For Physicians
What the evidence supports, and where it thins out
The neuromodulation literature is large, industry-funded, and methodologically uneven. This is an appraisal of what can be reasonably claimed from it — and of the questions it has not answered.
What You'll Learn
Approx. 12 min min read- Trace the evidence base from gate control theory to the current randomised trials
- Identify the recurring methodological limitations across the pivotal studies
- Distinguish claims supported by randomised data from claims supported by registries
- Articulate what the field still does not know
The shape of the evidence base
Deep in some places, notably shallow in others.
Spinal cord stimulation has been in clinical use since 1967, and the modern randomised evidence base begins with PROCESS in 2007. Since then a series of industry-sponsored pivotal trials — SENZA-RCT, ACCURATE, SUNBURST, EVOKE, SENZA-PDN, SENZA-NSRBP and ReActiv8-B — have tested newer approaches against an older comparator, medical management or sham. Their results are not interchangeable: some established superiority, while ReActiv8-B did not meet its prespecified primary superiority endpoint.
The structural feature worth noticing is that most of these trials compare a new therapy with conventional tonic stimulation or conventional medical management. Almost none compare one modern therapy directly with another. The result is a literature that supports several platforms within their studied populations but provides little evidence for ranking today's alternatives against each other.
Recurring Limitations
Four issues that apply across nearly the whole literature
Blinding is close to impossible
Paresthesia-based stimulation is perceptible by design, which makes participant blinding unachievable in any trial with a tonic comparator. Sub-perception waveforms permit better blinding, and the sham-controlled studies that have used it report more modest effects.
Industry sponsorship is near-universal
The pivotal trials are designed, funded and often analysed by the manufacturer of the device under test. This is standard for device trials and does not invalidate them, but it shapes comparator choice, endpoint selection and follow-up duration.
Attrition rises steeply with time
Twelve-month data are generally robust. Twenty-four and thirty-six month data carry substantial loss to follow-up, and analyses of remaining participants systematically favour responders.
Responder thresholds are conventions
The 50% relief threshold is inherited convention rather than a validated clinically important difference. Trials reporting responder rates at 50% and at 80% show how much the headline number depends on where the line is drawn.
The Pivotal Trials
What each one actually established
Stated at the level of what the design can support, not the level of the press release.
| Trial | Design | Established | Principal limitation |
|---|---|---|---|
| PROCESS (2007) | RCT, SCS vs conventional medical management, n=100 | Superiority of SCS over CMM for leg pain in persistent spinal pain syndrome at 6 months | Unblinded; high crossover from CMM arm dilutes later comparisons |
| SENZA-RCT (2015) | RCT, 10 kHz vs tonic, n=198 | Superiority of 10 kHz over tonic for both back and leg pain at 3, 12 and 24 months | Unblinded; tonic comparator programmed to a fixed protocol |
| ACCURATE (2017) | RCT, DRG vs tonic SCS, n=152 | Superiority of DRG stimulation for CRPS and causalgia of the lower limb at 3 and 12 months | Unblinded; DRG arm had a higher rate of lead-related adverse events |
| SUNBURST (2017) | Randomised crossover, burst vs tonic, n=100 | Non-inferiority and modest superiority of burst over tonic, with patient preference for burst | Crossover design with possible carryover; short exposure per arm |
| EVOKE (2021–2024) | Double-blind RCT, closed-loop vs open-loop, n=134 | Superiority of ECAP-controlled closed-loop over open-loop, sustained to 36 months | Single manufacturer; the strongest blinding in the field but a single comparator |
| SENZA-PDN (2021) | RCT, 10 kHz + CMM vs CMM, n=216 | Substantial benefit in painful diabetic neuropathy, with reported neurological improvement | Unblinded; CMM arm crossover permitted at 6 months |
| SENZA-NSRBP (2022) | RCT, 10 kHz + CMM vs CMM, n=159 | Benefit in non-surgical refractory back pain, an indication previously outside the therapy | Unblinded; short randomised phase before crossover |
| ReActiv8-B (2020–2024) | Sham-controlled RCT, restorative neurostimulation, n=204 | Primary superiority endpoint not met; supportive secondary outcomes and improvement among five-year completers | Sham crossed over at 120 days; five-year data available for 126/204 and uncontrolled after crossover |
Modern platforms are usually tested against tonic stimulation, medical management or sham — not against one another. That leaves device choice dependent on indication, trade-offs and clinical judgment rather than a direct ranking trial.
Open Questions
What the literature does not answer
Comparative effectiveness
Which waveform for which patient?
There is no adequately powered head-to-head trial between 10 kHz, burst, closed-loop and multiphase stimulation. Selection between them is currently reasoning from mechanism and from single-arm comparisons, not from evidence.
Durability
What happens at ten years?
Explant rates in registry data are non-trivial and rise with time. The reasons — loss of efficacy, hardware failure, infection, resolution of the original problem — are poorly disaggregated in the published literature.
Mechanism
Why does sub-perception stimulation work?
Gate control theory explains paresthesia-based stimulation reasonably well. It does not explain 10 kHz or burst. Glial modulation, medial pathway engagement and dorsal horn interneuron effects are all proposed and none is settled.
Selection
Can response be predicted before a trial?
No validated predictive model exists. Quantitative sensory testing, functional imaging and phenotyping have all been explored without producing a tool that reliably outperforms clinical judgement plus a trial.
Trial validity
Does the trial period predict long-term outcome?
The trial reliably identifies non-responders. Whether it predicts durable multi-year benefit among responders is considerably less clear, and some registry data suggest the correlation weakens substantially over time.
Sham
How large is the placebo component?
For an implanted device with a perceptible procedure, this is inherently difficult to quantify. The sham-controlled data that exist suggest a real specific effect, but a smaller one than unblinded trials report.
Implications for clinical decision-making
Three positions follow from the appraisal above. First, that selection deserves more attention than platform choice, because the effect of getting the phenotype and target right is larger than any documented difference between modern devices. Second, that patients should be told the evidence supports meaningful relief in a majority of well-selected candidates, not that it supports a specific percentage in their specific case. Third, that the absence of head-to-head data should be disclosed rather than papered over with whichever manufacturer's comparison happens to be most favourable.
None of this is an argument against the therapy. The randomised evidence for spinal cord stimulation in appropriately selected patients is stronger than that supporting a good deal of routine spine care. It is an argument for describing that evidence accurately.
Reading the primary sources
Study-level summaries with findings and limitations for the trials above are in the Clinical Evidence Library. Read the original publication when a source link is available; source verification and independent clinical review remain necessary before using a summary to guide care.
Discussing a case
Referring clinicians can use the patient-selection guidance to frame their questions, then consult a qualified neuromodulation clinician in their region. This site does not arrange referrals or consultations.