Nonsurgical interventions for lumbar spinal stenosis with neurogenic claudication: A systematic review and network meta-analysis

Abstract

Background

Lumbar spinal stenosis (LSS) with neurogenic claudication is a common cause of chronic pain and disability, but the comparative efficacy of nonsurgical interventions remains uncertain. This network meta-analysis evaluated and compared these interventions.

Methods

PubMed, Web of Science, Embase, and the Cochrane Central Register of Controlled Trials were searched for randomized controlled trials (RCTs) evaluating nonsurgical interventions for LSS with neurogenic claudication. The primary outcome was short-term pain, defined as change in pain intensity from baseline to immediately after treatment. Secondary outcomes included long-term pain (≥6 months after randomization), short- and long-term function, short- and long-term walking distance assessed by walking tests, and short-term response rate. Bayesian random-effects network meta-analyses were performed, and results were reported as mean differences (MDs) with 95% credible intervals (CrIs).

Results

Thirty-five RCTs involving 3147 participants were included. Low-certainty evidence indicated that, compared with placebo, acupuncture was associated with clinically important short-term improvements in pain (MD −10.02, 95% CrI −18.49 to −0.23) and function (MD −14.39, 95% CrI −25.26 to −3.69). Pairwise meta-analysis also suggested that acupuncture was associated with clinically important long-term pain reduction and statistically significant improvement in long-term function versus placebo. No other interventions showed clear benefits over placebo for pain or function. No intervention showed clear improvement in walking distance versus placebo or other interventions in the short or long term. All estimates were associated with substantial uncertainty.

Conclusion

The evidence was characterized by substantial uncertainty, with certainty ranging from low to very low. Although acupuncture showed potential benefit, the findings remain inconclusive and highlight the need for rigorously designed, high-quality RCTs.

Protocol registration

CRD42025649922

1. Introduction

Lumbar spinal stenosis (LSS) is characterized by age-related degenerative changes in the intervertebral discs, facet joints, and ligaments, resulting in narrowing of the spinal canal and compression of neurovascular structures.^1,2 It is a major cause of chronic pain, neurogenic claudication, and disability, substantially impairing quality of life and physical function.1, 2, 3 Clinically diagnosed LSS affects 11%–38% of the general population, and its prevalence and socioeconomic burden are expected to increase with global population aging.⁴ Degenerative osteoarthritic changes can compress the spinal nerves, often leading to ischemia and progressive walking limitation, which in turn contribute to disability and increased healthcare utilization.^1,2,5,6 Neurogenic claudication, the hallmark of symptomatic LSS, typically presents as buttock or lower-extremity pain, weakness, or heaviness that is aggravated by walking or standing and relieved by sitting or forward flexion.^1,7

The management of LSS remains controversial. Although surgery is generally reserved for refractory cases, most patients initially receive nonsurgical interventions, including pharmacotherapy, physical therapy, epidural injections, rehabilitation, and lifestyle modification.⁸ A recent review identified activity modification, analgesia, and physical therapy as first-line treatments.¹ A clinical guideline also recommended nonsurgical management before surgery, although the supporting evidence was inconsistent.⁹ A systematic review of 44 randomized controlled trials (RCTs) found that multimodal care combining manual therapy, exercise, and education had moderate efficacy, whereas epidural steroids showed no benefit.¹⁰ Evidence for other interventions, including calcitonin, gabapentin, and acupuncture, was inconclusive because of low-quality evidence.¹⁰ Likewise, the guideline developed by Bussières et al. in 2021 conditionally recommended multimodal nonpharmacologic care (moderate-certainty evidence) and acupuncture (very low-certainty evidence), but strongly recommended against nonsteroidal anti-inflammatory drugs (NSAIDs), opioids, gabapentin, and epidural steroids because of insufficient or conflicting evidence.¹¹

Despite these findings, the current evidence base is limited by methodological heterogeneity, inadequate statistical power, and a lack of direct comparative data. As a result, high-quality evidence to guide the selection of nonsurgical interventions for LSS with neurogenic claudication remains insufficient. To address this gap, we conducted a systematic review and network meta-analysis of RCTs to compare the effectiveness of nonsurgical interventions in the short term (post-treatment) and long term (≥6 months after randomization). Network meta-analysis integrates direct and indirect evidence across a connected evidence network, allowing simultaneous comparison of multiple interventions even when head-to-head trials are limited.^12,13 This study aimed to synthesize the available evidence to inform clinical decision-making for LSS with neurogenic claudication.

2. Methods

This systematic review and network meta-analysis was conducted in accordance with the Cochrane Handbook for Systematic Reviews of Interventions and reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension statement for Network Meta-Analyses (PRISMA-NMA).^14,15 The protocol was registered in PROSPERO (CRD42025649922).

2.1. Criteria for inclusion and exclusion

2.1.1. Study types

Eligibility criteria were established using the Participants, Intervention, Comparator, Outcome, and Study design (PICOS) framework (Supplement 1). Only RCTs published in peer-reviewed journals in English were included. Crossover trials and studies available only as abstracts were excluded.

2.1.2. Participant types

The target population comprised adults (aged ≥18 years) with imaging-confirmed LSS, including central or foraminal stenosis, with or without spondylolisthesis, accompanied by neurogenic claudication. Neurogenic claudication was defined as buttock or leg pain, aching, numbness, tingling, weakness, or fatigue, with or without back pain, precipitated by standing or walking.^1,7 Studies of radiculopathy caused by disc herniation were excluded. Mixed-population studies were included only when data for patients with LSS and neurogenic claudication were reported separately.

2.1.3. Intervention types and controls

Eligible interventions included all nonsurgical treatments compared with placebo, usual care, no intervention, other nonsurgical interventions, or surgery. Studies comparing nonsurgical treatments with surgery were included solely to preserve the treatment network for the network meta-analysis; evaluating the efficacy of surgery was not an objective of this study. RCTs investigating traditional herbal medications or multimodal nonsurgical interventions without clearly defined treatment details were excluded. To enable network meta-analysis, interventions were categorized according to the original study reports and their core therapeutic principles. For example, supervised exercise-based physical therapy (SE-PT) included interventions centered on prescribed physical activity delivered under supervision, whereas manual therapy comprised hands-on techniques targeting joints and soft tissues. Usual care generally included physical therapy, education or counseling with home exercise instruction, and NSAIDs when tolerated.16, 17, 18 This classification necessarily grouped protocols with varying components and may have introduced clinical heterogeneity.

2.1.4. Outcome measures

Only RCTs with extractable outcome data were included. Eligible studies were required to report at least one of the following outcomes: pain intensity, physical function, walking capacity, or response rate. For outcome timing, we defined short-term outcomes as those assessed immediately after completion of the intervention, and long-term outcomes as those assessed at 6 months or more after randomization. When multiple time points were available within the same time window, the time point closest to the end of treatment was selected for short-term analyses, and the time point closest to 6 months after randomization was selected for long-term analyses. The primary outcome was short-term pain, defined as the change in pain intensity from baseline to immediately after treatment completion. This outcome was selected because of its clinical importance and its consistent reporting across interventions, which strengthened the network analysis. If immediate post-treatment measurements were unavailable, outcomes assessed within one month after treatment completion were prioritized. When separate leg and back pain scores were reported, they were synthesized to derive a pooled pain intensity estimate. If only one component (leg or back pain) was available, that measure was extracted. Pain outcomes assessed using the visual analogue scale (VAS) or numerical rating scale (NRS) were prioritized because these are standard measures in chronic pain research.¹⁹ For comparability, all pain scores were converted to a 0–100 scale, with higher scores indicating worse pain. The minimum clinically important difference (MCID) was defined as 10–14 points.²⁰

Secondary outcomes included long-term pain, short- and long-term function, short- and long-term walking distance assessed by walking tests, and short-term response rate. Pain, function, and walking distance were analyzed based on changes from baseline. For functional assessment, the Roland-Morris Disability Questionnaire (RMDQ) and Oswestry Disability Index (ODI), both widely used measures, were prioritized.²¹ The RMDQ contains 24 items assessing disability, with scores ranging from 0 (no disability) to 24 (severe disability).²² The ODI evaluates the impact of low back pain on daily life, with scores ranging from 0% (no disability) to 100% (maximum disability).²³ For data synthesis, all functional scores were transformed to a 0–100 scale, with higher scores indicating worse function. The MCID for the RMDQ has been reported as 2–3 points on the original scale,²⁴ and the MCID for the ODI in LSS has been reported as 13.8 points.²⁵ Accordingly, 13.8 points was used as the MCID on the converted 0–100 scale. The self-paced walking test (SPWT) is a commonly used walking test for assessing walking performance in patients with LSS, in which walking distance is recorded until symptom-limited rest is required, up to a maximum of 30 min.26, 27 Safety outcomes were also assessed.

2.2. Literature searches

A comprehensive literature search was conducted on December 1, 2024, in PubMed, Web of Science, the Cochrane Central Register of Controlled Trials (CENTRAL), and Embase. Reference lists of relevant reviews and RCTs were also screened. The search strategy (Supplement 2) was developed according to the PICOS framework. Two independent researchers screened titles and abstracts, followed by full-text review of potentially eligible articles. Disagreements were resolved by consensus.

2.3. Data selection and extraction

Change-from-baseline outcome data were extracted whenever available; otherwise, baseline and post-intervention values were used to calculate change scores.¹⁴ Missing standard deviations were imputed from studies using the same outcome scale; when multiple candidate values were available, the largest reported standard deviation was used as a conservative approach.^13,14 Graphical data were extracted using GetData Graph Digitizer, version 2.2. Study design characteristics, baseline demographic data, intervention details, treatment and follow-up durations, and safety outcomes were also extracted. No additional data were requested from study authors.

2.4. Risk of bias assessment of included studies

Two independent authors assessed the risk of bias of each included study using version 2 of the Cochrane Risk of Bias tool for randomized trials (RoB 2).²⁸ RoB 2 evaluates five domains of bias related to trial design, conduct, and reporting. Disagreements were resolved through discussion and consensus.

2.5. Assessment of the certainty of the evidence

The certainty of evidence was assessed using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach according to the minimally contextualized framework recommended by Brignardello-Petersen et al.^29,30 Certainty was dichotomized as high certainty (moderate to high certainty) or low certainty (low to very low certainty). Interventions were classified into three categories: category 0 (least effective), category 1 (moderately effective), and category 2 (most effective).

2.6. Data analysis

The network meta-analysis was conducted using the multinma package (version 0.8.1)³¹ in R (version 4.4.2),³² within a Bayesian framework implemented in Stan.³³ Random-effects models were used to account for between-study heterogeneity arising from variations in treatment protocols, dosages, and durations. Evidence of a difference was inferred when the 95% credible interval (CrI) excluded the null value. For continuous outcomes, effect sizes were expressed as mean differences (MDs) with 95% CrIs. Global inconsistency was explored by comparing the deviance information criterion (DIC) between the consistency and inconsistency models;³⁴ a lower DIC for the inconsistency model by ≥5 points was considered to suggest potential inconsistency.³⁵ Node-splitting analyses were performed to assess local inconsistencies in individual comparisons.

The random-effects network meta-analysis assumed a normal likelihood and identity link and was implemented in Stan using four Markov chains.³¹ Each chain ran for 2000 iterations, including 1000 warmup iterations, yielding 4000 post-warmup draws for inference. Convergence was assessed using standard diagnostics.³⁶ Treatments were ranked at each posterior iteration,³¹ and treatment rankings were summarized using the surface under the cumulative ranking curve (SUCRA) and rankograms.³⁷ SUCRA values range from 0 (certainly the worst treatment) to 1 (certainly the best).³⁷ Sensitivity analyses for short-term pain and function were conducted by excluding trials with imputed standard deviations, small sample sizes (defined as ≤10 participants per study arm), or high risk of bias to evaluate the robustness of the findings. When network meta-analysis was not feasible, conventional pairwise meta-analysis was performed where appropriate. For continuous outcomes, treatment effects were synthesized as mean differences with 95% confidence intervals (CIs).

3. Results

3.1. Study characteristics

The search identified 35,015 records in total (34,986 from databases and 29 from reference lists). After duplicate removal, 27,170 records were screened by title and abstract, and 74 articles underwent full-text review. Of these, 38 were excluded (Supplement 3), leaving 36 reports corresponding to 35 RCTs for inclusion, as one trial was reported in two publications^38,39 (Fig. 1). Trial characteristics, including intervention details, are provided in Supplement 4. The included trials enrolled 3147 participants (47.4% female; mean age, 63.3 years) across 78 comparison groups. Evaluated nonsurgical interventions included, among others, calcitonin, gabapentin, oral glucocorticoids, supervised exercise-based physical therapy (SE-PT), multimodal physical therapy (MPT), manual therapy (e.g., spinal manipulation), acupuncture, epidural glucocorticoid injections (EGI), therapeutic ultrasound, and usual care. Follow-up ranged from immediately after treatment to 10 years. The trials were conducted across Asia, Europe, and North America, primarily in hospitals, clinics, and other healthcare settings. Given the overall comparability of study design, patient populations, and outcome measures, network meta-analysis was considered appropriate.

Fig. 1.

PRISMA flow diagram of study selection. RCTs, randomized controlled trials.

3.2. Risk of bias of included studies

Ten trials were rated as having some concerns,40, 41, 42, 43, 44, 45, 46, 47, 48, 49 eight as high risk of bias,50, 51, 52, 53, 54, 55, 56, 57 and the remainder as low risk of bias (Supplement 5). The main concerns related to deviations from intended interventions, missing outcome data, and selective reporting.

3.3. Primary outcome: short-term pain

Seventeen trials contributed to the short-term pain network (Fig. 2a), with pain most commonly assessed using the VAS or NRS (Supplement 4). The network included 16 nodes, and 15 comparators were presented against placebo in the forest plot (Fig. 2b). Compared with placebo, acupuncture was associated with clinically important short-term pain reduction (MD −10.02, 95% CrI −18.49 to −0.23), although the estimate was imprecise (Fig. 2b). In contrast, 14 of the 15 interventions, including EGI (MD −9.33, 95% CrI −21.68 to 2.90), calcitonin (MD −5.89, 95% CrI −15.03 to 4.28), and SE-PT (MD −0.38, 95% CrI −11.35 to 10.50), were not superior to placebo (Fig. 2b). Acupuncture had the highest SUCRA ranking for short-term pain reduction (SUCRA 0.82), followed by EGI (SUCRA 0.78) (Fig. 2b; Supplement 6). Node-splitting analyses indicated consistency between direct and indirect estimates (Supplement 7). However, because of network sparsity and reliance on indirect evidence, the SUCRA rankings should be regarded as hypothesis-generating and interpreted cautiously. The certainty of evidence for short-term pain reduction was low (low to very low certainty), and all interventions were classified as among the least effective within the minimally contextualized framework (Supplement 8).

Fig. 2.

Network meta-analysis of short-term pain. (a) Network plot; node size is proportional to the number of participants, and edge thickness is proportional to the number of trials. (b) Forest plot of interventions versus placebo, presented as mean difference (MD) with 95% credible interval (CrI). Values <0 favor the intervention and values >0 favor placebo. (c) League table for the main analysis; interventions are shown on the diagonal, and estimates whose 95% CrIs excluded 0 are shown in bold. Read comparisons from left to right; an estimate <0 indicates greater short-term pain reduction for the intervention in the column than for that in the row. EAI, epidural injection of local anesthetics; EGI, epidural injection of glucocorticoids; MPT, multimodal physical therapy; OG, oral glucocorticoids; SE-PT, supervised exercise-based physical therapy; SUCRA, surface under the cumulative ranking curve; SUS, sham ultrasound; UC, usual care.

Across network pairwise estimates, acupuncture, EGI, gabapentin plus usual care, and calcitonin were associated with clinically important pain reduction compared with usual care or no intervention (Fig. 2c). Sham ultrasound plus SE-PT also showed clinically important pain reduction versus no intervention, whereas no clear differences were observed for the other comparisons (Fig. 2c).

Sensitivity analyses excluding studies with small sample sizes,^58,59 imputed standard deviations,^46,55 or high risk of bias50, 51, 52^,54,55 yielded similar results. No treatment was superior to placebo for pain reduction (Supplement 9). EGI (MD −17.86, 95% CrI −34.79 to −0.84) and acupuncture (MD −16.77, 95% CrI −28.75 to −4.42) remained superior to usual care (Supplement 10).

Four additional trials^38,53,60,61 were analyzed separately. Epidural injection of local anesthetics (EAI) plus a tumor necrosis factor-alpha (TNF-α) inhibitor (etanercept) was associated with clinically important pain reduction compared with EGI plus EAI or EAI alone (Supplement 11).^38,53 SE-PT plus home exercise was superior to home exercise alone,⁶¹ whereas calcitonin plus MPT plus home exercise showed no difference compared with acetaminophen plus MPT plus home exercise (Supplement 12).⁶⁰

3.4. Secondary outcomes

3.4.1. Short-term function

The short-term function network comprised 19 nodes, with 18 comparators presented against placebo in the forest plot (Fig. 3a,b). Compared with placebo, acupuncture was associated with clinically important functional improvement (MD −14.39, 95% CrI −25.26 to −3.69), whereas home exercise was associated with poorer short-term function (MD 17.93, 95% CrI 2.72 to 32.18) (Fig. 3b). Across network pairwise estimates, acupuncture was also superior to acetaminophen, home exercise, and usual care (Fig. 3c). Acupuncture had the highest SUCRA ranking (SUCRA 0.94), followed by EGI plus EAI plus usual care (SUCRA 0.72) (Supplement 13). Sensitivity analyses confirmed the superiority of acupuncture over placebo (MD −14.85, 95% CrI −25.49 to −3.82) (Supplements 14 and 15), and acupuncture remained the highest-ranked treatment (SUCRA 0.96) (Supplement 14). However, because these sparse networks relied largely on indirect comparisons, the SUCRA rankings should be considered exploratory and interpreted cautiously. Three additional trials suggested that EAI plus epidural injection of TNF-α inhibitor was associated with greater functional improvement than EGI plus EAI or EAI alone,^38,53 whereas calcitonin plus MPT plus home exercise showed no difference compared with acetaminophen plus MPT plus home exercise (Supplements 16 and 17).⁶⁰

Fig. 3.

Network meta-analysis of short-term function. (a) Network plot; node size is proportional to the number of participants, and edge thickness is proportional to the number of trials. (b) Forest plot of interventions versus placebo, presented as mean difference (MD) with 95% credible interval (CrI). Values <0 favor the intervention and values >0 favor placebo. (c) League table for the main analysis; interventions are shown on the diagonal, and estimates whose 95% CrIs excluded 0 are shown in bold. Read comparisons from left to right; an estimate <0 indicates greater short-term functional improvement for the intervention in the column than for that in the row. Com-pharma, combination pharmacotherapy; EAI, epidural injection of local anesthetics; EGI, epidural injection of glucocorticoids; MPT, multimodal physical therapy; OG, oral glucocorticoids; SE-PT, supervised exercise-based physical therapy; SUCRA, surface under the cumulative ranking curve; SUS, sham ultrasound; UC, usual care.

3.4.2. Short-term walking distance

A total of 21 RCTs assessed walking capacity. Of these, 10 trials evaluating 13 interventions reported walking distance with extractable data and were included in the quantitative analysis.^42,44, 45, 46^,48,49,60, 61, 62, 63 Network meta-analysis of six trials showed no clear differences versus placebo (Supplement 18). The remaining four trials also showed no significant differences between interventions and controls (Supplement 19).

3.4.3. Short-term response rate

Among the 11 trials that assessed short-term response rate, substantial heterogeneity in response criteria precluded quantitative synthesis.^38,40,41,43, 44, 45^,48,61,63, 64, 65, 66 The response criteria used in each study and the corresponding responder counts are presented in Supplement 20.

3.4.4. Long-term pain, function, and walking distance

Ten trials reported long-term pain outcomes.^{17,38,53,55,58,62,}65, 66, 67, 68 Network meta-analysis of four trials showed no clear differences between interventions (Supplement 21). EAI plus epidural injection of TNF-α inhibitor was superior to EGI plus EAI or EAI alone (Supplement 22).^38,53 Pairwise meta-analysis suggested that acupuncture may provide clinically important long-term pain reduction versus placebo (sham acupuncture) (MD −13.74, 95% CI −18.76 to −8.71) (Supplement 23).^65,66 In addition, EGI was associated with greater long-term pain reduction than EGI plus MPT,⁶⁸ whereas no significant difference in long-term pain was observed between EGI and manual therapy (Supplement 24).⁵⁵

For long-term function, pairwise meta-analysis showed a statistically significant benefit of acupuncture versus placebo (sham acupuncture) (MD −7.50, 95% CI −11.94 to −3.06),^65,66 and network meta-analysis suggested benefit for EAI plus epidural injection of TNF-α inhibitor (Supplements 25–27).^38,53 Only two trials reported long-term walking distance, and neither showed a significant difference: combination pharmacotherapy versus SE-PT⁶³ and surgery versus usual care⁶² (Supplement 28).

Evidence for the secondary outcomes was also of low certainty (low to very low), and all interventions were classified as among the least effective within the minimally contextualized framework because of sparse data, potential heterogeneity, and imprecision.

3.5. Other comparisons

Matsudaira et al. reported that limaprost significantly improved pain, function, and subjective satisfaction compared with etodolac at 8 weeks.⁵⁰ Kim et al. evaluated an 8-week course of limaprost and pregabalin, administered as monotherapy or in combination, and found no significant advantage of combination therapy over either monotherapy at the end of treatment.⁵⁷ Song et al. evaluated translaminar epidural injections combined with selective nerve root spinal injections using glucocorticoids with or without local anesthetics.⁴⁷ No significant between-group differences in pain or function were observed at 1 or 3 months after randomization.⁴⁷ Eguchi et al. reported that combined Neurotropin and limaprost improved walking speed compared with either drug alone. Neurotropin may also improve low back pain, stride length, and standing balance.⁵⁶ Young et al. found that adding electrical dry needling and spinal manipulation to exercise, manual therapy, and electrothermal modalities resulted in greater improvement in low back, buttock, and leg pain, as well as related disability, at 3 months than exercise, manual therapy, and electrothermal modalities alone.⁶⁹ However, no significant between-group differences were observed at 2 or 6 weeks.⁶⁹

3.6. Adverse events

Adverse events were assessed qualitatively. Common intervention-related adverse events were reported across studies (Supplement 29). For example, gastrointestinal complaints were frequently reported with calcitonin. Additionally, acupuncture-related adverse events were generally mild and transient, such as subcutaneous hematoma and dizziness.

4. Discussion

4.1. Summary of key findings

This systematic review and network meta-analysis showed substantial uncertainty across all effect estimates, and the findings should therefore be interpreted cautiously. Low-certainty evidence suggested that, compared with placebo, acupuncture was associated with clinically important short-term improvements in pain and function. Pairwise meta-analysis also suggested clinically important long-term pain reduction and statistically significant improvement in long-term function versus placebo (sham acupuncture). No other nonsurgical intervention showed clear benefit over placebo for pain or function. No nonsurgical intervention showed clear improvement in walking distance in either the short or long term compared with placebo or other interventions.

4.2. Comparison with previous studies

Our findings suggest that the current evidence remains insufficient to draw definitive conclusions regarding the effectiveness of nonsurgical interventions for LSS with neurogenic claudication, consistent with previous reviews.^10,70,71 Two previous systematic reviews included Chinese-language as well as English-language publications.^72,73 Sun et al. systematically evaluated nonpharmacological Chinese therapies for degenerative LSS and reported that low-quality evidence suggested potential benefit of acupuncture, whereas findings for other traditional interventions were inconsistent and also limited by low-quality evidence.⁷² Similarly, a network meta-analysis of RCTs published between January 2000 and July 2021 suggested that Chinese herbal medicine and limaprost might provide short-term improvement in pain and function.⁷³ However, detailed methodological appraisal identified limitations in that review⁷³ that may affect the reliability of its conclusions.

4.3. Clinical and scientific implications

Our findings suggest that acupuncture may be beneficial for LSS with neurogenic claudication, although further confirmation is needed. Previous evidence supports multimodal care, including education, home exercise, and manual therapy, for managing LSS with neurogenic claudication.¹¹ Acupuncture may therefore serve as a complementary option within such multimodal strategies to improve symptom control. Mechanistically, acupuncture may alleviate LSS symptoms by increasing sciatic nerve blood flow, potentially influencing cauda equina and nerve root circulation, and activating endogenous pain-inhibitory pathways; some experimental evidence also suggests involvement of cholinergic nerves.74, 75, 76 At the same time, contextual and expectancy effects related to the treatment environment, patient-clinician rapport, and patient expectations may also contribute to the observed benefits,77, 78, 79 although sham acupuncture and standardized intervention protocols may improve blinding and reduce bias in effect estimates. Nevertheless, because two of the three acupuncture trials included in this review were conducted in Chinese populations, the generalizability of these findings may be limited.^65,66 In addition, only 164 patients received acupuncture across these trials, excluding sham acupuncture groups, underscoring the need for larger RCTs.

Among other nonsurgical interventions, calcitonin showed potential for pain relief but was associated with adverse effects such as nausea, vomiting, and headache in our qualitative analysis. Concerns have also been raised regarding possible long-term cancer risk, warranting caution in its use.^80,81 Given these concerns, together with prior guideline recommendations against calcitonin,¹¹ its safety and efficacy require further evaluation to inform risk-benefit assessment. Current guidelines also recommend against epidural steroid injections.¹¹ The limited benefit observed with EGI may reflect the predominantly neuroischemic rather than inflammatory pathophysiology of LSS with neurogenic claudication.⁶

Gabapentin plus usual care showed potential for pain relief compared with usual care alone or no intervention; however, this finding was based on a single small trial involving 55 patients and is therefore highly uncertain. Emerging evidence suggests that gabapentinoids such as gabapentin and pregabalin may be ineffective for low back pain and are associated with adverse events, further questioning their utility in this population.^11,82, 83, 84 Limaprost also showed short-term improvement in pain and function,⁵⁰ consistent with a previous meta-analysis,⁷³ but the evidence was of very low certainty because of small sample sizes. Epidural injection of TNF-α inhibitors, such as etanercept, has also been explored for LSS.^85,86 Although this intervention was associated with short- and long-term benefits for both pain and function in the present review, the evidence was derived from a single small, low-quality trial (n = 90), resulting in very low certainty. Overall, the evidence remains sparse, and previous studies have suggested limited clinical benefit of epidural TNF-α inhibitor injections for sciatica caused by herniated discs or spinal stenosis.⁸⁷

The findings for SE-PT should also be interpreted cautiously. Outcomes associated with SE-PT often depend on patient adherence, which can only be partially ensured in clinical trials. Despite its frequent use in clinical practice, further research is needed to establish standardized protocols and clarify its clinical value.

Despite the clinical importance of walking capacity in LSS, fewer than two-thirds of trials reported this outcome. In addition, the conduct and reporting of walking tests were often inconsistent or insufficiently detailed, limiting interpretation of the available evidence. No nonsurgical intervention showed superiority over placebo or other interventions in improving walking distance in either the short or long term. One possible explanation is that walking tests such as the SPWT require standardized space, prolonged assessment time, and trained personnel, which may limit their feasibility in RCTs. This may explain why walking tests were not adopted in some studies, thereby reducing the available data and statistical power to detect meaningful differences. The consistent lack of improvement in walking capacity across interventions may also reflect the limited ability of current nonsurgical treatments to modify the underlying structural pathology of LSS. Given this evidence gap and the potential signal of benefit associated with acupuncture, future adequately powered sham-controlled trials should prioritize walking outcomes, such as distance and time on the SPWT, to evaluate whether acupuncture improves walking capacity in patients with LSS and neurogenic claudication.

4.4. Strengths and limitations

This review has several strengths. It evaluated a broad range of nonsurgical interventions and applied rigorous methods aligned with Cochrane recommendations. Risk of bias and certainty of evidence were assessed using RoB 2 and the GRADE approach. Although this review cannot provide definitive conclusions regarding the comparative effectiveness of nonsurgical interventions for LSS with neurogenic claudication, it offers important insight into the current evidence base and highlights key knowledge gaps that warrant high-quality studies with adequate power, low risk of bias, and long-term follow-up.

Several limitations should also be acknowledged. First, classification of nonsurgical interventions was challenging because some modalities overlapped conceptually and pragmatically. Interventions were categorized according to their trial-reported components; however, clinical heterogeneity within categories was unavoidable and may have introduced confounding. Second, some broad intervention categories, such as SE-PT and usual care, necessarily encompassed heterogeneous protocols, which likely increased statistical heterogeneity and widened the credible intervals. Accordingly, the SUCRA rankings reflect average effects across diverse practices and should be interpreted with great caution; they cannot guide selection of specific treatment protocols. This heterogeneity precluded firm conclusions regarding individual protocols and rendered the findings primarily hypothesis-generating at the level of broader intervention classes. Third, substantial uncertainty surrounded all estimates. Challenges related to blinding, randomization, and the selection of appropriate controls are common in trials of nonsurgical interventions for LSS, particularly for nonpharmacological therapies, and may increase risk of bias and reduce validity. In addition, underlying clinical heterogeneity across studies may have obscured true treatment differences. Fourth, although the literature search was systematic, relatively few interventions were supported by multiple trials or direct head-to-head comparisons. As a result, the network was sparse, and many estimates relied heavily on indirect evidence; thus, the SUCRA rankings should be regarded as exploratory. Fifth, quality-of-life outcomes were reported inconsistently and therefore could not be synthesized.

4.5. Conclusions and suggestions for future research

This study provides a comprehensive but inconclusive synthesis of the current evidence because certainty ranged from low to very low. Although effect estimates and SUCRA rankings suggested that some interventions, particularly acupuncture, may be promising, these findings remain highly uncertain. The quality and heterogeneity of the included studies, together with the limited number of trials, precluded definitive conclusions regarding comparative effectiveness. Further high-quality RCTs with adequate sample sizes and long-term follow-up are needed to provide more reliable evidence for clinical decision-making.

Funding

This study was supported by the High Level Chinese Medical Hospital Promotion Project (HLCMHPP2023089).

Ethics statement

Not applicable.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

CRediT authorship contribution statement

He Chen: Formal analysis, Methodology, Software, Writing – original draft. Xun Chen: Formal analysis, Methodology, Software, Writing – original draft. Jiufei Fang: Investigation, Data curation, Formal analysis, Writing – original draft. Wei Wang: Investigation, Data curation, Formal analysis, Writing – original draft. Jiarong Fan: Investigation, Data curation, Formal analysis, Writing – original draft. Yuan Xie: Software, Visualization, Writing – review & editing. Hanwei Lun: Software, Visualization, Writing – review & editing. Zhishun Liu: Conceptualization, Supervision.

Declaration of competing interest

Zhishun Liu is an editorial board member of this journal but had no role in the peer review process or editorial decision for this manuscript. The authors declare no other conflict of interest.