Efficacy of therapies and interventions for repeated embryo implantation failure: a systematic review and meta-analysis

The aim of the present systematic review and meta-analysis was to assess the effect of the different therapeutic options for repeated embryo implantation failure (RIF) on a subsequent IVF cycle outcome. Twenty-two RCTs and nineteen observational studies were included. Pooling of results showed a beneficial effect of intrauterine PBMC infusion on both CPR (RR 2.18; 95% CI 1.58–3.00; p < 0.00001; OR 2.03; 95% CI 1.22–3.36; p = 0.006) and LBR (RR 2.41; 95% CI 1.40–4.16; p = 0.002; OR 3.73; 95% CI 1.13–12.29; p = 0.03), of subcutaneous G-CSF administration on CPR (RR 2.29; 95% CI 1.58–3.31; p < 0.0001) and of intrauterine PRP infusion on CPR (RR 2.45; 95% CI 1.55–3.86; p = 0.0001). Observational studies also demonstrated a positive effect of IVIG and intrauterine hCG infusion on both CPR and LBR and of atosiban on CPR. Studies investigating intrauterine G-CSF infusion, LMWH, intravenous intralipid, hysteroscopy, blastocyst-stage ET, ZIFT, PGT-A and AH failed to observe an impact on IVF outcome. The quality of the evidence that emerged from RCTs focused on intrauterine PBMC infusion and subcutaneous G-CSF administration was moderate. For all other therapies/interventions it varied from low to very low. In conclusion, intrauterine PBMC infusion and subcutaneous G-CSF administration are the most promising therapeutic options for RIF. However, further well conducted RCTs are necessary before their introduction into clinical practice.

www.nature.com/scientificreports/ three fresh or frozen cycles under 40 years of age 6 . However, the definition of good quality embryos is subjective and the authors often do not refer to shared classification criteria. Most of the previous meta-analyzes aimed at determining the efficacy of single therapeutic intervention for RIF included patients with at least two previous failed ET attempts. However, by applying these criteria, the rate of false positive RIF diagnosis is estimated to be considerable [at least 46%] 7 and, as a consequence, the studied population probably included a significant proportion of patients without a real obstacle to conception but who had not yet succeeded just because of statistical misfortune. Evidence about efficacy of therapeutic interventions deriving from meta-analyzes conducted with these assumptions cannot therefore be considered completely reliable.
In the present systematic review and meta-analysis, we defined RIF as the failure to obtain a clinical pregnancy after at least three ET attempts. By using this threshold, the risk of false positive diagnosis is significantly lower 7 . Importantly, these diagnostic criteria also exclude elements of subjectivity and are therefore easily replicable in any clinical setting.
In most cases, the abovementioned therapeutic interventions are promising. However, clinicians can hardly orient themselves toward such a plethora of options with often unproven efficacy 2 .
Aim. Considering the methodological weaknesses of the previous contributions and the uncertainties about the preferred treatment strategies, we conducted the present systematic review and meta-analysis with the aim to assess the effect of the different therapies and interventions for RIF on the subsequent IVF cycle outcomes.

Materials and methods
This literature overview was reported according to the PRISMA guidelines for systematic reviews 34,35 and the meta-analysis was conducted according to the MOOSE guidelines 36 . Since published de-identified data were used, this study was exempt from institutional review board approval.
Sources and study selection. The present systematic review and meta-analysis was restricted to published research articles that investigated the effect of all proposed therapies and interventions for RIF on the subsequent IVF cycle outcomes. Primary outcomes were Live Birth Rate (LBR) per patient and Clinical Pregnancy Rate (CPR) per patient. "Live birth" was defined as the delivery of one or more living infants. "Clinical pregnancy" was defined as the presence of one or more intrauterine gestational sacs on transvaginal ultrasound or other definitive clinical signs 37 . Secondary outcomes were implantation rate (IR) per embryo, multiple pregnancy rate (MPR) per patient and miscarriage rate (MR) per patient. "Implantation rate" was defined as the number of gestational sacs on transvaginal ultrasound divided by the number of embryos transferred. "Multiple pregnancy" was defined as the presence of two or more intrauterine embryos on transvaginal ultrasound. "Miscarriage" was defined as fetal loss before 20 weeks' gestation 37 .
We systematically searched Pubmed, MEDLINE, Embase and Scopus, from database inception to May 13th, 2020. Searches were limited to studies in humans. A first search was conducted using the following terms: 'therapy' OR 'intervention' OR 'treatment' AND 'implantation failure' OR 'repeated implantation failure' OR 'recurrent implantation failure' OR 'RIF' . A second search was carried out by combining each therapy or intervention emerged from the first search (i.e. endometrial injury; hysteroscopy; endometrial sampling for histology and microbiological investigations and endometritis treatment; atosiban; copper intrauterine device placement; sequential embryo transfer; embryo transfer medium enriched with hyaluronic acid; autologous embryo-cumulus cells co-culture; intracytoplasmic morphologically selected sperm injection; blastocyct stage embryo transfer; zygote intrafallopian tube transfer; assisted hatching; preimplantation genetic testing for aneuploidies; intravenous immunoglobulin; intrauterine administration of peripheral blood mononuclear cell; tacrolimus; subcutaneous administration of granulocyte colony stimulating factor; intrauterine infusion of autologous platelet-rich plasma; intravenous intralipid infusion; human chorionic gonadotropin; low-molecular-weight heparin; aspirin; Data extraction and analysis. Three authors (A.B., E.S. and F.C.) independently evaluated all articles and extrapolated the data on standardized forms. A final abstraction form was compiled from the three evaluation forms, after resolution of all the discrepancies among reviewers through a discussion with the two remaining authors.
The year of publication, location, study design, study period, criteria used to define RIF, investigations performed to exclude possible known causes of RIF, investigated therapy or intervention for RIF, primary and secondary outcomes were recorded.
Study outcomes were expressed using risk ratio (RR) with 95% confidence interval (95% CI) for RCTs and odds ratio (OR) with 95% CI for observational studies.
Risk estimates greater than 1 indicate an increased risk of the defined outcome; risk estimates less than 1 indicate a decreased risk of the defined outcome. We assessed statistical significance using 95%CI: if the 95%CI did not include the neutral value 1, we considered the risk statistically significant 41,42 . The inconsistency of the studies' results was measured using Cochrane Q and the I 2 statistic 38 . Negative values of I 2 are set equal to 0 so that I 2 lies between 0 and 100%. According to the Cochrane Handbook for Systematic Reviews of Intervention, an I 2 value of 0 indicates no observed heterogeneity, whereas I 2 values from 30 to 60% may represent moderate heterogeneity, I 2 values from 50 to 90% may represent substantial heterogeneity, and I 2 values from 75 to 100% represent considerable heterogeneity 38 . If the I 2 values indicated moderate, substantial, or considerable heterogeneity, we conducted sensitivity analyses to verify whether any one of the included studies unduly influenced the pooled effect size.
The risk estimates were combined in a meta-analysis using a fixed effects model when the heterogeneity found among the studies was absent to moderate (0% ≤ I 2 < 30%). When heterogeneity was moderate, substantial, or considerable (I 2 ≥ 30%), we used the DerSimonian and Laird method 43,44 for a random-effects model 45 . Funnel plots, which graph RR/OR on a log scale (effect) against standard error of log-RR/OR (precision), were generated and visually inspected for asymmetry to determine whether the included studies were non representative of the body of possible studies on the subject (as could result from a small-study effect or other biases, such as publication and poor-quality bias). The approach by Egger et al. was used to test the significance of funnel plot asymmetry 45 . All analyses were performed using Review Manager version 5.3 (Nordic Cochrane Centre, Cochrane Collaboration). Figure 1 summarizes the process of literature identification and selection of studies. Our literature searches yielded 746 studies, from which 22 duplicates were removed. After a review of the titles and abstracts, 154 studies were identified as potentially eligible for inclusion. After a full review, we excluded 19 systematic reviews or meta-analysis 2,5,22,23,37,46-59 , 8 case reports 60-67 , 4 letters to the editor [68][69][70][71] and 81 original studies [references and reasons for exclusion are reported in Table 1]. Data on the efficacy of therapies and interventions for RIF were extracted from the remaining 42 articles 8,12,13,18,20,21,24,27,28,31,75, . Included studies investigated uterine interventions, laboratory and procedural technologies and interventions and immunomodulatory therapies. Details of the characteristics of the selected studies are shown in Table 2. Seven of the included studies were case-control studies, 12 were prospective cohort studies and 22 were RCTs. Therapies and interventions that could be pooled included subcutaneous or intrauterine G-CSF administration, sequential ET, intravenous intralipid infusion, endometrial injury, subcutaneous LMWH, hysteroscopy, PGT-A, atosiban, IVIG administration, intrauterine hCG injection, blastocyst stage ET, ZIFT, intrauterine PBMC infusion, AH and intrauterine PRP infusion. www.nature.com/scientificreports/ Risk of bias and quality assessment results. Results obtained from the risk of bias assessment for RCTs and for observational studies are summarized in Fig. 2 and Table 3 respectively. The quality of the evidence for each single therapy/intervention is described in the 'Synthesis of results' section and summarized in Table 4.

Synthesis of results. Uterine interventions.
Intentional Endometrial injury. Three RCTs 8,146,164 and two observational studies 152,159 evaluated the impact of an intentional injury to the endometrium during the spontaneous menstrual cycles before IVF on the outcomes of the IVF cycle. Primary outcomes Meta-analysis of RCTs did not show significantly increased chances of pregnancy and live birth in women who underwent intentional endometrial injury (random effects model, RR 1.43; 95% CI 0.79-2.61; p = 0.24; I 2 = 52% and random effects model, 1.55; 95% CI 0.81-2.94; p = 0.18; I 2 = 46%, respectively) 8,146,164 (Fig. 4). On the contrary, pooling of results from observational studies showed a beneficial effect of endometrial injury on pregnancy rate (fixed effects model, OR 3.03; 95% CI 1.48-6.18; p = 0.002; I 2 = 0%) 152 146 .
Quality of the evidence We downgraded the quality of the evidence provided by RCTs by one level for risk of bias and, considering the low number of events, by one level for imprecision. The quality of the evidence provided by observational studies was downgraded by one level for risk of bias and, considering the wide confidence interval, by one level for imprecision and upgraded by one level for the large magnitude of the effect (Table 4).
Hysteroscopy. One RCT investigated whether outpatient hysteroscopy in the month before starting IVF treatment cycle could improve the outcome in women with RIF 143 .
Quality of the evidence The data reported in the present meta-analysis were extrapolated from a sub-analysis carried out by El-Thouky et al. 143 . Furthermore, the number of events is low. Hence, we downgraded the quality of the evidence by one level for imprecision ( www.nature.com/scientificreports/ Atosiban. One observational study 12 examined the effect of atosiban administered before transfer of frozenthawed embryo to women with RIF. Primary outcomes Authors observed an increased CPR in treated women when compared to controls (OR 2.63; 95% CI 1.08-6.40; p = 0.03) 12 (Fig. 4).
Quality of the evidence The quality of the evidence provided by He et al. was downgraded by one level for risk of bias (Table 4).
Laboratory and procedural technologies and interventions. Sequential ET. One RCT 13 and two observational studies 75,144 compared sequential ET (cleavage stage ET followed by blastocyst ET) vs blastocyst stage ET in women with RIF.

References Therapy/intervention Reason for exclusion
Madhavan et al. 113 Intrauterine Quality of the evidence We downgraded the quality of the evidence provided by Shahrokh Tehraninejad et al. by one level for risk of bias and, considering the low number of events, by one level for imprecision. The quality of the evidence provided by observational studies was downgraded by one level for risk of bias (Table 4).
PGT-A. Two RCTs 20,141 and three observational studies 145,160,167 investigated the potential role of PGT-A in improving IVF outcomes in women with RIF.
Quality of the evidence The evidence emerged from RCTs was downgraded by one level for risk of bias and, considering the low number of events, by one level for imprecision. For CPR, we downgraded the quality of the evidence provided by observational studies by one level for risk of bias. For LBR, we did not downgrade the quality of the evidence (Table 4).
Blastocyst-stage ET. One RCT compared blastocyst-stage ET outcomes with day 2-3 ET outcomes in women who failed to conceive after three or more day 2-3 IVF/ET cycles 149 .
Quality of the evidence The quality of the evidence was downgraded by one level for risk of bias and, considering the low number of events, by one level for imprecision (Table 4).
ZIFT. Three observational studies investigated the possible beneficial effect of ZIFT in women with RIF 18,150,163 .  www.nature.com/scientificreports/  163 .
Quality of the evidence The quality of the evidence was downgraded by one level for risk of bias (Table 4).
AH. One RCT 153 and one observational study 158 investigated the effect of AH on IVF outcomes in women with RIF. Primary outcomes 156 did not observe an increased chance of clinical pregnancy in women who underwent AH (RR 0.78; 95% CI 0.48-1.27; p = 0.31) 153 (Fig. 4). Primi (Fig. 4).
Quality of the evidence The quality of the evidence provided by Rufas-Sapir et al. was downgraded by one level for risk of bias and, considering the low number of events, by one level for imprecision 2 . We downgraded the quality of the evidence emerged from the study conducted by Primi et al., by one level for risk of bias (Table 4).
Immunomodulatory therapies. G-CSF administration. Six RCTs evaluated the possible beneficial effect of the subcutaneous or intrauterine G-CSF administration 140,142,147,148,161,162 .
Subgroup analysis Subcutaneous and intrauterine route of administration were analyzed separately (Fig. 3). Subcutaneous G-CSF administration resulted associated with an increased chance of clinical pregnancy (fixed effects model, RR 2.29; 95% CI 1.58-3.31; p < 0.0001; I 2 = 0%) when compared with no treatment 140,148,161,162 (Fig. 3). On the contrary, intrauterine administration had no impact on CPR (fixed effects model, RR 1.53; 95% CI 1.00-2.33; p = 0.05; I 2 = 0%) 142,147 (Fig. 3). Aleyasin et al. who investigated the subcutaneous route of administration observed a positive effect on embryo implantation chances (RR 2.94; 95% CI 1.24-5.01; p = 0.01) 140 .  Table 4. Summary of findings and certainty of the evidence. GRADE Working Group grades of evidence. High quality: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate quality: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low quality: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect. Very low quality: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect. RCT randomized clinical trial, G-CSF granulocyte-colony stimulating factor, LMWH low molecular weight heparin, PBMC peripheral blood mononuclear cells, AH assisted hatching, PGT-A preimplantation genetic testing for aneuploidy, IVIG intravenous immunoglobulin, PRP platelet rich plasma, hCG human chorionic gonadotropin, ZIFT zygote intrafallopian transfer, IR implantation rate, CPR clinical pregnancy rate, MR miscarriage rate, MPR multiple pregnancy rate, LBR live birth rate, 95% CI 95% confidence interval, RR risk ratio, OR odds ratio. www.nature.com/scientificreports/ www.nature.com/scientificreports/ In contrast, Davari-tanha et al. who focused on intrauterine G-CSF injection did not observe any impact on IR (RR 2.28; 95% CI 0.90-5.74; p = 0.08) 142 . Quality of the evidence In the majority of RCTs, the description of allocation concealment was unclear or the treatment providers were not blinded, hence we downgraded the quality of the evidence by one level for risk of bias for all outcomes. Considering the low total number of events, we also downgraded the quality of the evidence by one level for imprecision for all outcomes. For CPR evaluated in studies focused on subcutaneous G-CSF administration, we upgraded the quality of evidence by one level for the large magnitude of the effect (Table 4).

Intravenous intralipid infusion. One RCT investigated the effect of the intravenous infusion of intralipid 27 .
Primary outcomes Authors failed to show a benefit of the intravenous intralipid infusion on both the clinical pregnancy rate and the live birth rate (RR 1.30; 95% CI 0.80-2.10; p = 0.29 and 1.30; 95% CI 0.61-2.77, respectively) (Fig. 4).
Quality of the evidence Quality of the evidence was downgraded by one level for risk of bias and by one level for imprecision (Table 4).  (Fig. 4).
Quality of the evidence The quality of the evidence provided by RCTs was downgraded by two levels for risk of bias and by one level for imprecision. We also downgraded the level of the evidence provided by Berker et al. by one level for risk of bias (Table 4).
IVIG. One observational study 21 evaluated the efficacy of IVIG in women with RIF.
Quality of the evidence The quality of the evidence was downgraded by one level for risk of bias (Table 4).
Quality of the evidence The quality of the evidence was downgraded by one level for risk of bias (Table 4).
Quality of the evidence The quality of the evidence provided by RCTs was downgraded by one level for risk of bias, by one level for imprecision and upgraded by one level for the large magnitude of the effect ( Table 4). The quality of the evidence provided by observational studies was downgraded by one level for risk of bias and by one level for imprecision and upgraded by one level for the large magnitude of the effect (Table 4).
Intrauterine PRP infusion. Two RCTs 155,170 investigated whether administration of intrauterine PRP could improve IVF outcomes in women with RIF.
Primary outcomes Pooling of results showed a significantly increased chance of clinical pregnancy in treated women (fixed effects model, RR 2.45; 95% CI 1.55-3.86; p = 0.0001; I 2 = 0%) 155,170 (Fig. 4).  www.nature.com/scientificreports/ Quality of the evidence The quality of the evidence was downgraded by one level for risk of bias and, considering the low number of events, by one level for imprecision (Table 4).

Discussion
In the present study, meta-analysis of RCTs showed a beneficial effect of PBMC intrauterine infusion on both LBR and CPR and of subcutaneous G-CSF administration and intrauterine PRP infusion on CPR in women with RIF. Pooling of results of observational studies also demonstrated a positive effect of IVIG and hCG intrauterine infusion on both CPR and LBR and of atosiban administration on CPR. Meta-analysis of studies investigating the possible impact of intrauterine G-CSF infusion, LMWH, hysteroscopy, blastocyst-stage ET, ZIFT, PGT-A and AH failed to observe an impact on IVF outcome. Results about the effects of sequential ET and intentional endometrial injury are conflicting. The quality of the evidence that emerged from RCTs investigating the effect of intrauterine PBMC infusion and subcutaneous G-CSF administration was moderate. For all other therapies/ interventions it varied from low to very low.
Among the therapies that have been proven to be potentially effective, the intrauterine infusion of PBMC is supported by the most convincing evidence. In fact, meta-analyses of RCTs and of observational studies agree in demonstrating the positive effect on both primary outcomes and the magnitude of calculated effect estimates is considerable. Pourmoghadam et al. in an interesting meta-analysis had already shown a beneficial effect in women with at least three IVF failures 171 . The subsequent publication of the study conducted by Nobijari et al. 156 , which was the first RCT to report the chances of live birth, further strengthened the evidence. Nevertheless, data on the impact on the LBR as well as on the safety profile of this therapy should still be considered scanty.
The administration of G-CSF also emerged as a promising treatment option in women with RIF. Our findings confirmed those recently published by Kamath et al. who showed that in women with two or more IVF failures, G-CSF administration may improve CPR versus placebo 47 . Interestingly, we observed that of the two possible routes of administration, the only potentially effective seems to be the systemic one. Importantly, the magnitude of the effect was considerable and, as a consequence, we upgraded the quality of the evidence to moderate. Unfortunately, no data about the rate of live birth can be extracted from included studies that investigated this route of administration, which may impair the convincingness of the analysis. Reasons for discrepancies between the effects of systemic and intrauterine administration have yet to be fully elucidated. One could speculate that when administered systemically, G-CSF has a positive effect on oocyte maturation and embryonic development, while in locally endometrial cavity applications oocytes and embryos are deprived of this positive support 147 .
Intrauterine hCG infusion constitutes an excellent candidate to be tested in women with RIF. In fact, by acting as the homologous isomer of LH, hCG shares a common receptor with LH, namely, LHCGR, and their combination can regulate both endometrium receptivity and embryo implantation 172 . Importantly, in a recent metaanalysis, Gao et al., showed that infertile women who received intrauterine hCG injection before ET exhibited significantly higher rates of implantation, ongoing pregnancy and live birth and a lower rate of miscarriage 172 . In the present meta-analysis, pooling of results of observational studies focusing on patients with RIF showed a beneficial effect on both CPR and LBR. Unfortunately, the quality of the evidence was very low. In particular, the different volumes of culture medium (1 ml and 0.2 ml) and doses of hCG (1000 UI and 500 UI) impair the clinical homogeneity between studies and significantly limit the reliability of our results 28,151 .
Hypothesizing a key role of the immune response in the pathogenesis of RIF, IVIG, intravenous intralipid injection and PRP intrauterine infusion have also been proposed as possible treatments. Initial results regarding the efficacy of IVIG and PRP intrauterine injection are encouraging. However, even in these cases, the very low quality of the evidence does not allow reliable conclusions.
The decrease of the frequency and amplitude of uterine contractions obtained through the administration of atosiban, has also been theorized as a method to enhance the probability of embryo implantation and pregnancy in women with RIF. Our results were obtained from the data extrapolated from a single observational study and are in line with those of a recent meta-analysis conducted by Huang et al., who, using less stringent inclusion criteria [i.e. two or more consecutive failed IVF-ET attempts in which at least 1 ± 2 high quality embryos were transferred in each cycle], demonstrated increased chances of implantation, clinical pregnancy and live birth in women with RIF treated with atosiban 28 . Well conducted RCTs focusing on women with RIF diagnosed according to the criteria proposed in the present study are warranted.
Inconclusive results and demonstrations of inefficacy that emerged from the present meta-analysis are of particular importance. Over the years, we witnessed the emergence of a number of RIF treatment options of simple execution but characterized by weak rational bases. Nonetheless, their introduction into current clinical practice occurred rapidly without waiting for adequate evidence of efficacy and safety. Such conduct evidently conflicts with the principle of the traditional medical ethics summarized in the injunction "primum non nocere" and with the duty to protect patients, already psychologically frustrated, from false hopes and to avoid waste of resources.
In this perspective, the results about the effect of intentional endometrial injury deserve to be commented. The biological plausibility and relative ease of execution of this intervention attracted the attention of many clinicians around the world. Endometrial scratching is a safe procedure. However, it is somewhat painful. When performed in the luteal phase, patients reported pain scores between 3 and 7 of 10, and the procedure was discontinued due to pain in a number of cases 173 . Its efficacy in women with RIF is debated. Nonetheless, an online survey distributed to 189 fertility clinics across Australia, New Zealand and the UK found that 92% of clinicians recommend endometrial scratching to women with RIF 173 . In our study, meta-analysis of RCTs demonstrated the inefficacy of this intervention in increasing CPR and LBR. On the contrary, pooling of results of observational studies suggested a beneficial effect on CPR. These discrepancies combined with the relatively small sample size of the included studies and the statistical moderate/substantial heterogeneity do not allow conclusive interpretations. www.nature.com/scientificreports/ A recent RCT showed a potentially harmful effect of the endometrial biopsy performed in the follicular phase. In fact, authors reported a higher incidence of clinical miscarriages in the context of in-cycle scratching, which led to the study premature halt 174 . This considered, we conducted a sub-analysis on the basis of endometrial injury timing without however observing the superiority of one strategy over the others. Importantly, a recent retrospective study questioned the existence of RIF due to endometrial effect. In a cohort of 4229 women whose endometrium was sonographically normal and who underwent up to three frozen euploid single embryo-transfers, authors found a cumulative sustained implantation rate of 95.2%. As a result, RIF incidence was estimated < 5% 175 .
At present, there is no evidence to support the routine use of hysteroscopy as a screening and treatment tool in the population of women with RIF and a normal uterine cavity on ultrasound or hysterosalpingogram to improve the reproductive success rate. However, available data are scanty. Notably, there is compelling rationale that hysteroscopy might be effective in women with RIF. In fact, intrauterine pathology has been reported in as many as 50% of women with RIF leading to suggest that the correction of such pathology could improve IVF outcome 143 . Benefit could also be due to the negotiation of the cervical canal, thus, facilitating the subsequent embryo transfer 176 . Hysteroscopy has also the considerable advantage of allowing targeted endometrial biopsies. In this regard, a recent interesting meta-analysis showed that chronic endometritis therapy might be beneficial in patients suffering from RIF even if, according to the authors, the body of evidence on this topic is still insufficient to recommend routine chronic endometritis screening as intervention in such patients 37 . Future RCTs are thus welcomed in order to test such multiple hypothetical beneficial function of hysteroscopy in women with RIF.
Notably, we also failed to show a significant impact of LMWH administration on both CPR and LBR in nonthrombophilic women with RIF. However, the reliability of the results is limited by the very low quality of the evidence. Furthermore, the absence of data regarding the undesirable effects of LMWH administration [e.g. risk of bleeding] does not allow to grasp the whole picture.
Pooling of results of studies investigating the possible role of PGT-A did not show a positive effect on both clinical pregnancy and live birth chances per patient. Future research efforts should probably test this intervention on a population of older women in whom one may suspect with higher confidence that aneuploidy constitutes the cause of RIF. In this regard, it has however to be highlighted that PGT-A cannot be expected to increase the chance of live birth per patient 177 . It can at most only alleviate the burden of treatment to patients by reducing the number of transfers.
Finally, as for the sequential ET, the evidence is conflicting: pooling of results of observational studies showed a significantly increased CPR while the results of the only included RCT demonstrated no benefit. Safety of this intervention is questionable. The transfer of two embryos at a distance and the transfer of the second one at the blastocyst stage may increase the risk of dizygotic and monozygotic twinning respectively 41 . Published data about these possible complications are reassuring but still insufficient. The potential serious obstetric and neonatal consequences and the unconvincing results on the efficacy discourage the conduct of further studies. Moreover, data demonstrating no differences in CPR for the first 6 IVF cycles deserve careful study on the role of chance and even of different multiple factors influencing CPR and LBR 178 .
Other treatment hypotheses might be valid and some RCTs are ongoing in order to test them. In this context, of particular relevance is the study protocol published by Lu et al. 179 . Authors aim to determine if prednisone can enhance live birth in women with RIF undergoing IVF. Interestingly, studies have shown that prednisone could not only suppress the inflammatory response in pre-implantation endometrium, but also stimulate the secretion of hCG and promote proliferation and invasion of trophoblast 179 . The efficacy of ad hoc treatments in women with known diseases and RIF also deserves to be clarified. In this context, the benefits and risks of aspirin and/ or heparin in women with persistent antiphospholipid antibodies and RIF have been rather neglected until now.

Strengths and limitations.
To the best of our knowledge the present meta-analysis is the first to give a comprehensive view of the efficacy of all therapies or interventions proposed in order to improve IVF outcome in women with RIF. The population was selected according to strict inclusion criteria in order to reduce as much as possible the risk of misleading conclusions due to the high incidence of false positive diagnosis and, consequently, of inappropriate treatment. Moreover, being aware in advance of the limited available evidence, we decided to include also observational studies rather than limiting our analyses to RCTs. This choice allowed us also to also report on options that could become of interest in the future, i.e. once properly tested with RCTs.
Several limitations need to be considered in the interpretation of our results. First, many of the included studies suffered from serious risk of bias. Additionally, in the majority of cases, they recruited too few women to have enough statistical power to detect clinically relevant effect sizes, as is common in our field. Second, some studies included only frozen-thawed embryo replacement cycles while others only fresh IVF cycles. Furthermore, the protocols for ovarian stimulation, endometrial preparation, luteal phase support and the proposed interventions themselves also present marked variations between studies. In most cases, a proper investigation of this clinical heterogeneity was not feasible due to the limited number of studies. Third, in the present meta-analysis we focused on patients who had been investigated as much as possible to rule out possible known causes of RIF. However, it cannot be sustained with certainty that the selected population is affected by unexplained RIF. In fact, some contributions also included women of advanced age. In this context, it is pretty impossible to exclude the embryonic cause of RIF, without the use of PGT-A. Finally, there are few data addressing the safety profile of these treatments and their effect on the development and health of conceived children. Future studies focusing on treatment-related side effects and long-term follow-up data among the offspring are needed before introducing such interventions into daily clinical practice. www.nature.com/scientificreports/

Conclusion
In women with RIF, moderate quality evidence suggests that intrauterine PBMC infusion improves chances of clinical pregnancy and live birth and that subcutaneous G-CSF administration has a beneficial effect on CPR. These treatment options are the most promising among those investigated. However, prior to their introduction into routine clinical practice, high quality RCTs are needed. Trials design should include an identical placebo in the control arm to reduce performance bias and report ongoing pregnancy or live birth rate as primary outcome. The major and minor adverse effects of their administration should also be captured in any future studies. Notably, our results should limit the use of many adjunct or add-on interventions in women with RIF whose prescription is currently extremely popular in IVF clinics around the world. In this regard, the administration of LMWH is not supported by evidence either regarding its efficacy or its safety profile. We also strongly discourage intentional endometrial injury with the aim of improving IVF outcome outside of registered experimental protocols.
RIF of unknown cause significantly hampers IVF success. An effective treatment strategy would constitute a revolution in the field. In this context, future research should focus on confirming therapeutic approaches for which robust efficacy data are already available [i.e. intrauterine PBMC infusion and subcutaneous G-CSF administration] before investigating new interventions or therapies or retest those supported by preliminary flabby evidence. Finally, regardless of the option to be tested, we plea for collaborative efforts that could allow to run large and robust RCTs. In recent years, RIF has become extremely popular with entire meetings exclusively dedicated to the argument. The time has now come for facts rather than speculations. www.nature.com/scientificreports/