Study selection

The literature search yielded a total of 1521 potentially relevant publications. Among them, 606 studies were removed due to duplication. Subsequently, 356 articles were excluded based on the initial screening. Upon examination of the full text of 534 research articles, 25 studies [22, 25, 26, 34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55] met the inclusion criteria and were finally included in this meta-analysis. The literature screening flow chart is shown in Fig. 1.

Fig. 1

Study flow diagram. (Abbreviation: POF: premature ovarian failure)

Connection between keywords

The visual mapping showed that keywords were divided into 3 clusters: “Stem cell therapy for POF”, “Ovaries and oocytes-related organismal processes”, “EVs and ovarian cancers” (Fig. 2A). Research focuses include studying the pathophysiological mechanisms of POF and exploring stem cell-based EV therapies and innovative bioengineering methods. The citation track map in Web of Science database depicts the citation between important works of the research field (Fig. 2B). This field experienced an outburst between the year 2015–2020. We also found that in the PubMed database, published documents related to EVs and POF are gradually increasing (Fig. 2C). It indicates that research in this field has developed rapidly and is in a rapid rising stage, highlighting the necessity of such meta-studies.

Fig. 2

Bibliometric analysis. A Connection between keywords map. Cluster 1 (blue): “Stem cell therapy for POF”. Cluster 2 (green): “Ovaries and oocytes-related organismal processes”. Cluster 3 (red): “EVs and ovarian cancers”. B Citation track map. Each circle represents documents (including DOI numbers) that have been cited more than 20 times, and are defined as representative documents in the field. The larger the circle, the more times it has been cited. C Analysis of annual publishing trends in PubMed database. (Abbreviation: EVs: extracellular vesicles; POF: premature ovarian failure; DOI: digital object identifier)

Study characteristics

The characteristics of the included studies are shown in Supplementary Table 1. The 25 studies spanned from 2016 to 2023 (Fig. 3A), with the majority (21 trials) conducted in China [25, 26, 34,35,36,37,38,39, 41,42,43,44, 46,47,48,49,50,51, 53,54,55], followed by 2 trials in Iran [45, 52], one in the United States [22], and one in Egypt [40] (Fig. 3B). The total sample size across studies was 339 animals, with 173 allocated to the EVs treatment group and the remaining to the control group. Seventeen studies utilized mice as the animal model, while the remainder employed rats (Fig. 3C). Various methods were employed for disease modeling, include cyclophosphamide (CTX), CTX + busulfan (BUS), cisplatin, 4-vinylcyclohexene diepoxide (VCD), and D-galactose (D-gal) (Fig. 3D). The EVs utilized in the trials were derived from different stem cells sources, including umbilical cord mesenchymal stem cells (UC-MSCs), bone marrow mesenchymal stem cells (BMSCs) and adipose tissue mesenchymal stem cells (ADSCs), etc. Despite EVs being derived from cells of different species, each study independently demonstrated their effectiveness against POF in animal models. The route of drug administration primarily involved intravenous injection, although ovary and intraperitoneal injections were also used (Fig. 3E).

Fig. 3

An overview of studies characteristics, including distribution of (A) publications by year (B) region (C) animal model (D) disease model (E) route of administration

Risk of bias in the eligible studies

The included articles did not specify whether animal groups were randomly generated, nor did they provide details on randomization and allocation concealment methods, resulting in an unclear risk of selection bias (Fig. 4). While these studies did not extensively discuss blinding of personnel, it is evident that the animals were unaware of the group assignment, thus ensuring blinding of participants. Overall, the risk of reporting bias in these articles was low. Thirteen studies demonstrated a low risk of attrition bias, while one study had incomplete outcome data, resulting in a high risk of attrition bias.

Fig. 4Outcomes of the meta-analysisOvary weight

Eight studies reported the effect of stem cell-derived EVs on ovary weight [34, 38, 42, 44, 48, 51, 52, 54]. The meta-analysis revealed a significant increase in ovary weight following the administration of stem cell-derived EVs (SMD = 3.88; 95% CI: 2.50 ~ 5.25; P < 0.00001; I2 = 70%, P = 0.0007) (Fig. 5A).

Fig. 5

Forest plots depicting the comparison between the stem cell-derived EVs and control groups: A Ovary weight; B Pregnancy rate; C Count of births. (Abbreviation: EVs: extracellular vesicles)

Pregnancy rate

Four studies, encompassing 6 trials, examined the efficacy of stem cell-derived EVs on pregnancy rate, demonstrating a significantly improvement with EV administration (RR = 3.88; 95% CI: 1.94 ~ 7.79; P = 0.0001; I2 = 0%) (Fig. 5B) [22, 48, 49, 52].

Count of births

Analysis of data from 12 trials revealed that administration of stem cell-derived EVs significantly increased the count of births in POF animals (SMD = 2.17; 95% CI: 1.31 ~ 3.04; P < 0.00001; I2 = 69%) (Fig. 5C) [22, 25, 34, 40, 44, 45, 48, 49, 52, 54].

Follicle count

The effects of stem cell-derived EVs on primordial, primary, secondary, and antral follicle counts were assessed across 16, 14, 14, and 15 included trials, respectively. The results demonstrate that administration of stem cell-derived EVs significantly increased primordial follicle count (SMD = 3.75; 95% CI: 2.30 ~ 5.19; P < 0.00001; I2 = 86%), primary follicle count (SMD = 2.99; 95% CI: 1.83 ~ 4.14; P < 0.00001; I2 = 78%), secondary follicle count (SMD = 3.21; 95% CI: 2.03 ~ 4.38; P < 0.00001; I2 = 81%), antral follicle count (SMD = 3.41; 95% CI: 2.31 ~ 4.51; P < 0.00001; I2 = 81%) (Fig. 6).

Fig. 6

Forest plots depicting the comparison between the stem cell-derived EVs and control groups: A Primordial follicle count; B Primary follicle count; C Secondary follicle count; D Antral follicle count. (Abbreviation: EVs: extracellular vesicles)

The serum levels of FSH, E2 and AMH

We analyzed data from 18, 22 and 21 trials, respectively, to evaluate the effect of stem cell-derived EVs on the serum level of AMH, E2, and FSH. All studies employed enzyme-linked immunosorbent assay (ELISA) for detection. The analysis showed that administration of stem cell-derived EVs significantly increased the level of AMH (SMD = 4.15; 95% CI: 2.75 ~ 5.54; P < 0.00001; I2 = 88%) and E2 (SMD = 2.88; 95% CI: 2.02 ~ 3.73; P < 0.00001; I2 = 80%), while reducing the level of FSH (SMD = -5.05; 95% CI: -6.60 ~ -3.50; P < 0.00001; I2 = 90%) (Fig. 7).

Fig. 7

Forest plots depicting the comparison between the stem cell-derived EVs and control groups: A AMH; B E2; C FSH. (Abbreviation: EVs: extracellular vesicles; AMH: anti-Müllerian hormone; E2: estradiol; FSH: follicle-stimulating hormone)

Subgroup analysis results

The efficacy of EVs and heterogeneity in meta-analysis may be influenced by various factors including the source of EVs, animal species, disease model, EVs administration route, and test timepoint. Therefore, we conducted a series of subgroup analyses on the primary outcome indicators based on these conditions (Supplementary Table 2).

Based on the source of EVs, the main subgroups included UC-MSCs, clonal mesenchymal stromal cells (cMSCs), BMSCs, amniotic fluid mesenchymal stem cells (AFSCs), ADSCs. UC-MSCs, and cMSCs were found to significantly increase ovary weight, count of births and pregnancy rate. Additionally, UC-MSCs demonstrated an increase in the follicle count. However, the effects of BMSCs, cMSCs, and AFSCs on different types of follicles showed unstable statistical differences.

The animal species utilized in the study comprised mice and rats. Subgroup analyses indicated statistical differences in all outcome indicators, except for the rat subgroups of primordial follicles and antral follicles, which showed no statistical difference (p > 0.05). Furthermore, the heterogeneity was reduced in these analyses.

The disease models of POF primarily involved CTX and CTX + BUS. Subgroup analyses based on disease models showed statistical differences in all outcome indicators. However, heterogeneity was not consistently reduced across the analyses.

The main methods of EVs administration route include ovary injection, tail vein injection, and intraperitoneal injection. The results of subgroup analysis showed that there was no statistically significant difference between ovarian injection + tail vein injection on primary follicle and antral follicle count, and there was also no statistical difference on the effect of tail vein injection on primary follicle count. Additionally, the EVs administration route showed statistical differences in increasing ovary weight and count of births, accompanied by reduced heterogeneity.

The follow-up period and testing timepoint, mainly ranging from 1 day to 12 weeks after the last transplantation, was analyzed in subgroup analyses. Results indicated that when the test timepoint was 1 day, subgroup analysis outcomes were not statistically significant (p > 0.05), whereas significant statistical differences were observed for the remaining timepoints.

Sensitivity analysis and publication bias

Table 1 summarizes the sensitivity analysis and publication bias of this meta-analysis. Each study was individually excluded to assess its impact on the final effect, with results remaining consistent with those of all included studies, indicating the stability and reliability (Supplementary Fig. 1). However, in Egger’s test, only primary follicle count did not exhibit publication bias (p > 0.05). Subsequent trim and fill method analysis showed that ovary weight, count of births, primary follicle count, secondary follicle count, level of FSH were not trimmed, and the data in the funnel plot remained unchanged, suggesting no significant publication bias (Supplementary Fig. 2–4). Pregnancy rate, antral follicle count, level of AMH and E2 were trimmed using the trim and fill method (Supplementary Fig. 4 and Supplementary Table 3).

Table 1 Sensitivity and publication bias analysisTrial sequential analysis

TSA can assess whether results are supported by sufficient data. The TSA results indicate that there is enough data to draw definite conclusions about the count of births, ovary weight, as well as primordial, primary, secondary, and antral follicle counts. However, regarding pregnancy rate, the evidence is inconclusive according to TSA. This uncertainty may result in false negative or false positive conclusions due to the small number of experimental animals. Therefore, more animal experiments are warranted in the future to further validate these findings (Fig. 8).

Fig. 8

Trial sequential analysis results depicting the comparison between the stem cell-derived EVs and control groups. A Ovary weight. The cumulative Z curve crossed the conventional line, but did not reach the RIS. B Pregnancy rate. The cumulative Z curve crossed the conventional line, but did not reach the RIS. C Count of births. The cumulative Z curve crossed the conventional and reached the RIS. D Primordial follicle count. The cumulative Z curve crossed the conventional, but did not reach the RIS. E Primary follicle count. The cumulative Z curve crossed the conventional, but did not reach the RIS. F Secondary follicle count. The cumulative Z curve crossed the conventional and reached the RIS. G Antral follicle count. The cumulative Z curve crossed the conventional, but did not reach the RIS. (Abbreviation: EVs: extracellular vesicles; RIS, required information size)

Safety of stem cell-derived EVs

In the 25 studies included, the safety of EVs is mentioned in 6 of them [22, 34, 40, 48, 50, 55]. Two studies explicitly mention experiments on the safety of EVs, reporting safety outcomes, and indicating that none of the participants treated with EVs during the study period experienced EVs-related adverse events or complications [22, 48]. In the other 4 studies, EVs are generally accepted as safe [34,