To retrieve relevant articles, we searched databases, including the Cochrane Library, Web of Science, PubMed, Embase, Chinese Scientific and Technological Journal Database (VIP), China National Knowledge Infrastructure (CNKI), Wanfang Database (WF), and China Biomedical Literature Database (CBM), from their inception to 28 November 2024. The search strategy is the Medical Subject Headings (MeSH) and free text terms, with the objective of published studies. Table 1 demonstrates the search strategies applied when searching the PubMed database. Supplementary Table S2 shows the search strategy for all databases in detail.

To ensure the accuracy of the data, two researchers (Du and Ye) checked and verified with each other after independently extracting the data. In the event of inconsistent results, discrepancies were resolved either through discussion or, when necessary, by consulting an independent third party (Wei) to reach a consensus. The extracted data included the following: (a) basic information; (b) baseline characteristics of participants and the treatment methods applied; (c) key elements to assess the risk of bias; and (d) outcome measurements and associated data.

Two researchers (Du and Ye) independently assessed the risk of bias of the included RCTs and cross-checked their evaluations to ensure consistency. In case of disagreement, a third researcher (Wei) was consulted for reassessment. The risk of bias assessment of the included studies was conducted using the tool recommended in the Cochrane Handbook for Systematic Reviews (Version 5.1.0).

This study employed the mvmeta package in Stata 16.0 and Stata 12.0 software for network and routine meta-analyses, using mean ± standard deviation as effect size metrics for quantitative data analysis. All effect indicators were presented with mean difference (MD) and 95% confidence intervals (CIs). Literature quality and risk of bias were evaluated via Review Manager (RevMan), Version 5.3. Clinical and methodological features were compared to evaluate evidence network similarity, with outcome comparisons between any two interventions presented in league tables. A significance level of α = 0.05 was adopted. Heterogeneity was quantified using the I² statistic, with values exceeding 50% indicating substantial heterogeneity. Fixed-effects models were used when p ≥ 0.05 and I² ≤ 50%; otherwise, random-effects models were applied. When significant heterogeneity was detected, effect sizes were aggregated using a random-effects model, and subgroup analyses were performed for all studies according to different interventions, with sensitivity analyses performed using a dropout method for subgroups with inhibitions. The robustness of the results was assessed by systematically excluding high-risk studies or subgroups exhibiting severe heterogeneity on a leave-one-out method. Surface under the cumulative ranking curve (SUCRA) values (0%–100%) probabilistically ranked intervention efficacy, with higher values indicating greater likelihood of being the optimal treatment (19, 20).

The preliminary search yielded a total of 1,842 relevant articles across multiple databases, including the Cochrane Library (n = 31), Web of Science (n = 40), PubMed (n = 54), Embase (n = 129), VIP (n = 653), CNKI (n = 271), WF (n = 295), and CBM (n = 369). After multiple rounds of rigorous screening, 51 studies (21–71) were ultimately included in this meta-analysis. The publication dates of these studies spanned from 1999 to 2024. The detailed literature screening flowchart and results are presented in Figure 1 .

This NMA included a total of 51 studies involving 3,754 patients, all of which were RCTs (21–71). Twelve various methods of treatment were included in these studies: conventional treatment (CT), acupuncture (Acu), moxibustion (Moxi), abdominal acupuncture therapy (AAT), CIAA, EA, AST, awn needle acupuncture (ANA), umbilical acupuncture (UA), massage therapy (MT), acupoint application (AA), and umbilical moxibustion (UM). ( Supplementary Table S3 details the abbreviations and definitions of different therapies and conventional treatment.) For investigations using a three-arm experimental design, subgroup merging was used during the data processing phase. Data analysis was then conducted using a two-arm trial design. Supplementary Table S4 provides further details about the included RCTs. A comprehensive network diagram illustrating various comparisons across all outcomes is shown in Figure 2 .

This study included a total of 51 RCTs, with their quality comprehensively evaluated. Regarding the method of random sequence generation, five studies (42, 44, 51, 59, 70) assigned patients based on order of registration, visit sequence, or outpatient or inpatient number, resulting in the possibility of selection bias, which was judged to be high risk. Conversely, 31 studies (21, 25–29, 33–37, 39, 43, 47, 48, 50, 53–56, 58, 60–62, 64, 66–68, 71) were considered low risk because they employed random number tables or computer-based random sequence generators for participant allocation. However, 15 studies (22–24, 30–32, 38, 40, 41, 45, 46, 49, 52, 57, 65) mentioned randomization but did not describe the method of randomization and were therefore classified as having unclear risks.

Concerning allocation concealment, all studies (21–71) failed to clearly describe their allocation concealment methods, resulting in an unclear risk rating. Only one study (30) implemented blinding of participants and investigators, assessed as low risk, while the remaining studies (21–29, 31–71) did not mention blinding and were rated as unclear risk. All studies (21–71) maintained complete datasets. The presence of selective reporting bias was determined by verifying that the methods were aligned with the presented outcomes. All studies (21–71) adequately reported their findings and were deemed low risk in this aspect. One study (58) demonstrated a rigorous design with no apparent sources of bias not covered by other domains, rated as low risk, while the remaining studies (21–57, 59–71) retained unclear risks regarding other potential bias sources. Figure 3 illustrates the risk of bias assessment conducted with RevMan 5.3. Specific details of the evaluation are shown in Supplementary Table S5 .

A total of 13 studies (24, 26, 32, 37, 42, 50, 52–54, 56, 66, 67, 69) reported adverse events. Of these, six studies (24, 26, 37, 42, 53, 69) did not report specific adverse events, and the remaining seven studies (32, 50, 52, 54, 56, 66, 67) reported minor adverse events. All adverse events reported in these seven studies resolved after discontinuation of treatment and symptomatic treatment, which could be judged as mild adverse events, except for the adverse events reported in the study of Hou, where follow-up outcomes were not reported. There was no statistically significant difference in the incidence of adverse events in any of these studies compared with the control group. Specific details are presented in Table 3 .

The funnel plots for LH and E₂ demonstrate that the study points are approximately symmetrically distributed on each side of the center axis. This symmetry suggests a relatively low chance of publication bias in this study. By contrast, the funnel plots of AFC, FSH levels, and the Kupperman scores exhibit poorer symmetry, indicating potential biases or impact from small study effects. These findings are presented in Figure 9 .

The results of the heterogeneity analysis showed that all studies had an I ² ≥ 50%, and therefore, a sensitivity analysis was performed by adopting a leave-one-out method for all outcome indicators, the results of which are presented below. Forest plots and sensitivity analysis plots are presented in Supplementary Figures S1 – S10 .

In summary, inadequate standardization of testing methods, differences in intervention protocol design, inconsistent regimens, flawed randomization methods, and confounding interventions in the control group were the primary sources of heterogeneity. Furthermore, the limited number of original studies stems from the inability to reliably identify sources of heterogeneity for the Kupperman score and unspecified metrics. Future investigations with adequately powered samples are required to validate these findings.

This study represents the first NMA to include all AS therapies in the treatment of POI. Involving a total of 51 studies with 3,754 patients, this meta-analysis examined commonly utilized AS therapies for POI and their effectiveness in comparison to conventional drug therapy or placebo. The results indicate a comparative advantage of these AS therapies over CT, particularly in improving the hormone levels and menopausal symptoms in patients with POI.

Through SUCRA, the NMA assessed the effectiveness of all AS therapies to identify the most optimal treatment plan for POI. The results suggest that moxibustion combined with other AS therapies may be the most effective in improving the hormone levels, such as the AFC, E₂, and Kupperman scores. AST therapy has been shown to be significantly effective in improving FSH levels in patients with POI. Moxi + MT therapy benefits particularly in improving E₂ and Kupperman scores in patients with POI.

In summary, POI is a complex disorder that is caused by the interplay of genetic factors, congenital causes, autoimmune disorders, bad habits, and psychological pressure (12, 72), which lead to ovarian hypoplasia by age 40, infertility, and various menopausal symptoms. POI is recognized by elevated levels of gonadotropins and decreased levels of E₂, which disrupt the normal menstrual cycle and lead to symptoms such as hot flashes, night sweats, insomnia, and decreased bone mineral density (73, 74). This hormonal imbalance can also lead to other related symptoms such as cardiovascular risks and mood disorders (74, 75).

Our research found that combination therapies—especially those based on moxibustion—are more successful than monotherapies in treating POI. Specifically, moxibustion combined with other treatments such as acupuncture or MT has demonstrated superior outcomes in improving the hormone levels and alleviating clinical symptoms compared to monotherapies. Moxibustion is used to warm the meridians and reconcile qi and blood by burning strips or pillars of wormwood upon acupoints near the skin’s surface (10). POI is often accompanied by hormonal imbalances and psychological symptoms. Research shows that moxibustion has been shown to significantly improve E₂ levels and reduce FSH and LH levels, thereby enhancing ovarian function (76). Research (77) found that acupuncture with moxibustion at the lumbosacral area and acupuncture combined with warm palace moxibustion effectively alleviated the clinical symptoms and the levels of FSH and LH, while improving E₂ levels in the treatment of patients with POI.

Multiple studies have demonstrated that moxibustion significantly reduces Kupperman scores for clinical symptoms in POI, either alone or in combination with acupuncture, with superior efficacy to pure HRT. The underlying mechanisms are hypothesized to involve multi-target regulation of the neuro–endocrine–immune network, alleviating clinical symptoms such as hot flashes, night sweats, insomnia, and anxiety and promoting menstrual cycle restoration (30, 32, 46, 50, 54, 66, 70). A study (30) showed that tuina–moxibustion combined with Chinese herbal medicine for treating the spleen and kidney yang deficiency type of POI significantly reduced its Kupperman score, which was significantly better than that of the control group. The therapy resulted in a 96.67% improvement in menstruation-related symptoms in patients with POI. Research (54) found that the treatment group (acupuncture combined with herbal UM) showed a significant reduction in Kupperman scores, which were prominently lower than those from the control group, with particularly pronounced improvements in menopause symptoms and anxiety symptoms. The study (66) showed that acupuncture combined with moxibustion in treating the liver and kidney yin deficiency type of POI resulted in a lower Kupperman score than the control group, which remained consistently lower than the control group at 3 months of follow-up. The underlying mechanism of the immediate effect and long-term efficacy may be associated with the role of moxibustion in stabilizing HPO axis homeostasis. Research (50) evaluated acupuncture combined with fire dragon moxibustion in the kidney-deficiency type of POI and found that the treatment group achieved a Kupperman score of 10.13 ± 1.96, significantly lower than that from the Western medicine group (14.52 ± 2.27). The thermal effect of fire dragon moxibustion is proposed to activate cutaneous receptors and enhance local blood circulation, thereby alleviating genitourinary and somatic symptoms.

Nine RCTs (27–29, 32, 43, 45, 50, 59, 69) included in this study had single-group sample sizes of less than 30, which is considered to be a small study. In meta-analysis, studies with smaller sample sizes tend to report larger effect sizes, whereas studies with larger sample sizes report relatively smaller effect sizes, which can lead to small-study effects. The phenomenon suggests the possibility of publication bias, selective bias, or clinical heterogeneity, which can affect the reliability of the results of systematic reviews and meta-analyses. The funnel plots of LH and E₂ had better symmetry, indicating lower inter-study heterogeneity and better consistency of results, suggesting a low possibility of publication bias. In contrast, there was an asymmetric distribution of AFC, FSH, and Kupperman scores, suggesting that it might be related to the small sample study effect, selective publication, and heterogeneity in methodology, leading to the potential risk of bias they might have.

Studies (72, 78) show that acupuncture may improve the ovarian function of POI rats by regulating serum metabolite markers such as divanillyltetrahydrofuran ferulate, trans-ferulic acid, tryptamine, and neuraminic acid to exert antioxidant and antiapoptotic effects, which not only increase vascular permeability by upregulating the content of vascular endothelial growth factors in serum but also inhibit the expression of the JNK-mediated apoptosis pathway-related factors under oxidative stress by downregulating ROS levels in ovarian tissue, slowing down the process of follicular atresia, and improving ovarian function in POI model rats (79, 80). Moreover, research shows that EA could improve ovarian reserve function in POI rats by reducing the number of autophagosomes and autolysosomes, upregulating p62 expression, and downregulating Beclin-1 and LC3 expression, thus inhibiting autophagy of ovarian granulosa cells, and regulating the serum levels of FSH, E₂, AMH, and INHB (73, 81). Furthermore, studies have validated that moxibustion can improve impaired ovarian function by enhancing steroidogenesis in GCs through the activation of the cAMP/PKA/CREB pathway (74, 82), and improve ovarian reserve in rats with Cy-induced POI by attenuating mitochondrial dysfunction and NLRP3 inflammatory activation (75, 83). Recent studies indicate that Acu + Moxi can effectively improve clinical symptoms, regulate sex hormone levels, increase endometrial thickness, and increase the number of sinus follicles in patients with POI, thus promoting the recovery of ovarian function (84–86).

Autophagy, particularly the autophagic level in ovarian granulosa cells, is closely associated with POI (87). During autophagy regulating the occurrence and development of POIF, numerous autophagy-related genes and signaling pathways are involved, among which the PI3K/PTEN/AKT/mTOR pathway is relatively standard (88, 89). This pathway can influence the activation of primordial follicles by regulating the proliferation of granulosa cells in primordial follicles, and it plays a significant role in the development, growth, and atresia of primordial follicles (90). The study found that high expression of PTEN inhibits the phosphorylation of AKT. At the same time, stimulation of Guanyuan (CV4) and Sanyinjiao (SP6) can relieve the excessive inhibition of AKT by PTEN, ensuring the normal growth and development of follicles and the reserve function of the ovary (87). Research (91) demonstrated that acupuncture significantly increased the mRNA expression levels of PI3K, Akt, and mTOR in the ovaries of POF rats, accompanied by an increase in the number of mature follicles. Another study (92) indicated that acupuncture reduced granulosa cell apoptosis, which was potentially associated with the upregulation of the PI3K/Akt signaling pathway via hormonal regulation. Notably, EA was found to protect the primordial follicle pool and improve fertility in POF mice by downregulating the phosphorylation of PI3K/AKT/mTOR pathway proteins (93). Additionally, moxibustion therapy promoted ovarian function in POF rats, possibly by inhibiting the phosphorylation of the PI3K/Akt/mTOR signaling pathway (94).

Existing research usually compares a particular AS treatment with a traditional medication treatment or placebo, and the present NMA of POI is quite constrained. For example, a meta-analysis in 2024 found that acupuncture substantially decreased serum levels of LH, FSH, and FSH/LH ratio, and elevated E₂ levels (14). Another meta-analysis in 2024 on acupuncture treatment, which involved 13 RCTs, indicated that acupuncture had proven superiority over non-acupuncture in decreasing FSH levels and increasing E₂ levels, AMH, and general efficacy in women with confirmed POI (13). However, these studies were mainly focused on acupuncture or acupuncture plus medicine, with less comparison of varied AS therapies. In our work, we include all AS therapies in the management of POI. Differing from traditional meta-analyses, NMA facilitates the identification of optimal therapy options by directly and indirectly comparing and analyzing various therapies. This study included 51 articles covering almost all existing AS therapies for POI. It systematically assessed the improvement effects of multiple AS therapies on ovarian function, sex hormone levels, and clinical symptoms. Our results are consistent with those of other meta-analyses, indicating that AS therapies—specifically acupuncture and moxibustion—are effective and safe in enhancing menopausal symptoms and hormone levels in patients with POI, as well as significantly improving their quality of life (QOL).

There are certain limitations in this study. First, the selection criteria were confined to literature in Chinese and English, and papers that were written in other languages were excluded. This restriction may lead to failure to have a global perspective. Although our search focused on peer-reviewed journals, we acknowledge that materials such as conference abstracts and dissertations were not included due to data quality and reproducibility considerations. Second, there were only one or two studies available for some AS therapies. This lack of data might induce bias and limit the generalizability of the results. Third, it is difficult to completely blind individuals owing to the nature of the therapies. Interaction and communication between the patient and the practitioner are frequently necessary for AS therapy, and this might affect how the patient views and reacts to the treatment. Fourth, adverse events are significant for evaluating the safety of acupuncture. However, many original studies have not reported adverse events, and future research should focus on assessing the safety of acupuncture. Future multicenter, standardized RCTs with large samples and long courses of treatment are needed to extend the follow-up period and add evaluation indexes such as anti-Müllerian hormone (AMH), psychological status, and safety to further validate the efficacy and safety of AS therapy.