The systematic review protocol has been registered on the prospective international register of systematic review (PROSPERO: registration number is CRD42020206790) (https://www.crd.york.ac.uk/PROSPERO/).

All contents and report details were strictly referred to as Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) (22), as shown in Supplementary Material S1.

References were retrieved from nine databases: four English databases (i.e., PubMed, Embase, Cochrane Library, and Web of Science), and five Chinese databases, i.e., SinoMed (formerly Chinese Biomedical Database), Chinese National Knowledge Infrastructure (CNKI), Wanfang Data, China Biomedical Literature Database, and China Science Journal Database (VIP database) from inception to 25 February 2022. The search strategy was based on the guidance of the Cochrane handbook. The language was limited to English and Chinese. The search strategies are shown in Supplementary Material S2.

In addition, to reduce the publication bias, we searched for lists of relevant references. Clinical trial registries, i.e., Menstrual Disorders and Subfertility Group (MDSG) Specialized Register, Cochrane Central Register of Controlled Trials (CENTRAL), World Health Organization International Clinical Trials Registry Platform, Chinese clinical registry, and Clinical Trials. Gov, and we manually searched key journals and meetings such as the European Society for Human Reproduction and Embryology (ESHRE) and American Society of Reproduction Medicine (ARSM) for relevant articles, including relevant journals and conferences abstracts, by connecting with the coordinator.

Two reviewers (XYZ and HMY) independently selected the studies, extracted data, and downloaded the citations into Note Express software Version 2.6.1 (Aegean Sea software company, Beijing, China) for data management and eliminated the duplicate research by software. Any discrepancies were resolved through further discussion with the third reviewer (FGW).

For trials to be eligible, intervention dosage (duration and frequency) for acupuncture treatment is extracted in detail. We categorized the acupuncture dosages into three groups: low, moderate, and high. Moderate-dosage group was defined as “one menstrual cycle” of acupuncture duration [including the acupuncture conducted during the controlled ovarian hyperstimulation (COH) procedure in IVF or during frozen-thawed embryo transfer (FET)]. For the duration of acupuncture that was less than “one menstrual cycle,” we extracted studies in the “low-dosage group,” while for acupuncture duration that was more than “one menstrual cycle,” we extracted the studies in the “high-dosage group.”

We used the RevMan5.4.1 provided by the Cochrane Collaboration to analyze data. For dichotomous data (CPR, LBR, and EMP), we expressed the results for each study as the odds ratio (OR) with a 95% confidence interval (CIs). For continuous data (EMT, RI, PI, and S/D), we expressed the results as the difference or standardized mean difference (SMD) with 95% CI. If data could be synthesized, we used descriptive analysis to solve this problem. For pooled data, we used I-square (I²) statistics, which indicates the proportion of variability across trials not explained by sampling variation alone, and the Cochran Q-test for heterogeneity assessing (31). A fixed-effects model was performed for low heterogeneity (I² < 50% statistics or the Cochran Q-test, p > 0.05), otherwise, random-effects were performed. We used network construction between acupoints-associated and duration-associated to show their relationship visually.

To find out the dose-related efficacy of acupuncture, we divided the extracted data into three groups based on the different dosages of acupuncture intervention separately (2 or 4 days only, one menstrual cycle, or three menstrual cycles). We assessed the likelihood of publication bias by constructing funnel plots and the Eggers' test.

For pooled data, characteristics varied from different studies. We used subgroup analysis instead of meta-regression to find the source of heterogeneity, mainly considering the insufficient number of included studies. Subgroup analysis was based on four clinical characteristics: i) acupuncture duration; ii) intervention of different routine treatment (IVF-ET, FET, or clomiphene citrate (CC)/letrozole (LE) for ovulation induction); iii)acupuncture used alone or with other intervention; iv) mean age of participants (≥35 years or not) and three methodological characteristics: i) risk of blinding of participants and personnel (low risk or high risk); ii) risk of random sequence generation (low risk or high risk); and iii) risk of allocation concealment (low risk or high risk).

For each subgroup analysis, we performed a single covariate weighted random effects with 95% confidence interval in Stata version 16 (StataCorp) to investigate whether differences in effects of adjuvant acupuncture between the covariate's two subgroups were statistically significant. For each single covariate subgroup analysis, we calculated the interaction p-value of the test and the percentage of the heterogeneity explained by the covariate (I²).

Figure 1 shows the details of the study selection process.

The PRISMA flow diagram (Figure 1) shows the details of the study selection proces. Of the 101 identified trials, 76 studies were excluded after screening full-text (6 studies were not available for full-text. Six studies were not RCTs, seven studies had no outcomes for interests, 34 studies had the only outcome of EMT, which cannot represent ER in our opinion, and 34 other studies were also excluded because of low quality or repeated publication). Thus, 14 RCTs with a total of 1,564 participants met the inclusion criteria. Twelve (14, 16, 20, 33–41) studies were conducted in China [ten studies (14, 20, 33, 34, 36–41) were from the mainland, China and two were (16, 35) from Taiwan, China], and two (13, 42) were from Germany.

Table 1 shows the characteristics of RCTs included in the review. The differences in trial eligibility criteria were that six trials (14, 20, 34, 36, 39, 41) included women with RIF and one trial (37) included women with PCOS. The duration of the acupuncture sessions differed among trials (Table 1). Three studies (13, 16, 42) conducted two sessions around ET and one study (35) conducted 4 sessions during COH. Some women received acupuncture for almost one menstrual cycle [acupuncture was performed for 1 month before FET in one study (38) and during FET in three studies (36, 39, 40)]. Meanwhile, the other six trials (14, 20, 33, 34, 37, 41) lasted for three menstrual cycles before embryo transfer. Six studies (13, 16, 33, 35, 38, 42) conducted IVF-ET of routine treatment, seven studies (14, 20, 34, 36, 39–41) were performed in FET, and one study (37) administered letrozole (LE) and human chorionic gonadotropin (HCG) for the induction of ovulation in routine treatment.

For the frequency of the collected studies, one study (41) did not mention the frequency of acupuncture, three studies (33, 34, 40) conducted acupuncture every day, three studies (36, 37, 39) were conducted every other day, two studies (20, 38) were conducted three times a week, while one study (14) conducted for six times per cycle. However, all of the collected studies did not mention the total dosage of acupuncture treatments, except that three studies (13, 16, 42) were conducted only for 2 days (25 min before and after ET), and one study (35) was conducted for 4 days in total.

Besides, the acupoints-associated and duration-associated network construction chart (Figure 2) shows that various acupoints were selected in treating PER. Among the whole trails, SP6, EX-CA1, RE4, ST36, and LR3 were the top five acupoints, and three menstrual cycles were the most frequently used duration for the whole treatment.

The RCTs included in this study varied in study design, especially in ways of intervention of treatment group, such as, manual acupuncture (MA) alone (16, 36), electro-acupuncture (EA) alone (35, 38, 39), MA or EA with moxibustion (20, 33, 34, 37, 41), MA with auricular acupuncture (13, 42), and TEAS (14, 40). Meanwhile, the duration of acupuncture differed for trails, such as 2 or 4 days of acupuncture, one or three menstrual cycles. Finally, we categorized these trails according to the duration of acupuncture treatment in three groups: the high-dosage group [three menstrual cycles; (14, 20, 33, 34, 37, 41)], the moderate-dosage group [one menstrual cycle; (36, 38–40)], and the low-dosage group [2 or 4 days; (13, 16, 35, 42)].

To the best of our knowledge, this is the first comprehensive review addressing the pooled analysis of whether dose-related acupuncture therapeutic efficacy improves ER for infertile women. Finally, we collected 14 trials in our review, including four (36–39) updated relevant studies.

Generally, acupuncture can significantly improve the CPR and the ER parameters (EMT, EMP, RI, PI, and S/D) with moderate or high heterogeneity (I²: 56–95%), whereas the LBR was not statistically significant between the two groups with high heterogeneity (I² = 83%), also. For the primary outcome, CPR was significantly increased in the moderate and high acupuncture groups without heterogeneity, whereas in the low-dosage group, CPR failed to show a significant improvement with considerable heterogeneity (I² = 82%) (Table 3). For the ER parameters, EMP was significantly improved in the high-dosage group with a small heterogeneity (I²= 10%) and was not statistically significant in the moderate group with considerable heterogeneity (I² = 85%) (Table 3). Though the total heterogeneity of EMP was moderate (I² = 71%), we failed to find the sources of the heterogeneity (Figure 11A). The decrease of PI showed a significant difference in the high-dosage and moderate group between the acupuncture group vs. the controlled one with considerable heterogeneity (I² = 96% and I² = 79%, respectively), whereas, in the low-dosage group, no significant difference was found with considerable heterogeneity (Table 3). The total heterogeneity of PI was high (I² = 93%) in our analysis, and not only the duration of acupuncture but also the different routine treatments for infertile women were attributed to the source of heterogeneity (interaction P < 0.01, <0.01, respectively) (Figure 11B). EMT was significantly improved in the moderate- and high-dosage groups with significant heterogeneity (I² = 87%, I² = 80%, respectively), and was not statistically different in the low-dosage group with considerate heterogeneity (I² = 81%). The total heterogeneity of EMT was high (I² = 87%) too, and we found out that the different routine treatment was attributed to the source of heterogeneity (interaction P = 0.003). The heterogeneity was decreased and was 65% in the IVF-ET group and 84% in the FET group (Figure 11C). Though the significant decreases of RI were shown in the moderate- and high-dosage acupuncture groups, the heterogeneities were significant too (I² = 88%, I² = 77%, respectively) (Table 3). In our subgroup analysis, the routine treatment and the method of the compared intervention (placebo acupuncture was utilized or not) may be the sources of heterogeneity (Figure 11D). LBR was statistically improved in the high-dosage group while it was non-significant in the low-dosage group. S/D was significantly decreased in both moderate and high-dosage groups. However, the indicators of S/D and LBR showed high heterogeneities in the studies (I² = 95, I² = 83%, respectively) (Table 3). Considering the small numbers of included studies that measured the LBR and the S/D, the heterogeneity analysis was not performed. Above all, comparisons were evaluated as a very low to moderate level of evidence, even very low and low, mostly based on the GRADE (Table 3).

Thus, the therapeutic effect of acupuncture in different dosage groups is relatively weak so far. Though it is difficult to draw a definitive conclusion that acupuncture is more effective, we discussed the findings based on our subgroup analysis under the method of study design.

Above all, the evaluation among the outcomes from the included studies and acupuncture dosage resulted in different therapeutic effects. Though significant heterogeneities were not found, the trend of relatively more dosage showed better effects in CPR, EMT, PI (moderate or high dosage showed significantly improved while the low dosage did not), and EMP (high dosage was significantly improved while the moderate did not; Table 3).

Our findings of the relationship between acupuncture therapeutic efficacy and the dosage were consistent with the previous studies. White et al.(45) thought the dose may be affected by the state of the patient (e.g., nervous, immune, and endocrine systems); different doses may be required for different conditions. For the analgesia of acupuncture, a system review (46) evaluated whether the effect of acupuncture is dose-related for symptom management in knee osteoarthritis. Based on the consideration of the number of points needled during each treatment; de qi response; frequency of treatment per week; and several treatment sessions, the study found out that the effect may be associated with the dose of acupuncture, with a higher dosage (the number of points needled ≥9; de qi response; frequency ≥2 sessions a week; the total number of treatment sessions ≥8) related to better treatment outcomes in terms of relief of pain and dysfunction in patients with knee osteoarthritis. Gao (47) utilized functional MRI (fMRI) in evaluating acupuncture efficacy in treating migraine. Four weeks of acupuncture treatments strengthened the brain connection to improve therapeutic efficacy in migraine when compared with 2 weeks. For the acupuncture modulation in default mode network (DMN) of the brain with fMRI, Lin et al. (48) suggested that enhancing the acupuncture dose could potentially be applied as a means of modulating brain activity. For improving the outcome of women who underwent IVF, Magarelli et al.(12) suggested that patients receiving acupuncture weeks before the stimulation of IVF (usually 4 weeks at two treatments per week until the day of retrieval) and in the times before egg retrieval and ET may benefit more than those patients treated just pre-/post-ET. In a systematic and meta-review, Zheng et al. (49) reported that the pooled biochemical pregnancy rate (BPR) was significantly higher in the acupuncture group than in the controls (OR = 2.07, P = 0.02). The CPR, LBR, and IR results tended to be higher when acupuncture treatment was conducted during the COH time (from the starting time of the menstrual cycle to the oocyte retrieval). Whereas, the pooled BPR, CPR, OPR, LBR, IR, and MR showed no significant differences when acupuncture was performed only around the day of ET (i.e., CPR: p = 0.48).

Infertility for women with PER is a complex medical issue with multiple factors, often associated with previous gynecological issues/pathology (e.g., fibroids, endometriosis, autoimmune disease, etc.). Although patients may benefit from two sessions of acupuncture treatment (13), many studies (15–17, 33) failed to repeat this outcome. Thus, chronic conditions of long-term duration need larger dosages of acupuncture, especially in patients with multiple health issues.

Why S/D is lower in three-cycle treatments than in one cycle? It was found that there were more stressful and uncomfortable feelings during the IVF process in the acupuncture group (50). Also, the S/D is a parameter of the uterine artery. The S/D of women will change under too much stress and discomfort. Besides, how many acupuncture sessions should be performed for participants who suffered PER is necessary to reconsider because of the financial burden and time cost for PER (51).

Heterogeneity was expected due to the variables of the routine treatment, acupuncture treatment duration, performance bias (blinding of participants and personnel), and the age of participants.

For routine treatment for infertile women, to the best of our knowledge, IVF-ET is a procedure in which fresh embryos would be transferred after ovarian stimulation, while FET is frozen embryo transfer. Ovarian stimulation for IVF would affect the luteal phase and the window of implantation (WOI) by altering the endocrinological environment of the endometrium. That is, ovarian stimulation will reduce ER in IVF-ET (52), whereas FET in subsequent cycles avoids possible adverse effects without ovarian stimulation. For women with PCOS, LE has been used as the traditional medication to induce ovulation for being superior to CC in ER (53). But it is uncertain whether LE for PCOS will improve the EMT compared to the programmed COH cycle (54).

For blinding of participants and personnel, placebo acupuncture treatment is not inert in numerous studies (55), and the possibility that some of the sham interventions may have acupuncture-specific efficacy (56, 57). Even in Madsen's review (58), a greater effect of acupuncture with penetrative placebo needles compared with the non-penetrative placebo needles was found (P = 0.04). Furthermore, the device or method of different placebo interventions is possible resource of heterogeneity. To the best of our knowledge, four different types of placebo acupuncture: i) placebo acupuncture device (with a blunt needle tip, pressing the skin without penetrating), including Streitberger's needle [(15, 16, 59); the needle tip is blunt but a pricking sensation is felt by the patient, simulating the puncturing of the skin] and Park needle [(60, 61); this device has a retractable needle shaft and a blunt tip. Acupuncturists were instructed to lightly place the sham needle on the surface of the skin with no manipulation of the needle to minimize any physiological effect]; ii)shallow acupuncture (39), which stimulates at the real acupoints (sham acupuncture was standardized as minimal-depth needling without stimulation “DeQi”); iii) non-acupoints (17) or non-therapeutic acupoints (42) utilization in the control group, and iv)Mock electrical stimulation [(14, 62); the technique has been shown to constitute a successful placebo treatment in functional brain networks (61). However, the difference of placebo efficacy among the above placebo intervention was not found in an accepted way (63). Thus, the potential heterogeneity may be also related to the placebo intervention.

For the age of participants, one of the major factors related to the decline in fertility with age is the aging uterus. However, the underlying factors that decrease endometrial receptivity in older women are still unclear.

Though different interventions in the treatment group (acupuncture was used alone or not) showed no significant heterogeneity, the previous meta-analysis showed the intervention in the treatment group accounts for some heterogeneity (64). Further analysis should be conducted. Furthermore, other potential variables, such as the way of measurement of three-dimensional ultrasound, may cause heterogeneity and care should be taken when the therapeutic efficacy of acupuncture treatment is evaluated.

To reduce the risk of bias, we only included the studies of randomized trials. We assessed the GRADE score to determine the strength of evidence. We excluded the RCTs with the outcome of EMT only for a more reliable result. To strengthen the efficacy of our findings, we explored the cause of heterogeneity associated with the clinical characteristics and methodological characteristics.

The review included moderate or low-quality studies. In our review, the duration of acupuncture was assessed. However, based on the study design of acupuncture trial guideline, Standards for Reporting Interventions in Controlled Trials of Acupuncture (STRICTA), and the frequency and starting time of acupuncture should be also considered in acupuncture dosage evaluation. Also, the variables subgroup analysis did not decrease the heterogeneity, which indicated that there were still potential factors in the collected studies. Therefore, a standardized individualized acupuncture program with a homogeneity trial design needed to be performed to evaluate the therapeutic efficacy of acupuncture, such as enough fixed acupuncture treatment dosage. The latest RCTs or negative unpublished trials were unlikely to be included in our analysis, so there was some publication bias.