Different Acupuncture Therapies for Postoperative Residual Pain Following PKP/PVP for Osteoporotic Vertebral Compression Fractures: A Protocol for a Systematic Review and Bayesian Network Meta-Analysis

Abstract

Background

Osteoporotic Vertebral compression fracture (OVCF) is a common consequence of osteoporosis and accounts for a substantial proportion of osteoporotic fractures. Percutaneous vertebroplasty (PVP) and percutaneous kyphoplasty (PKP) are among the common treatment modalities for OVCF. In patients undergoing PVP/PKP, approximately one quarter experience postoperative residual pain, which seriously affects the health and quality of life of patients. Acupuncture has exhibited promising efficacy in alleviating residual pain after PKP/PVP; still some controversy remains. Clinical researches have demonstrated the effectiveness of acupuncture; however it is still unclear which acupuncture technique is best for relieving pain in patients who have had PKP/PVP. A Bayesian network meta-analysis (NMA) will be used to examine the relative effectiveness of several acupuncture modalities in postoperative OVCF patients.

Methods

The search strategy covers PubMed, Cochrane Library, Web of Science, Embase, Wanfang Database, VIP Database, and China National Knowledge Infrastructure (CNKI) from database inception to March 15, 2026, with language restrictions applied to English or Chinese publications. Pain intensity serves as the primary outcome. Secondary outcomes include the Oswestry Disability Index (ODI), serum procollagen type I N-terminal propeptide, serum beta-C-terminal telopeptide of type I collagen, and adverse events. A Bayesian NMA will be performed using MetaInsight and Stata 14. The protocol for this review has been registered with PROSPERO (No. CRD420261336974) on March 10, 2026.

Results

This Bayesian network meta-analysis will generate relative effect estimates for each acupuncture modality. The interventions will be ranked by the surface under the cumulative ranking curve (SUCRA), and the certainty of evidence for each outcome will be graded using the CINeMA framework.

Conclusion

Our findings are expected to provide guidance for residual pain management after PKP/PVP. This could help reduce the risk of analgesic‑related side effects and ultimately improve patients’ quality of life.

Introduction

Osteoporosis (OP) results in a predisposition to bone fragility and subsequent fractures.¹ Pathophysiologically, this condition is driven by a disequilibrium between bone degradation and deposition, driving progressive bone loss. Nationwide in China, OP affects 32% of adults of age ≥65 years, with nearly half of cases occurring in females and roughly one‑tenth in males.²

Osteoporotic vertebral compression fracture (OVCF) occurs at a rate that is notably high among osteoporotic fractures, predominantly involving the T10–L2 vertebral segments. It typically presents with low back pain, reduced mobility, loss of height, and kyphosis, ultimately leading to decreased quality of life and even shortened life expectancy.³ Prognostic modeling in post‑orthopedic populations has further underscored the substantial clinical burden and mortality risk associated with these patients.⁴ It is estimated that approximately 750,000 cases of OVCF occur annually in the United States. The occurrence of OVCF among Chinese adults aged ≥50 years is documented at 152.13 cases per 100,000 individuals.^5,6 By 2035, fragility fractures in China are projected to total 4.83 million yearly, increasing to 5.99 million by 2050.²

Treatment strategies for OVCF can be broadly classified as non‑surgical or surgical. The former encompasses bed rest, analgesics, physical therapy, and anti‑osteoporosis interventions. However, this approach requires not only a prolonged treatment duration but also prolonged bed rest, which may further lead to complications such as lower limb bone loss, deep vein thrombosis, and pulmonary infection.⁷ The primary surgical approaches are percutaneous vertebroplasty (PVP) and percutaneous kyphoplasty (PKP), both of which contribute to pain relief, faster recovery, and a well‑established safety profile. Due to these advantages, they are widely utilized in clinical practice and are considered effective treatment modalities for OVCF.

Studies have shown that 1.8% to 15.6% of patients may still experience residual pain after PVP.⁸ Research has found that in the early postoperative period following PKP or PVP, 49.6% of patients did not experience significant pain relief, and 28.3% still reported pain during long-term follow-up.⁹ Further studies indicated that late postoperative residual pain was associated with bone mineral density, treated vertebral level, surgical technique, bone cement dosage, and restoration rates of anterior and central vertebral body height. The persistence of postoperative residual pain may prolong functional recovery, reduce patient satisfaction, and potentially lead to further medical interventions.

Therefore, further exploration is needed to identify new approaches for alleviating residual pain after PKP/PVP.

Currently, acupuncture has spread to 183 countries and regions worldwide.¹⁰ A study performed in the US has indicated that within the top 99 applications of acupuncture, pain represents the largest category.¹¹ Acupuncture has shown favorable effectiveness for knee osteoarthritis,^12,13 low back pain,^14–16 lumbar disc herniation,^17–19 postoperative pain,^20–22 and migraine,^23,24 among others. Numerous reports have explored the effectiveness of acupuncture for residual pain after OVCF surgery. However, considerable heterogeneity exists in the types of acupuncture used across these studies, including manual acupuncture, warm acupuncture, silver needle, fire needle, and scalp acupuncture, electroacupuncture, etc. Previous meta-analyses have categorized different acupuncture modalities into a single group for comparison with pharmacotherapy or placebo, thereby overlooking the characteristics of individual acupuncture techniques and lacking comparative evaluations among them. Therefore, further investigation is warranted to elucidate the comparative effectiveness of various acupuncture modalities for postoperative residual pain in OVCF patients.

Network Meta-Analysis (NMA) is an extension and advancement of traditional meta-analysis. Its core value lies in synthesizing evidence from both direct and indirect comparisons, enabling the comprehensive evaluation and ranking of multiple interventions within a unified analytical framework. NMA currently employs two main approaches: the classical frequentist method and the Bayesian method derived from Bayes’ theorem. The frequentist approach can produce biased estimates due to issues such as iterative maximum likelihood estimation and model instability. In contrast, the Bayesian approach not only addresses these shortcomings but also enables the ranking of all included interventions through posterior probabilities, yielding more reliable and robust results. For these reasons, Bayesian NMA is the preferred approach.²⁵

A recently published protocol has also proposed a systematic review and meta-analysis on acupuncture for residual pain after PKP. However, that study similarly focused only on overall safety and efficacy without ranking individual acupuncture modalities. In contrast, our study employs a Bayesian network meta-analysis to provide comparative rankings of different acupuncture techniques, incorporates a broader range of interventions, and prespecifies more comprehensive subgroup and sensitivity analyses. These features distinguish our work from the existing protocol and extend the current evidence base.⁷

To identify the optimal acupuncture intervention strategy for residual pain in OVCF patients following PKP/PVP, we will carry out a Bayesian NMA evaluating the effectiveness of diverse acupuncture modalities.

Methods and Design

This study will be conducted as a systematic review and NMA to evaluate the efficacy of different acupuncture modalities for postoperative residual pain following PKP/PVP in patients with OVCF. We will report this systematic review in accordance with the PRISMA‑NMA extension, and the protocol was developed following the PRISMA‑P 2015 Checklist (Supplementary Materials PRISMA-P2015 Checklist). The protocol for this review has been registered with PROSPERO (No. CRD420261336974) on March 10, 2026. Any modifications will be meticulously recorded in the PROSPERO record.

Eligibility Criteria

Study Designs

Only randomized controlled trials (RCTs) written in English or Chinese will be considered eligible. Any studies not meeting these criteria, including non‑RCTs, will be removed.

Participants

The eligibility criteria were informed by the Guidelines for the Diagnosis and Treatment of Primary Osteoporosis (2022 Edition)²⁶ and the OVCF treatment guidelines.²⁷ Eligible participants are patients aged ≥60 years diagnosed with OVCF who experience persistent back pain following PKP/PVP, with a Visual Analog Scale (VAS) score of ≥4 at one month postoperatively, irrespective of sex, race, nationality, educational background, economic status, or disease severity.²⁸ Patients will be excluded if they present with new vertebral fractures or vertebral infections after surgery, intraspinal leakage of bone cement with compression of the spinal cord and/or nerve roots, or pathological fractures caused by malignant tumors.

Interventions

Eligible interventions consist of multiple acupuncture modalities, including manual acupuncture, silver needle, abdominal acupuncture, warm acupuncture, Fu’s subcutaneous needling, fire needle, scalp acupuncture, electroacupuncture, and wrist‑ankle acupuncture.

Comparison

The control group will receive either no intervention, sham acupuncture, placebo, nonsteroidal anti-inflammatory drugs (NSAIDs), or Chinese herbal medicine as control interventions. Due to substantial clinical heterogeneity among the control interventions, we will conduct separate network meta-analyses for distinct types of comparators to preserve the transitivity assumption.

For Network 1, no intervention, sham acupuncture, and placebo will be treated as separate intervention nodes rather than being pooled as a common reference arm, to preserve clinical distinctiveness. Only in a sensitivity analysis will we consider combining them, with the rationale and limitations explicitly discussed.

For Network 2, the active comparator‑controlled comparisons will include nonsteroidal anti‑inflammatory drugs (NSAIDs) and Chinese herbal medicine as active controls.

Outcomes

The Primary Outcome

Pain intensity, measured by standard pain scales (eg, VAS, NRS). To minimize heterogeneity arising from variations in follow‑up duration across trials, we will define a primary time window for the main analysis: the assessment closest to 1 month post‑treatment (within a window of 4 to 8 weeks). If a study reports multiple time points within this window, the one closest to the primary endpoint will be prioritized. For studies not reporting outcomes within this window, the earliest available post‑treatment time point will be used, and the deviation will be clearly noted.

Secondary Outcomes

1. Oswestry Disability Index (ODI)

2. Serum procollagen type I N-terminal propeptide (PINP)

3. Serum beta-C-terminal telopeptide of type I collagen (β-CTX)

4. Adverse events, including needling-related reactions such as syncope, hematoma, needle bending, needle retention, and local infection.

The integration of bone turnover biomarkers, including PINP and β‑CTX, with clinical characteristics has been highlighted as a valuable approach for assessing treatment response in osteoporotic populations.²⁹

Exclusion Criteria

1. Non‑English and non‑Chinese studies will be excluded.

2. Non‑RCTs (eg, cohort, case‑control) will be excluded.

3. Other non‑original publications (eg, case reports, editorials, conference abstracts) will be excluded.

4. Studies with insufficient outcome data for meta‑analysis will be omitted.

5. Duplicate publications, retracted articles, and fabricated data studies will be ineligible.

Search Strategy

Databases will be searched from their establishment to March 15, 2026. The search will cover the following sources: PubMed, the Cochrane Library, Web of Science, and Embase for English‑language publications, alongside the China National Knowledge Infrastructure (CNKI), Wanfang Database, and the VIP Database for Chinese‑language publications. Additionally, we will search trial registries including ClinicalTrials.gov and the Chinese Clinical Trial Registry (ChiCTR) to identify unpublished or ongoing trials. Grey literature will be sourced from the CNKI Conference Proceedings Database. Reference lists of included studies and relevant systematic reviews will also be manually screened to identify additional eligible studies. Detailed search strategies are available in the Supplementary Materials (Appendix Tables 1–9).

Study Selection

We will employ EndNote X9 to organize and manage the identified records. Duplicate records will be removed in EndNote prior to screening. Two researchers will independently screen titles/abstracts and subsequently evaluate full texts. Differences will be settled by consultation with the corresponding author, and the overall screening workflow will be shown in Figure 1.

Figure 1.

Flowchart of study selection.

Data Extraction

Data will be extracted using Microsoft Excel 2019 as follows:

1. Study characteristics: title, authors, and publication date.

2. Patient characteristics: study location, sex distribution, mean age, and number of treated vertebral levels.

3. Methodological characteristics: random sequence generation, allocation concealment, blinding, incomplete outcome data, selective reporting, other sources of bias, and baseline characteristics.

4. Outcome data: primary outcomes, secondary outcomes, and adverse events.

For missing data, the relevant authors will be approached. If no feedback is provided after two contact attempts within two weeks, we will include them in a sensitivity analysis using best-available published data and compare the results with analyses where they are excluded, to assess the potential impact of non-response bias.

Risk of Bias Assessment

Two researchers will assess risk of bias independently using the Cochrane Risk of Bias 2.0 (RoB 2.0) tool. The tool evaluates bias across five domains: the randomization process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result.³⁰ For each domain, studies will be classified as having low risk, high risk, or some concerns. Any discrepancies will be settled by consulting the corresponding author.

Network Meta-Analysis

To assess the feasibility of conducting a network meta‑analysis, we performed a preliminary scoping search in PubMed and CNKI prior to finalizing the protocol. After screening titles and abstracts, we identified 24 potentially eligible randomized controlled trials (23 in Chinese, 1 in English), providing reasonable confidence that a quantitative synthesis is feasible.

The analysis will be performed using MetaInsight (https://apps.crsu.org.uk/MetaInsight), a free, interactive web‑based tool developed by the UK NIHR Complex Reviews Support Unit.³¹ The analysis relies on the R packages gemtc and BUGSnet, with simulations conducted using Markov chain Monte Carlo (MCMC) methods. Multi‑arm trials will be included without being split. The correlation between treatment arms will be accounted for by using a multivariate normal likelihood for the effect estimates. We will adopt the R packages gemtc and BUGSnet for their automatic handling of this structure. We will ensure full reproducibility of the results.

Prior to any synthesis, network connectivity will be assessed using network plots. If the network is disconnected, indirect comparisons between disconnected subnetworks will not be performed; only pairwise meta-analyses will be presented. For nodes supported by only a single trial, the limited evidence base will be clearly noted and rankings interpreted with caution.

Bayesian estimation will be carried out using four Markov chains, with 20,000 burn‑in iterations followed by 50,000 simulation iterations. We will assess convergence via the Potential Scale Reduction Factor. Figures between 1.00 and 1.05 indicate satisfactory convergence, with lower values reflecting greater stability. If PSRF values exceed 1.05, we will sequentially (1) increase burn‑in to 50,000 and total iterations to 100,000, (2) apply thinning (eg, every 10th iteration), (3) adjust prior distributions (eg, using a half‑normal prior with smaller scale for τ), and (4) re‑parameterize the model if needed. Persistent non‑convergence will be reported as a limitation. We will adopt a random‑effects model to adjust for potential diversity in treatment effects across trials. A fixed‑effects model will be applied instead when data are very sparse or when heterogeneity is minimal, defined as I² <25%. The presentation of comparative effect estimates will use MD or OR along with 95% CIs, determined by variable type. The presence of significant heterogeneity will prompt exploration of its potential sources using subgroup analysis or meta‑regression. The ranking order of interventions will be determined by the surface under the cumulative ranking curve (SUCRA), where greater values reflect superior positions.

Assessment of Transitivity and Inconsistency

Transitivity is a fundamental assumption for how trustworthy indirect comparisons in NMA are. This assumption demands that effect modifiers be comparably spread among the included trials to ensure interpretable and unbiased estimates. Evaluation of transitivity will be performed by examining key study‑level and patient‑level characteristics, such as age, OVCF history, sex, surgical segment, and follow‑up duration.

Transitivity requires that effect modifiers be similarly distributed across comparisons. Given the fundamental difference between placebo-type controls (sham acupuncture, no intervention) and active pharmacological controls (NSAIDs, herbal medicine), we will not directly compare these two categories within the same network. Instead, we will conduct separate NMAs as described above. Within each network, we will compare the distribution of key effect modifiers (age, sex, baseline pain severity, OVCF history, treated segments) across studies to ensure transitivity holds.

In the presence of closed loops within the evidence network, both global and local inconsistency between direct and indirect evidence will be examined. Global inconsistency will be assessed by contrasting the consistency and inconsistency models via the Deviance Information Criterion. The node‑splitting method will evaluate local inconsistency.

In cases where significant inconsistency is detected, we will re-examine the original studies and the direct and indirect evidence sets. We will focus specifically on the comparability of key effect modifiers—such as age, male/female ratio, and OVCF history—to determine whether clinical heterogeneity might explain the observed inconsistency.

Assessment of Publication Bias

We will identify publication bias by using Stata 14. To detect small‑study effects or publication bias, we will construct comparison‑adjusted funnel plots and carry out Egger’s test.

Sensitivity Analysis and Subgroup Analysis

We will evaluate the reliability of the findings by conducting sensitivity checks that exclude reports judged as high risk and those with fewer than 30 participants per arm. If this prespecified threshold leaves too few studies for a meaningful sensitivity analysis, we will first attempt an alternative threshold (eg, ≥20 participants per arm) where feasible. Should this also prove insufficient, we will abandon the sample‑size‑based sensitivity analysis. In its place, we will assess the robustness of the findings by excluding studies at high risk of bias or those with unclear reporting. The limitations imposed by small sample sizes will be explicitly discussed.

Robust statistical procedures, including convergence diagnostics and sensitivity analyses, are essential for ensuring the reliability of predictive models and meta‑analytic findings.³² To explore heterogeneity and potential sources of variation in treatment effects, prespecified subgroup analyses will be performed according to gender, OVCF history, age (dichotomized at 70 years), number of treated segments, and risk of bias.^33,34 Subgroup analyses will be performed only when at least three studies are available for a given subgroup; otherwise, the findings will be described qualitatively. For any covariate with at least 10 available studies, meta‑regression will be used to examine whether these potential effect modifiers explain the observed heterogeneity or inconsistency. If fewer than 10 studies are available for any covariate, we will not perform meta‑regression; instead, we will conduct descriptive subgroup analyses where possible and qualitatively discuss the potential influence of the covariate based on available study characteristics.

Additionally, we will conduct a sensitivity analysis that restricts the control arm to only sham acupuncture/no intervention to test the robustness of the ranking results when active comparators are removed.

Furthermore, to assess the potential impact of time‑related heterogeneity, we will perform a sensitivity analysis stratified by follow‑up duration (eg, ≤4 weeks vs >4 weeks) to examine whether treatment rankings are consistent across different assessment time frames. If sufficient studies are available, follow‑up duration will also be explored as a covariate in meta‑regression; otherwise, we will qualitatively discuss its potential influence in the Discussion section.

Certainty of Evidence

We will assess the certainty of evidence for the NMA results of each outcome by using the CINeMA (Confidence in Network Meta-Analysis) online tool, which evaluates six domains: within-study bias, between‑study bias, indirectness, imprecision, heterogeneity, and inconsistency.³⁵ It incorporates additional domains into the GRADE assessment system, including study limitations, indirectness of evidence, inconsistency of results (comprising heterogeneity and incoherence), imprecision of findings, and publication bias. Unlike the traditional GRADE approach, CINeMA introduces a contribution matrix that plays a pivotal role by revealing the degree to which individual studies shape the outcomes of the network meta-analysis.

Discussion

In this NMA, we will rank treatment efficacy to provide clinicians with an evidence‑based hierarchy of acupuncture modalities, thereby guiding clinical decision‑making. We define pain intensity as the primary outcome and prespecify subgroup analyses based on patient characteristics to explore the influence of potential effect modifiers on treatment effects. Stratified analyses will be performed according to gender, OVCF history, age (dichotomized at 70 years), number of treated segments, and type of acupuncture to assess whether the relative efficacy of acupuncture modalities differs across subgroups.

Strengths and Limitations

This study has several methodological strengths designed to ensure transparency and reproducibility throughout the research process. A systematic search of English and Chinese databases will be performed to minimize the risk of publication bias. Study selection and data extraction will be performed independently and in duplicate by two reviewers, with any disagreements resolved through discussion or arbitration by a third reviewer to minimize bias and errors. Risk of bias will be assessed using the RoB 2.0 tool, ensuring rigor and standardization in study quality evaluation. Moreover, CINeMA will be used to determine the certainty of evidence across comparisons, providing a thorough and open assessment of the credibility of the overall results.

Admittedly, this study also has certain limitations. We acknowledge that non‑response from authors may introduce bias, and we will assess its potential impact through a sensitivity analysis as described in the Data Extraction section. The search strategy is restricted to studies published in English or Chinese, which may result in the omission of relevant literature published in other languages. The inclusion of both inactive and active controls in a single NMA could violate transitivity. To address this, we have revised the analysis plan to perform separate NMAs for placebo/sham controls and active comparators. This approach reduces heterogeneity in effect modifiers across comparisons and improves the validity of indirect estimates. Moreover, given the inherent complexity of acupuncture interventions, considerable heterogeneity may arise from variations across studies in acupoint selection, needling techniques, electroacupuncture parameters (eg, frequency, waveform, intensity), treatment frequency and duration, as well as practitioner qualifications and experience. Additionally, the implementation of blinding in acupuncture‑related RCTs faces unique challenges: blinding participants is difficult, and blinding practitioners is nearly impossible. This may introduce within‑study bias (eg, detection or performance bias), thereby affecting the reliability of effect estimates. Furthermore, given the large number of acupuncture modalities and the splitting of control arms into separate networks, the evidence network may be disconnected or supported by only sparse evidence, which could limit the feasibility of indirect comparisons and affect the reliability of the ranking results. We will assess network connectivity prior to synthesis and interpret the findings with caution.

Funding Statement

This study is supported by the Famous Traditional Chinese Medicine Studio Construction Project of Shanghai Pudong New Area Health Commission(No. PDZY-2025-0707), the “Hu Zongde” Shanghai Famous Traditional Chinese Medicine Practitioner Studio Grassroots Workstation Project (No. JCGZZ-2024012) and the Scientific Research Project of Shanghai Pudong New Area Health Commission (No. PW-2024B-16).

Systematic Review Registration

PROSPERO CRD420261336974.

Disclosure

Hao Wang and Siyao Wang are co‑first authors for this study. Rui Li and Zouqin Huang are co-correspondence authors for this study. The authors declare that no competing interests in relation to this work.