (0‐100) at immediate‐term follow‐up (< 1 month) Show forest plot

1

Mean Difference (IV, Random, 95% CI)

‐15.10 [‐25.69, ‐4.52]

1.5.1 Conventional SCS

1

Mean Difference (IV, Random, 95% CI)

‐9.80 [‐24.09, 4.49]

1.5.2 Burst SCS

1

Mean Difference (IV, Random, 95% CI)

‐20.60 [‐35.16, ‐6.04]

1.6 Function (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo) Show forest plot

1

Mean Difference (IV, Fixed, 95% CI)

Subtotals only

1.6.1 Burst SCS

1

Mean Difference (IV, Fixed, 95% CI)

‐1.30 [‐3.91, 1.31]

1.7 Health‐related quality of life (0‐1 index) at immediate‐term follow‐up (< 1 month) Show forest plot

1

Mean Difference (IV, Fixed, 95% CI)

Subtotals only

1.7.1 High‐frequency SCS

1

Mean Difference (IV, Fixed, 95% CI)

0.02 [‐0.10, 0.13]

1.8 Health‐related quality of life (0‐1 index) at medium‐term follow‐up (≥ 3 mo to < 12 mo) Show forest plot

1

Mean Difference (IV, Fixed, 95% CI)

Subtotals only

1.8.1 Burst SCS

1

Mean Difference (IV, Fixed, 95% CI)

0.04 [‐0.08, 0.16]

Figures and Tables -
Comparison 1. Spinal cord stimulation (SCS) versus placebo
Comparison 2. Spinal cord stimulation (SCS) plus medical management versus medical management alone

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

2.1 Low back pain intensity (0‐100) at short‐term follow‐up (≥ 1 mo to < 3 mo) Show forest plot

1

Mean Difference (IV, Random, 95% CI)

Subtotals only

2.1.1 Conventional SCS

1

98

Mean Difference (IV, Random, 95% CI)

‐8.70 [‐18.95, 1.55]

2.2 Low back pain intensity (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo) Show forest plot

3

430

Mean Difference (IV, Random, 95% CI)

‐25.97 [‐56.17, 4.23]

2.2.1 High‐frequency SCS

1

140

Mean Difference (IV, Random, 95% CI)

‐54.60 [‐61.03, ‐48.17]

2.2.2 Conventional SCS

2

290

Mean Difference (IV, Random, 95% CI)

‐11.78 [‐16.74, ‐6.81]

2.3 Leg pain intensity (0‐100) at short‐term follow‐up (≥ 1 mo to < 3 mo) Show forest plot

1

Mean Difference (IV, Fixed, 95% CI)

Subtotals only

2.3.1 Conventional SCS

1

98

Mean Difference (IV, Fixed, 95% CI)

‐32.30 [‐42.26, ‐22.34]

2.4 Leg pain intensity (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo) Show forest plot

2

Mean Difference (IV, Random, 95% CI)

Subtotals only

2.4.1 Conventional SCS

2

290

Mean Difference (IV, Random, 95% CI)

‐18.84 [‐33.21, ‐4.47]

2.5 Function (0‐100) at short‐term follow‐up (≥ 1 mo to < 3 mo) Show forest plot

1

Mean Difference (IV, Random, 95% CI)

Subtotals only

2.5.1 Conventional SCS

1

94

Mean Difference (IV, Random, 95% CI)

‐12.60 [‐20.05, ‐5.15]

2.6 Function (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo) Show forest plot

3

430

Mean Difference (IV, Fixed, 95% CI)

‐16.19 [‐19.36, ‐13.01]

2.6.1 High‐frequency SCS

1

140

Mean Difference (IV, Fixed, 95% CI)

‐28.80 [‐33.81, ‐23.79]

2.6.2 Conventional SCS

2

290

Mean Difference (IV, Fixed, 95% CI)

‐7.72 [‐11.82, ‐3.62]

2.7 Health‐related quality of life (0‐100) at medium‐term follow‐up (≥ 3 mo to < 12 mo) Show forest plot

2

Mean Difference (IV, Random, 95% CI)

Subtotals only

2.7.1 Conventional SCS

2

289

Mean Difference (IV, Random, 95% CI)

7.63 [‐0.61, 15.87]

2.8 Global assessment of efficacy at medium‐term follow‐up (≥ 3 mo to < 12 mo) Show forest plot

3

430

Risk Ratio (IV, Random, 95% CI)

7.40 [2.34, 23.39]

2.8.1 High‐frequency SCS

1

140

Risk Ratio (IV, Random, 95% CI)

30.00 [7.60, 118.38]

2.8.2 Conventional SCS

2

290

Risk Ratio (IV, Random, 95% CI)

4.23 [2.12, 8.42]

2.9 Global assessment of efficacy at long‐term follow‐up (≥ 12 mo) Show forest plot

1

100

Risk Ratio (M‐H, Random, 95% CI)

1.96 [0.93, 4.12]

2.10 Withdrawals due to adverse events at longest follow‐up Show forest plot

1

159

Risk Difference (M‐H, Random, 95% CI)

0.02 [‐0.02, 0.06]

2.10.1 High frequency SCS

1

159

Risk Difference (M‐H, Random, 95% CI)

0.02 [‐0.02, 0.06]

2.11 Proportion with any adverse event at longest follow‐up Show forest plot

2

336

Risk Ratio (M‐H, Random, 95% CI)

2.32 [0.39, 13.79]

2.11.1 High‐frequency SCS

1

140

Risk Ratio (M‐H, Random, 95% CI)

5.77 [2.34, 14.20]

2.11.2 Conventional SCS

1

196

Risk Ratio (M‐H, Random, 95% CI)

1.03 [0.74, 1.42]

2.12 Proportion with serious adverse event at longest follow‐up Show forest plot

1

140

Risk Ratio (M‐H, Random, 95% CI)

1.73 [0.51, 5.87]

2.12.1 High‐frequency SCS

1

140

Risk Ratio (M‐H, Random, 95% CI)

1.73 [0.51, 5.87]

2.13 Medication use 1 (number (%) taking opioid medicines) at medium‐term follow‐up (≥ 3 mo to < 12 mo) Show forest plot

2

Risk Ratio (IV, Random, 95% CI)

Subtotals only

2.13.1 Conventional SCS

2

290

Risk Ratio (IV, Random, 95% CI)

0.85 [0.73, 1.00]

2.14 Medication use 2 (daily MME) at medium‐term follow‐up (≥ 3 mo to < 12 mo) Show forest plot

3

430

Mean Difference (IV, Random, 95% CI)

‐9.36 [‐19.89, 1.16]

2.14.1 High‐frequency SCS

1

140

Mean Difference (IV, Random, 95% CI)

‐9.20 [‐20.49, 2.09]

2.14.2 Conventional SCS

2

290

Mean Difference (IV, Random, 95% CI)

‐10.46 [‐39.55, 18.64]

2.15 Work status 1 (number returned to work) at medium‐term follow‐up (≥ 3 mo to < 12 mo) Show forest plot

1

Risk Ratio (IV, Random, 95% CI)

Subtotals only

2.15.1 Conventional SCS

1

100

Risk Ratio (IV, Random, 95% CI)

3.69 [0.43, 31.89]

Figures and Tables -
Comparison 2. Spinal cord stimulation (SCS) plus medical management versus medical management alone