All English-language RCTs were considered eligible for inclusion, irrespective of geographical or publication status. In the case of randomized cross-over trials, only the first treatment period was included in the analysis. Exclusion criteria were applied to non-randomized clinical studies, quasi-RCTs, cluster RCTs, case reports, and studies where data were unavailable.

Eligible trials included those enrolling adults diagnosed with any form of cancer who were undergoing chemotherapy with drugs known for peripheral neurotoxicity and exhibited symptoms indicative of peripheral nerve injury.

Acupuncture therapies were considered, including manual acupuncture (MA) plus usual care (UC), electroacupuncture (EA) plus UC were included.

The control interventions comprised conventional treatment, usual care (UC), sham acupuncture (SA), and wait-list (WL) groups.

Studies were included if they measured at least one relevant outcome. The primary outcome in our meta-analysis were the pain scores, which were assessed using tools such as the Brief Pain Inventory (BPI), Visual Analog Scale (VAS), Neuropathic Pain Scale (NPS) for pain, specifically targeting neuralgia (11). Secondary outcomes, which indicate an improved quality of life, included the Functional Assessment of Cancer Therapy/Gynecologic Oncology Group-Neurotoxicity (FACT/GOG-Ntx) and current EORTC recommendations (12, 13). Outcomes were calculated as the absolute difference between post-treatment and pre-treatment values (mean ± standard deviation).

We systematically searched six databases—PubMed, Embase, Web of Science, and the Cochrane Central Register of Controlled Trials, WHO ICTRP, ChiCTR—from inception to August 2024 for relevant RCTs. The search was conducted using keywords such as ‘chemotherapy-induced peripheral neuropathy’, ‘CIPN’, ‘acupuncture’, ‘manual acupuncture’, ‘electroacupuncture’, ‘electro’, ‘randomized controlled trials’, and ‘RCT’. Two reviewers independently screened the literature by examining titles and abstracts, followed by a full-text review of potentially eligible studies to determine inclusion in the meta-analysis. See in Supplementary 2 .

Data extraction was conducted independently by two reviewers, with disagreements being resolved through discussion. Titles, abstracts, and keywords were independently screened by TH and QL to identify duplicate trials and to exclude clearly ineligible studies. The full texts of the studies were then examined to confirm adherence to the inclusion criteria. Any disagreements on study eligibility were adjudicated by a third reviewer.

The risk of bias of methodological quality was assessed using the Cochrane Risk of Bias (ROB) Tool 2 (14). This tool comprised seven parts (randomization process, deviation from intended interventions, missing outcome data, measurement of the outcome, selection of the reported result, overall biases) and ranked the methodological quality as unclear, low, or high. A third party (MSS or FRL) was consulted to assist in the final decision-making process.

In this study, we utilized the REMR method to evaluate REMR analysis to investigate the potential effects of various time factors. We observed a nonlinear relationship between the number of sessions, duration, frequency, and outcomes. This phenomenon may be attributed to different analytical models. In clinical practice, the efficacy of acupuncture is closely related to the accumulation of “dose”, and the dose-response relationship does not follow a simple linear pattern. Therefore, a nonlinear model may be more appropriate to describe this relationship (15, 16). Employing Stata 18.0 software, we incorporated random effects to account for heterogeneity across studies. We standardized acupuncture treatment parameters, such as the number of sessions, frequency, and duration, along with outcome measures pain scores, to ensure comparability across studies.

To identify potential dose-response relationships and account for variability across studies, we applied a random-effects model with a significance threshold of P < 0.05. We addressed potential non-linear relationships by employing a restricted cubic spline model with two knots positioned at the 25th and 75th percentiles of the dose distribution (17). The model fit was evaluated using the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC), and the reasonableness of the data representation was confirmed through visual inspection of the dose-response curves (17, 18). These steps established a comprehensive framework for analyzing the dose-response relationship between acupuncture treatment parameters and outcomes in CIPN patients

For the assessment and visualization of the risk of bias, we utilized R 4.3.1 software along with the “robvis” package. Meta-analysis and subgroup analysis were performed using the ‘meta’ package, which employs a random-effects model (DerSimonian-Laird method) and I² statistics to quantify heterogeneity. The ‘forestplot’ package was used to generate forest plots. Sensitivity analysis, defined by I² ≥ 50% or Cochran’s Q test p < 0.05, was conducted using a leave-one-out method to assess the influence of each study on the overall effect size and heterogeneity (19).

In the analysis of treatment efficacy, effect sizes were determined by computing the differences in mean values and standard deviations between the intervention and control groups. These effect sizes were interpreted according to the following criteria: small effect size (0.2), moderate effect size (0.5), large effect size (0.8), and very large effect size (1.2). Hedges’s g was used for continuous outcomes, especially in studies with smaller sample sizes, with the 95% confidence interval (CI) determining statistical significance. Heterogeneity was assessed using Cochran’s Q test (p<0.05) and I² statistic, categorized as not be important; (<40%), moderate (30%–60%), substantial (50%–90%), or considerable (>75%). Where meta-analysis was infeasible, a narrative synthesis was conducted (19).

Pain scores: Ten studies (20, 22–30) were evaluated in this manner, including 686 patients. The results indicated considerable heterogeneity among the studies, so a random effects model was used for statistical analysis. Acupuncture effectively alleviates pain in patients with CIPN. As shown in the figure, there was a significant difference between the experimental and control groups [SMD = -1.23, 95% CI = (-2.22, -0.24); p < 0.01; I² = 95%].

In the sensitivity analysis for the pain outcome, high heterogeneity was confirmed (I² = 95%). Sequential exclusion of individual studies showed minimal impact on the overall effect size and heterogeneity. Heterogeneity slightly decreased to I²= 91%, with the pooled effect size adjusted to SMD = -0.84 (95% CI: -1.53 to -0.14). Overall, the effect size remained statistically significant (P < 0.05) throughout the leave-one-out analysis, indicating the robustness of the findings ( Figure 3 ).

FACT/GOG-Ntx: A total of six studies (21, 23, 25–28), including 368 patients, were evaluated, and the results showed a beneficial effect of acupuncture compared to the control group [SMD = 0.95, 95% CI = (0.02, 1.88); P < 0.01] Figure 4 .

In the sensitivity analysis for the FACT/GOG-Ntx outcome, high heterogeneity was confirmed (I² = 93%). Sequential exclusion of individual studies revealed that after removing one study with some quality concerns (27), reducing heterogeneity to I² = 61% and slightly altering the pooled effect size from SMD = 0.95, 95% CI = (0.02,1.88) to SMD = 0.49, 95% CI = (0.05,0.92). This suggests that while the exclusion of this study decreases heterogeneity, the overall effect size remains statistically significant, indicating the robustness of the findings.

EORTC QLQ-C30 was reported in four studies (23, 25, 28, 30), including 276 patients. A random effects model analysis revealed that acupuncture significantly increased EORTC QLQ-C30 in CNPI patients compared to controls (SMD = 0.36, 95% CI = 0.03,0.68; P = 0.14; I² = 46%). Further sensitivity analysis demonstrated the robustness Figure 5 .

In the sensitivity analysis for the EORTC QLQ-C30 outcomes, substantial heterogeneity was observed (I² = 95%). Sequentially excluding individual studies revealed that after removing one small-sample study (25), heterogeneity was notably reduced to I² = 0%. This adjustment led to a change in the pooled effect size from SMD = 0.36, 95% CI = (0.03, 0.68) to SMD = 0.47, 95% CI = (0.21, 0.73). These findings suggest that this particular study significantly contributes to the overall heterogeneity. However, even after its exclusion, the intervention’s effect on quality of life remains statistically significant, underscoring the robustness of the overall results.

This subgroup analysis focused on studies related to changes in pain and FACT/GOG-Ntx outcomes. The results indicate that, in terms of pain indicators, manual acupuncture (20–23, 25, 27–30) [SMD = -1.46, 95% CI = (-2.66,-0.27); P <0.01; I² = 96%] showed significant improvement compared to the control group. The duration of acupuncture treatment, whether it weeks>8 weeks (24, 27, 29, 30) [SMD =-2.33, 95% CI = (-4.46,-0.19); P <0.01; I² = 98%) or ≤ 8 weeks (20–23, 25, 26, 28) [SMD =-0.47, 95% CI = (-0.84, -0.09); P = 0.02; I² = 61%], both demonstrated statistical significance. In the control group, whether it was patients receiving only usual care (20, 23–25, 27, 29, 30) [SMD = -1.55, 95% CI = (-2.92,-0.18) P <0.01; I² = 97%] or those on a WL (28), [SMD = -0.99, 95% CI = (-1.65,-0.03)] the treatment group showed statistically differences when compared to them. Furthermore, studies with a larger sample size (more than 15 patients) (20, 22, 23, 26–30) [SMD = -1.46, 95% CI = (-2.65,-0.27); I² = 96%; P <0.01] also showed a statistically significant association with treatment efficacy. However, when the experimental group used EA [SMD = -0.29, 95% CI = (-0.70,0.13); I² = 95%; P = 0.79], or the control group used SA [SMD = -0.27, 95% CI = (-0.65,0.10); I² = 0; P = 0.89], or when the sample size was ≤15 [SMD = -0.30, 95% CI = (-0.84,0.25); I² = 0; P = 0.78], no statistically significant association was observed Table 2 .

The study utilized the REMR method to identify the non-linear, quantitative-effectiveness relationship between acupuncture time parameters (treatment duration, frequency, and session) and pain outcomes in CIPN patients.

Acupuncture duration (week): Within a certain range of treatment duration, there is a negative correlation between duration and pain levels, with pain scores continuously decline, forming an L-shape. At the baseline (0 weeks), the pain score was 4.80 (95% CI: 3.03–6.57). As the treatment duration increased, the pain score gradually decreased, with a substantial drop at 6 weeks to 2.95 (95% CI: 1.13–4.76). Subsequently, the reduction in pain score stabilized, ranging from 2.67 (95% CI: 0.95–4.38) at 8 weeks to 2.37 (95% CI: 1.75–2.99) at 20 weeks, indicating that the pain-relief effect reached a stable level Figure 6 .

Acupuncture session (times): Based on the data, as the number of acupuncture sessions increases, the pain score exhibits a V-shaped trend. Initially, the pain score is 4.82 (95% CI: 3.04–6.60). With an increase in sessions, the pain score gradually decreases, showing a noticeable drop at lower doses (between 4 and 8 sessions), reaching close to zero. However, within this session range from 4 sessions 1.11 (95% CI: -3.54–5.77) to 12 sessions: 0.97 (95% CI: -2.53–4.46). At higher doses (between 16 and 18 sessions), the pain score begins to rise, reaching 2.29 (95% CI: 0.93-3.64) and 2.95 (95% CI: 2.29-3.61), with confidence intervals entirely above zero. This indicates a statistically significant pain relief effect at these doses. Therefore, while lower to moderate doses of acupuncture show a decreasing trend in pain scores, this effect is not statistically significant; at higher doses, the pain relief effect stabilizes and becomes statistically significant Figure 7 .

Acupuncture frequencies (times per week): The chart shows a clear downward trend in pain scores as the acupuncture frequency increases which exhibits an inverted V-shaped trend. At the baseline, the pain score is relatively high at 4.81 (95% CI: 3.02-6.61). As the weekly frequency gradually increases to 1 and 1.5 times per week, the pain score steadily decreases to 4.46 (95% CI: 3.73-5.19) and 3.13 (95% CI: 2.90-3.36), indicating a negative correlation between frequency and pain level. At higher frequencies, such as 2 and 3, the pain score significantly drops to 1.44 (95% CI: 0.52-2.36). Subsequently, the pain score decreases further to -2.06 (95% CI: -5.03–0.90), suggesting substantial pain relief, despite the initial non-significant result Figure 8 .

Integrating results. Based on the longitudinal analysis of time parameters, this study shows that acupuncture treatment duration exceeding 8 weeks, achieves the Minimum Clinically Important Difference (MCID). Within the first 6-8 weeks of treatment, the pain score shows a significant reduction, dropping to 2.95 (95% CI: 1.13-4.76) by the 6th week and further decreasing to MCID of 2.67 (95% CI: 0.95-4.38) at the 8th week, with minimal additional benefit from extending the treatment further. In addition, when the total number of sessions reaches 16-18, the pain relief effect becomes notable, with a score of 2.29 (95% CI: 0.93-3.64) at the 16th session, which is considered the optimal therapeutic dose, and then it slightly increases to 2.95 (95% CI: 2.29-3.61) at the 18th session. A frequency of twice times per week also significantly enhances pain relief, where the score drops to 1.44 (95% CI: 0.52-2.36). This treatment plan achieves optimal pain relief within a reasonable timeframe.

Adverse events: One trial (22) reported mild events, while other adverse reactions included discomfort, minor swelling, and bruising after acupuncture needle withdrawal (21, 26–28, 30). Four trials (20, 23, 25, 29) reported no adverse events, and one trial (24) did not mention any adverse events.