Endometrial preparation for frozen–thawed embryo transfer cycles: a systematic review and network meta-analysis

Abstract

Purpose

To compare the effects of different endometrial preparation protocols for frozen–thawed embryo transfer (FET) cycles and present treatment hierarchy.

Methods

Systematic review with meta-analysis was performed by electronic searching of MEDLINE, the Cochrane Library, Embase, ClinicalTrials.gov and Google Scholar up to Dec 26, 2020. Randomised controlled trials (RCTs) or observational studies comparing 7 treatment options (natural cycle with or without human chorionic gonadotrophin trigger (mNC or tNC), artificial cycle with or without gonadotropin-releasing hormone agonist suppression (AC+GnRH or AC), aromatase inhibitor, clomiphene citrate, gonadotropin or follicle stimulating hormone) in FET cycles were included. Meta-analyses were performed within random effects models. Primary outcome was live birth presented as odds ratio (OR) with 95% confidence intervals (CIs).

Results

Twenty-six RCTs and 113 cohort studies were included in the meta-analyses. In a network meta-analysis, AC ranked last in effectiveness, with lower live birth rates when compared with other endometrial preparation protocols. In pairwise meta-analyses of observational studies, AC was associated with significant lower live birth rates compared with tNC (OR 0.81, 0.70 to 0.93) and mNC (OR 0.85, 0.77 to 0.93). Women who achieved pregnancy after AC were at an increased risk of pregnancy-induced hypertension (OR 1.82, 1.37 to 2.38), postpartum haemorrhage (OR 2.08, 1.61 to 2.78) and very preterm birth (OR 2.08, 1.45 to 2.94) compared with those after tNC.

Conclusion

Natural cycle treatment has a higher chance of live birth and lower risks of PIH, PPH and VPTB than AC for endometrial preparation in women receiving FET cycles.

Supplementary Information

The online version contains supplementary material available at 10.1007/s10815-021-02125-0.

Introduction

Frozen–thawed embryo transfer (FET) is a widely accepted procedure used for the storage and transfer of excess embryos produced in fresh in vitro fertilization (IVF) or intracytoplasmic sperm injection (ICSI) cycles. Increasing live birth rates resulting from improvements in technology, as well as increasing demand of preimplantation genetics testing and fertility preservation, has led to a progressive increase in the amount of FET cycles over the past decade [1]. In order to optimize pregnancy rates in FET, the development of embryo and endometrium should be synchronized; however, the best-individualized approach to prepare endometrium is still a matter of debate [2, 3].

Prior reviews had compared several endometrial preparation techniques in FET; a majority of the comparisons were in single, small trials; and there was no consistent superiority of one protocol over another [3–5]. These studies have focused on comparing two arms and provided limited insights into the treatment hierarchy, which makes it difficult for both women and healthcare to interpret current evidence regarding the plethora of possible comparisons. In this paper, we report the clinical effectiveness component, which aimed to bring together all prior randomised controlled trials (RCTs) and more recently published RCTs. These data are incorporated into a network meta-analysis comparing the clinical effectiveness of endometrial preparation protocols for FET.

Systematic reviews and meta-analyses of RCTs are considered the highest level of evidence for the evaluation of treatment effects. However, several reports have found little evidence of significant differences in effect estimates between RCTs and observational studies [6–9]. The addition of observational studies in meta-analyses increases sample size, which could enable the evaluation of small treatment effects and infrequent outcome measures and thus provide real-world evidences [10, 11]. In our work, we decided to perform meta-analyses of observational studies to obtain all the available evidence.

While most studies have focused on the rates of live birth or miscarriage, a few studies published in recent years have investigated the associations between the method of endometrial preparation and the obstetric and neonatal outcomes [12–15]. No systematic review or meta-analysis has been performed, so we included them in our secondary outcomes with the aim to investigate the effect of different regimens for endometrial preparation in FET on obstetric complications and perinatal outcomes.

Material and methods

For this systematic review and meta-analysis, we searched Ovid MEDLINE, the Cochrane Library, Embase, ClinicalTrials.gov and Google Scholar for RCTs and observational studies published from the date of database inception to June 27, 2020, and updated the search on Dec 26. No language or publication-type limits were applied. We used MeSH headings “cryopreservation”, “gonadotrophin-releasing hormone (GnRH)”, “ovulation induction”, “menotropins”, “follicle stimulating hormone (FSH)”, “estrogens” and “progesterone” and combined them with text words and word variants. The reference lists of selected articles and reviews were hand searched to identify any relevant articles. Detailed search strategy can be found in Supplemental Appendix 1.

We followed the PRISMA guidelines for network meta-analysis [16]. The study protocol was registered with PROSPERO (CRD42020193531). We selected studies for inclusion in two stages, duplicate or irrelevant articles were excluded by screening of titles and abstracts and all remaining articles were screened in full text. Two authors independently did the screening and assessed them for eligibility with discrepancies resolved with an additional reviewer.

We included RCTs and observational studies comparing at least two of the following endometrial preparation protocols in FET cycles: true natural cycle (tNC), modified natural cycle (mNC), artificial cycle without suppression (AC), AC suppressed with GnRH agonists (AC+GnRH), ovarian stimulation with aromatase inhibitor (AI), ovarian stimulation with clomiphene citrate (CC), ovarian stimulation with gonadotropin or FSH (Gn/FSH). True natural cycle was defined as spontaneous ovulatory cycle without the use of any medication before luteal phase support (LPS). Modified natural cycle was defined as ovulation triggered using a human chorionic gonadotropin (hCG) injection. Regarding tNC and mNC, studies that did or did not employ LPS were both included. AC was defined as an artificial, hormonally controlled cycle using sequentially administered exogenous oestrogen and progesterone. AC+GnRH was defined as pituitary suppression by administration of GnRH agonist before AC. Case-series, case-control studies and conference abstracts were excluded. Studies involving donor recipient cycles or complex endometrial preparation protocols were also excluded.

Two independent reviewers undertook study quality assessment and data extraction. Any discrepancies were resolved by discussion within the review team. RCTs were assessed according to the Cochrane Handbook for Systematic Reviews of Interventions [17]. The review authors categorized studies as “low risk”, “high risk” or “unclear risk” of bias. The quality of observational studies was assessed using the Newcastle-Ottawa Scale for evaluating the quality of nonrandomized studies [18]. The scale evaluates study bias and assigns points in the following 3 domains: appropriate selection of participants, appropriate measures of exposure and outcome variables and appropriate control of confounding; study quality was judged to be high if the Newcastle-Ottawa Scale score was at least 7 points (maximum score, 9 points).

We considered live birth for our primary analyses. Live birth was defined as the delivery of one or more living infants. Our secondary outcomes were ongoing pregnancy (defined as a pregnancy beyond 12 weeks gestation), clinical pregnancy (defined as the presence of a gestational sac, with or without a fetal heartbeat on ultrasonography) and miscarriage (defined as pregnancy loss after confirmation of clinical pregnancy). The interested pregnancy-related complications involved in this meta-analysis were pregnancy-induced hypertension (PIH), gestational diabetes mellitus (GDM), preterm premature rupture of the membranes (PPROM), placenta previa (PP), placental abruption (PA) and postpartum haemorrhage (PPH). The perinatal outcomes included were low birth weight (LBW; < 2500 g), very low birth weight (VLBW; < 1500 g), preterm birth (PTB; < 37 weeks of gestation), very preterm birth (VPTB; < 32 weeks of gestation), small for gestational age (SGA; birth weight < 10th percentile), large for gestational age (LGA; birth weight > 90th percentile) and perinatal mortality (stillbirth and early or late neonatal death). In case of certain discrepancy in the definition, we accepted the primary study authors’ definition when relevant.

Categoric data of relevant outcomes were extracted from the study and collated into 2 × 2 tables. An odds ratio (OR) with 95% confidence interval (CI) was calculated based on the sample number and number of events in each treatment group within each study. For pregnancy outcomes, the number of FET cycles was regarded as the denominator. For obstetric and neonatal outcomes, the number of pregnancy or live birth was regarded as the denominator.

Concerning various preparation protocols and inclusion criteria across the studies, we supposed there would be some statistical heterogeneity; therefore, we did meta-analyses of all the outcomes with the random-effects model (Mantel-Haenszel method). We assessed statistical heterogeneity in each comparison with the I² statistic and p value [19]. Publication bias would be assessed by means of Begg test if more than 10 studies were included [20]. For network meta-analysis, we used a continuity correction for studies with no events by adding 0.5 to both the event count and the total sample size. To visualise network geometry and node connectivity, we produced network plots for each outcome [21]. Afterward, we performed network meta-analysis using the methodology of the multivariate meta-analysis model. We prepared league tables presenting mixed comparisons for the inspection of both types of evidence [22]. We estimated the ranking probabilities for all treatments of being at each possible rank for each intervention and the treatment hierarchy was summarized as surface under the cumulative ranking curve [23].

The assessment of statistical heterogeneity in the networks was based on the magnitude of the heterogeneity variance parameter (τ²) estimated from the network meta-analysis models. To check the assumption of consistency in the entire network, we used the design-by-treatment model and judged the presence of inconsistency based on a Chi² test [24]. Local inconsistency between direct and indirect sources of evidence was statistically assessed by calculation of the difference between direct and indirect estimates in all closed loops in the network [21]. To evaluate the presence of small study effects, we visually inspected comparison-adjusted funnel plots for each outcome [21]. We prepared to analyse primary outcome in the following subgroups if enough studies were available: ovulatory patients vs. anovulatory patients. To assess the robustness of our findings from observational studies, we performed three predefined sensitivity analyses: including only prospective studies, including only low risk of bias studies and including only studies published from 2010. The certainty of evidence produced by the synthesis for each outcome was evaluated using the GRADE approach and the framework described by Salanti and colleagues [25, 26]. Statistical analysis was performed with STATA (version 14.0).

Results

Overall, 3505 citations were identified by the search and 178 potentially eligible articles were retrieved in full text. We excluded 39 articles due to various reasons shown in Fig. 1, resulting in 26 RCTs and 113 prospective or retrospective cohort studies published between 1990 and 2020 (full references for all trials are given in Supplemental Appendix 2). The RCT by Li et al. had a tNC group in the contemporaneous period; it was also regarded as a prospective cohort study and included in the meta-analysis of observational studies.

Fig. 1.

Article retrieval and screening

The characteristics of the 139 studies, involving 6372 FET cycles in RCTs, 105,239 FET cycles and 21,644 pregnancy or live birth in observational studies are summarized in Supplemental Table 1. Seventy-eight studies were published in the past 5 years. There were 50 studies (36.0%) conducted in China, 38 (27.3%) in Europe, 10 trials each in the USA or Iran and the remaining studies recruited participants from other countries. Out of the 26 RCTs, 21 trials were classified as low risk for random sequence generation. Most studies lacked the blinding of participants, personnel, which was mainly inherent to the nature of the study interventions (Supplemental Figures 1 and 2). Twelve prospective and 102 retrospective cohort studies were evaluated; 106 of them had NOS scores > 6, and were considered to be high quality. Seventy-two studies were found to match maternal age and 41 studies were found to match at least 4 of the potential confounding factors, including maternal age, body mass index, previous pregnancy, infertility cause, infertility duration, endometrial thickness and no. and stage of embryo transferred (Supplementary Table 1).

Figure 2 shows the network of eligible comparisons of endometrial preparation protocols in RCTs for pregnancy outcomes. The results of the network meta-analyses for live birth are presented in Fig. 3. In the network meta-analysis, AC ranked the last in effectiveness with significant lower live birth rates compared with AI and Gn/FSH. AI was found to be more efficacious than AC in the meta-analyses with low or very low evidence in terms of ongoing pregnancy and clinical pregnancy (Figs. 4 and 5). In terms of miscarriage, there was no evidence of any significant difference in comparisons of different endometrial preparation protocols (Supplemental Figure 3). The test of global inconsistency showed no significant difference between the consistency and inconsistency models for pregnancy outcomes (p = 0.37 to 0.98). We also found no evidence for local inconsistency in all closed loops (Supplemental Figure 4). Small study effects were found to be suspicious in terms of ongoing pregnancy and miscarriage in the network meta-analysis (Supplemental Figure 5). Sensitivity analysis and subgroup analysis were not available due to the limited number of RCTs.

Fig. 2.

Network plots of network meta-analyses of randomised controlled trials: a live births, b ongoing pregnancy, c clinical pregnancy, d miscarriage. The size of the nodes corresponds to the number of trials evaluating the comparison. The thickness of the lines corresponds to the number of trials evaluating the comparison. The colours of nodes refer to the risk of bias: low (green), moderate (yellow) and high (red). tNC, true natural cycle; mNC, modified natural cycle; AC, artificial cycle without suppression; AC+GnRH, artificial cycle with gonadotropin-releasing hormone cycle; Gn/FSH, ovarian stimulation with gonadotropin or follicle stimulating hormone; AI, aromatase inhibitor; CC, clomiphene citrate

Fig. 3.

Live birth league table. Results of the network meta-analysis are presented in the left lower half and results from pairwise meta-analysis in the upper right half, if available. Comparisons between treatments should be read from left to right and the estimate is in the cell in common between the column-defining treatment and the row-defining treatment. Effect sizes represent summary odds ratios and 95% confidence intervals. In the left lower half, values greater than 1 favour the column-defining treatment and in the upper right half, those values greater than 1 favour the row-defining treatment. Cells in bold print indicate significant results. The evidence is graded using CINeMA system approach for network meta-analysis. Colours in the cells indicate the confidence in the evidence: grey, low; pink, very low. tNC, true natural cycle; mNC, modified natural cycle; AC, artificial cycle without suppression; AC+GnRH, artificial cycle with gonadotropin-releasing hormone cycle; Gn/FSH, ovarian stimulation with gonadotropin or follicle stimulating hormone; AI, aromatase inhibitor; CC, clomiphene citrate; NA, not applicable

Fig. 4.

Ongoing pregnancy league table. Results of the network meta-analysis are presented in the left lower half and results from pairwise meta-analysis in the upper right half, if available. Comparisons between treatments should be read from left to right and the estimate is in the cell in common between the column-defining treatment and the row-defining treatment. Effect sizes represent summary odds ratios and 95% confidence intervals. In the left lower half, values greater than 1 favour the column-defining treatment and in the upper right half, those values greater than 1 favour the row-defining treatment. Cells in bold print indicate significant results. The evidence is graded using CINeMA system approach for network meta-analysis. Colours in the cells indicate the confidence in the evidence: grey, low; pink, very low. tNC, true natural cycle; mNC, modified natural cycle; AC, artificial cycle without suppression; AC+GnRH, artificial cycle with gonadotropin-releasing hormone cycle; Gn/FSH, ovarian stimulation with gonadotropin or follicle stimulating hormone; AI, aromatase inhibitor; CC, clomiphene citrate; NA, not applicable

Fig. 5.

Clinical pregnancy league table. Results of the network meta-analysis are presented in the left lower half and results from pairwise meta-analysis in the upper right half, if available. Comparisons between treatments should be read from left to right and the estimate is in the cell in common between the column-defining treatment and the row-defining treatment. Effect sizes represent summary odds ratios and 95% confidence intervals. In the left lower half, values greater than 1 favour the column-defining treatment and in the upper right half, those values greater than 1 favour the row-defining treatment. Cells in bold print indicate significant results. The evidence is graded using CINeMA system approach for network meta-analysis. Colours in the cells indicate the confidence in the evidence: yellow, moderate; grey, low; pink, very low. tNC, true natural cycle; mNC, modified natural cycle; AC, artificial cycle without suppression; AC+GnRH, artificial cycle with gonadotropin-releasing hormone cycle; Gn/FSH, ovarian stimulation with gonadotropin or follicle stimulating hormone; AI, aromatase inhibitor; CC, clomiphene citrate; NA, not applicable

The results of cohort studies for pregnancy outcomes are presented in Fig. 6, Fig. 7 and Supplemental Figures 6–8. AC was found to be less efficacious than tNC (OR 0.81, 95% CI 0.70 to 0.93, I² = 75.6%) and mNC (OR 0.85, 95% CI 0.77 to 0.93, I² = 33.5%) in terms of live birth, but not in terms of clinical pregnancy. AC+GnRH was found to be more efficacious than AC in terms of live birth and clinical pregnancy. Patients who received AC were found to be at an increased risk of miscarriage than those after tNC and mNC. Subgroup analysis was not available because few studies recruited only anovulatory patients; therefore, we performed a sensitivity analysis including ovulatory patients instead. Since most of studies had high quality, we included those with NOS scores of 8 or 9 points for sensitivity analysis. Except the sensitivity analysis including limited prospective studies, other sensitivity analyses did not affect the main results (Fig. 8).

Fig. 6.

Endometrial preparation by NC compared with AC on live birth in cohort studies. tNC, true natural cycle; mNC, modified natural cycle; AC, artificial cycle without suppression

Fig. 7.

The effects of different endometrial preparation protocols on live birth in cohort studies. tNC, true natural cycle; mNC, modified natural cycle; AC, artificial cycle without suppression; AC+GnRH, artificial cycle with gonadotropin-releasing hormone cycle; Gn/FSH, ovarian stimulation with gonadotropin or follicle stimulating hormone; AI, aromatase inhibitor; CC, clomiphene citrate

Fig. 8.

Sensitivity analyses of live birth in cohort studies. tNC, true natural cycle; mNC, modified natural cycle; AC, artificial cycle without suppression; AC+GnRH, artificial cycle with gonadotropin-releasing hormone cycle; Gn/FSH, ovarian stimulation with gonadotropin or follicle stimulating hormone; AI, aromatase inhibitor; CI, confidence interval; NA, not applicable. *Only studies with NOS scores of 8 or 9 points were included

The results of the meta-analyses for obstetric and neonatal outcomes are presented in Figs. 9, 10 and Supplementary Figure 9. AC was chosen as the reference group because it was used in most studies. Women who achieved pregnancy after tNC or mNC were associated with a lower risk of PIH compared with those after AC (OR 0.55, 95% CI 0.42 to 0.73, I² = 0% for tNC; OR 0.54, 95% CI 0.33 to 0.87, I² = 54.5% for mNC). Women who achieved pregnancy after tNC had a lower chance of PPH compared with those after AC (OR 0.48, 95% CI 0.36 to 0.62); however, it was only based on a single study. For GDM and PP, there was no evidence of any significant difference. The tNC group showed a lower VPTB rate than the AC group (OR 0.48, 95% CI 0.34 to 0.69, I² = 0%). A higher risk of LGA was found in the AC group compared with other groups. There was no significant difference in terms of other neonatal outcomes. Funnel plots did not indicate evidence for publication bias in terms of live birth (Supplementary Figure 10).

Fig. 9.

Endometrial preparation by tNC compared with AC on obstetric and neonatal outcomes. tNC, true natural cycle; AC, artificial cycle without suppression; PIH, pregnancy-induced hypertension; GDM, gestational diabetes mellitus; PP, placenta previa; PPH, postpartum haemorrhage; PTB, preterm birth; VPTB, very preterm birth; LBW, low birth weight; VLBW, very low birth weight; LGA, large for gestational age; SGA, small for gestational age; PM, perinatal mortality

Fig. 10.

Endometrial preparation by mNC compared with AC on obstetric and neonatal outcomes. mNC, modified natural cycle; AC, artificial cycle without suppression; PIH, pregnancy-induced hypertension; GDM, gestational diabetes mellitus; PP, placenta previa; PTB, preterm birth; VPTB, very preterm birth; LBW, low birth weight; LGA, large for gestational age; SGA, small for gestational age; PM, perinatal mortality

Discussion

This network meta-analysis represents the most comprehensive synthesis of data for currently available endometrial preparation protocols for FET cycles. Based on the current published literature, there is no consistent superiority of any endometrial preparation for FET. We found that AC appeared to yield at most equivalent or lower live birth rates when compared with other endometrial preparation protocols, accompanied with higher risk of miscarriage in AC found in observational studies. The pregnancy outcomes were found to be promising in women receiving ovarian stimulation with either Gn/FSH or AI; however, the results should be interpreted with caution due to the limited number of studies included. Pregnancies after AC were associated with increased risks of PIH, PPH, VPTB and LGA compared with those after tNC. The overall quality of all the evidence was low or very low.

Though the absence of any medical intervention is an advantage of tNC, this protocol entails frequent visits to the clinic for endocrine and ultrasound monitoring and a high risk of cycle cancellation of 6%, which can be overcome by mNC [27]. In a recent RCT by Mackens et al., mNC FET was associated with fewer visits for blood samplings compared with tNC FET [28]. In our meta-analysis, no statistically significant differences were observed between tNC and mNC at the level of all the pregnancy outcomes. These results are in line with previously published meta-analyses on this subject [3–5]; however, we included four RCTs and 12 observational studies, triple the number of prior included studies.

The necessity for LPS in tNC is still controversial. Various studies have been published with conflicting results. In one retrospective study, a lower clinical pregnancy rate was observed when LPS was administered in tNC FET, although this observation did not reach statistical significance (p =0.069) [29]. Bjuresten et al. reported a higher live birth rate when progesterone supplementation was initiated on the evening of the day of embryo transfer [30]. Others showed a similar result in patients receiving tNC FET [31]. Meta-analyses were performed with administration of LPS or not; comparable clinical pregnancy rates were demonstrated [3, 5]. Among the included studies in our work, various LPS or not was used in tNC, which might have brought in confounding bias.

Five RCTs have been performed comparing tNC or mNC versus AC; there appears to be no significant difference for live birth favouring any protocol in the meta-analyses; however, the number of studies was small and AC ranked the last in effectiveness. The most powerful multi-center RCT (ANTARCTICA trial) reported comparable live birth rates when mNC was compared with AC; however, more cycles were cancelled in AC and the dropout rate of the study was higher than 10%. Concerns were raised due to the overall low success rate reported and the high miscarriage rates [32]. The results found from observational studies partly support the results of network meta-analysis. Some of the studies compared tNC or mNC in ovulatory women with AC in patients with irregular cycles. Patients with irregular cycles were probably older with poorer ovarian function and more patients with polycystic ovary syndrome were submitted to AC, which could influence pregnancy rates. However, the clinical pregnancy rates were comparable between the groups; the sensitivity analyses including studies with either low risk of bias or ovulatory patients did not affect the main results. Furthermore, the odds ratio of live birth for tNC or mNC versus AC remained around 0.8, which was comparable with that from RCTs. When comparing tNC or mNC with AC+GnRH, no significant difference in live birth rate was found in both RCTs and observational studies. Of note, AC or AC+GnRH appeared to be associated with a higher risk of miscarriage when compared with mNC in observational studies. The retrospective nature of these studies may have introduced selection or information bias, leading to the discrepancy of results between RCTs and observational studies.

The largest number of participants that could be combined in meta-analysis of RCTs related to the comparison of AC versus AC+GnRH (11 RCTs); no significant differences were found in terms of clinical pregnancy rate and live birth rate. The results were in line with a previously published meta-analysis [4]. In observational studies, we found AC+GnRH was more efficacious than AC in terms of live birth and clinical pregnancy; this discrepancy was perhaps due to the selection bias in retrospective studies.

In network meta-analysis, ovarian stimulation with Gn/FSH and AI were ranked top two for efficacy, with a significant higher live birth rate than AC; the clinical interpretation of this finding is limited by the small number of trials in each node. The results from RCTs could not be duplicated in the meta-analysis of observational studies. Further powerful RCTs are warranted to make definitive conclusions.

To our knowledge, our work is the first systematic review and meta-analysis to compare the pregnancy-related complications and perinatal outcomes resulting from different endometrial preparation protocols in FET cycles. Compared to tNC and mNC, AC was associated with a higher risk of PIH and PPH. Our results are in accordance with a recent study suggesting an increased rate of PIH and PPH in programmed FET cycles [33]. The finding of a higher rate of PIH in our study is also supported by a recent study, in which they found a higher rate of preeclampsia in programmed FET cycles compared to other FET protocols, and therefore may partly explain a higher risk of VPTB in AC in our findings [34]. We found newborns conceived after AC FET were more likely to be LGA than those born after tNC, mNC or Gn/FSH FET, findings in line with a large-scale retrospective study including 9267 live-born singletons conceived after FET [35]. The mechanism by which AC FET affects birthweight remains to be elucidated; the hypothesis is that the early maternal endocrine milieu resulting from exogenous oestrogen and progesterone may have already set the stage for fetal growth in later pregnancy.

The beneficial impact of nature cycles on pregnancy outcomes could be mediated through improved maternal circulatory function. A higher risk of thromboembolic events leading by exogenous hormones could damage placentation, which may in turn cause miscarriage [36, 37]. Earlier closure of the implantation window at high estradiol levels might also be responsible for decreased pregnancy rates in AC [38, 39]. It is believed that the increased risk of obstetric complications may be, to some extent, attributable to the absence of the corpus luteum. This hypothesis is biologically plausible because the corpus luteum produces not only oestrogen and progesterone, but also vasoactive products such as relaxin, vascular endothelial growth factor and angiogenic metabolites of oestrogen [40, 41]; these products are hypothesized to be important for initial placentation, and abnormal early placentation has been proposed to be a critical step in the development of preeclampsia [42–45]. Apart from this, we assume that the intake of exogenous oestrogens during the period of trophoblastic vessel invasion may also be involved. Imudia et al. found that an elevated serum oestrogen concentration during controlled ovarian hyperstimulation was associated with higher risk of maternal preeclampsia and SGA newborns for fresh embryo transfer [46].

Our study has unique strengths. First, this review is an up-to-date evidence study and the first study to investigate the association of different endometrial preparation protocols and obstetric complications and perinatal outcomes comprehensively. Second, we included 139 studies, enabling us to perform meta-analyses separately for RCTs and observational studies, which therefore avoided introducing selection bias. Some results were hardly changed in sensitivity analyses. Finally, we included studies from all over the world; most of the studies were published in the last 5 years and most of the cohort studies were of high quality, which makes our results more representative and credible.

There are several important limitations to this study. First, because this systematic review included observational studies and separate analysis of RCTs was not performed, the data used to generate the summary ORs might have been affected by confounding bias. In the GRADE framework, most comparisons were assessed as low or very low quality, which largely restricts the interpretation of these results. Second, heterogeneity exists in some comparisons, which might be a consequence of a cohort effect that related to different eligibility criteria. In the studies, we used random-effects model to combine the data and the results were consistent in sensitivity analyses. Third, drug dosage and timing, freezing method, transfer policy and LPS varied across the studies, which might have influenced our results. Fourth, some outcomes, such as endometrial thickness, cycle cancellation and adverse events, were not compared in our analysis, and cost-effectiveness analysis was not performed in our work, which is a big concern for clinical practice. Obstetric complications and perinatal outcomes were only reported in cohort studies, with calls to focus on these outcomes in future RCTs. Most of the results were based on some large retrospective cohort studies, which should be interpreted with caution. Finally, publication bias is always a concern. We searched ClinicalTrials.gov and included non-English language trials; funnel plots did not indicate evidence for publication bias.

Conclusions

Natural cycle treatment has a higher chance of live birth and lower risks of PIH, PPH and VPTB than AC for endometrial preparation in women receiving FET cycles. Ovarian stimulation with Gn/FSH or AI may be promising, but the evidence is scare and needs to be evaluated in future studies. Except for pregnancy rates, patient convenience and cost efficiency, pregnancy-related complications and perinatal outcomes need to be evaluated in well-conducted RCTs.

Author contribution

S.Y.Z. conceived and designed the study. H.L.W. selected articles for inclusion, extracted data, performed the statistical analyses and drafted the first version of the manuscript. P. Z. conducted literature searches, and selected articles for inclusion. S.S.W. and X.N.L. extracted data and assessed study quality. All the authors contributed to the revision of the first draft of the manuscript, critically checked its content and approved its final version.

Funding

This work was supported by the National Natural Science Foundation of China grant (81671435 to S. Z).

Declarations

Conflict of interest

The authors declare no competing interests.