The impact of endometrial injury on reproductive outcomes: results of an updated meta‐analysis

Abstract Background It is still unclear whether endometrial injury (EI) has a beneficial effect on reproductive outcomes, and if so, the optimal procedure characteristics are not clear. All previous papers concluded that more research is needed, and as additional studies were recently published, the insights on EI have changed significantly. Methods Searches were conducted in MEDLINE, Embase, Web of Science, and Cochrane Library, to identify randomized controlled trials examining the EI effect on IVF outcomes in women at least one previous failed cycle. Results 2015 references were identified through database searching. Ultimately, 17 studies were included, involving 3016 patients. Clinical pregnancy rate (CPR) (RR = 1.19, [95% CI 1.06–1.32], P = .003) and live birth rate (LBR) (RR = 1.18, [95%CI 1.04–1.34], P = .009) were significantly improved after EI. Number of previous failed cycles, maternal age, and hysteroscopy were found to be relevant confounders. Higher CPR and LBR were found when EI was performed twice, while performing EI once did not significantly improve reproductive rates. Conclusion According to the present meta‐analysis, EI may be offered to younger patients with few previous failed cycles and should be additionally studied in an RCT comparing different timing and more than one EI before treatment.

endometrium is locally intentionally damaged, usually by a Pipelle catheter.
Many studies have been published on the efficacy of EI and its true benefit on reproductive outcome, including several reviews and meta-analyses, and basic science studies. 6,7 Our recently published meta-analysis of randomized controlled trials (RCT) studied the EI effect in women with a least one previous failed IVF cycle. 8 We showed that improved clinical pregnancy rates (CPR) and live birth rates (LBR) were apparent mainly in younger patients. However, in the subgroup of women with at least two previous failed cycles, the EI effect was not found beneficial.
Later, similar reviews were published. [9][10][11] Vitagliano et al. showed improved reproductive outcomes in women with two or more previous failed cycles, with the greatest beneficial effect seen when double luteal EI was performed. 10 Van Hoogenhuijze et al. found improved CPR but no improved LBR in women with at least two previous failed cycles, concluding that it is still unclear whether EI improves IVF outcomes. 9 In line with this meta-analysis, Gui et al. did not find any significant difference in CPR or LBR when including only RCT in their analysis. 11 The recently published RCT by Lensen at al. concluded that EI did not improve LBR. 12 Further published editorial recommending stated that it is "Time to Stop" offering EI to patients. 13 However, in this RCT EI was performed in a time window that may have potentially skewed the results. The EI was performed between day three of the cycle preceding the IVF cycle and day three of the IVF cycle.
However, in previous studies EI was mostly studied when performed during the preceding cycle. Moreover, two studies examining EI effect when performed during the same cycle presented harmful reproductive results. 14,15 Basic science studies proving the beneficial EI effect entailed two or more EI procedures, 5,7,16,17 thus raising the question whether it takes more than one EI to induce a proper immunological response.
Optimal timing and quantity of EI have not yet been extensively discussed, yet they are potential confounders.
As all previous papers concluded that more research is needed, and due to accumulating new data on EI, we thought that an updated meta-analysis is needed, emphasizing on analyzing the clinical outcomes when EI is performed more than once.

| MATERIAL S AND ME THODS
This is an updated meta-analysis, conducted according to the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) statement, 18 with search strategies, data extraction, and synthesis thoroughly described in our former paper. 8 Study protocol, as previously described, is available at PROSPERO International prospective register of systematic reviews (registration number CRD42018092773). As no substantial changes were made, a new protocol was not required. 19 Searches were conducted in the following databases: MEDLINE(R) by OvidSP interface and PUBMED, Embase, Web of Science and Cochrane Library, on January 28 th , 2020

| Study selection
Considered for inclusion were RCTs examining the EI effect on reproductive outcomes in women with at least one previous failed IVF cycle. In addition, we considered for inclusion studies that presented a subgroup analysis of patients with prior failed IVF attempts.
We contacted authors by email if insufficient information was published.

| Outcomes measured
Our main outcomes were CPR and LBR. CPR was defined as the presence of a gestational sac presenting a positive heartbeat on transvaginal ultrasound. LBR was defined as the delivery of one or more live infants.
Secondary outcomes were multiple pregnancy and miscarriage rates. Multiple pregnancy rate was defined as the presence of more than one gestational sac on transvaginal ultrasound. Miscarriage rate was defined as fetal loss prior to the 20th week of gestation per clinical pregnancy.
Subgroup analyses were performed for known confounders such as at least two previous failed IVF cycles, maternal age, the use of hysteroscopy, and the number of times EI was performed before IVF treatment.

| Assessment of risk of bias
Quality of RCTs was determined by the Cochrane Collaboration's Risk of Bias tool Two independent reviewers made the assessment and if disagreements arose, the issues were resolved by discussion.
Publication bias was assessed by contour-enhanced funnel plots, as well as the Begg and Mazumdar's test and Egger regression asymmetry test. According to Cochrane Handbook for Systematic Reviews of Interventions, testing for publication bias by funnel plot asymmetry should not be conducted when less than ten studies are included in the meta-analysis in order to avoid a false result. Thus, funnel plots were assessed only in comparisons including at least ten trials.

| Data synthesis
RevMan 5.3 (Cochrane Collaboration, Oxford, UK) was applied for our quantitative synthesis. Heterogeneity across studies was assessed by the I-squared statistic (an I-squared statistic <25%-low level of heterogeneity, 25%-50%-moderate level, and >50%-high level). According to the heterogenicity, pooling of the results was performed using either the Mantel-Haenszel fixed-effects model or the Der Simonian-Laird random-effects model. The results were measured by risk ratio (RR), presenting the confidence interval (CI) and P value. A two-tailed P < .05 was considered statistically significant. Sensitivity analyses were performed by omitting studies one-by-one from the analyses. Quality assessment was conducted according to the GRADE criteria.

| Study selection
Altogether, 2015 titles and abstracts were identified through database searching. All potentially relevant studies were reevaluated for inclusion. Figure 1 Table 1 summarizes the characteristics of the included RCTs.

| Study characteristics
Five studies with appropriate subgroup analyses answering our inclusion criteria were included in our meta-analysis. 12,27,29,30,35 Due to high risk of bias in the randomization process and allocation according to the clinical case record number, the study published by Matsumato et al. was excluded from our analysis. 36 Eleven studies included patients with at least one previous failed cycle, 12,[20][21][22]26,27,[29][30][31]33,35 four studies included patients with at least two previous failed cycles, [23][24][25]34 and the remaining two studies 28,32 included patients with at least three previous failed IVF cycles (Table 1). Three studies, in addition to presenting the data for patients with at least one previous failed cycle, provided further data for patients with at least two previous failed cycles. 12,22,33 The average age of patients in five studies was up to and including 30 years old, 24,25,31,33,34 and above 30 in nine studies. 12,[20][21][22][23]26,28,32,35 Three studies did not report the average age of included patients with previous failed cycles as the included data originated from a subgroup analysis 27,29,30 (Table 1).
Hysteroscopy was part of the protocol in four studies. 21,25,31,32 Gurgan et al. 32 Figure 2 presents the risk of bias summary. One study had unclear risk of selection bias due to lack of description of the allocation sequence method, 23 while seven studies had unclear risk of selection bias because allocation concealment method was not noted. 23,24,26,28,[32][33][34] Most studies were not blinded due to the nature of the procedure.

| Risk of bias of included studies
However, we believe that lack of blinding was unlikely to affect the results, thus risk for detection bias was rated low for all studies.
The risk for attrition bias was high in one study that did not present a CONSORT flow diagram or describe the follow-up of patients. 23 Reporting bias was rated high in four studies due to presentation of the results as percentage, presentation of ongoing pregnancies and LBR as one outcome or due to presentation of only one outcome in the subgroup analysis. 26,28,30,35 Unclear risk of reporting bias was also found in nine studies due to absent or retrospective clinical trial registration. 21

| Clinical pregnancy rate
CPR forest plots are presented in Figure 3.  (Figure 4).

| Live birth rate
LBR forest plots are presented in Figure 5.

| Miscarriage rate
Two studies reported miscarriage rate per positive pregnancy test or per cycle initiated (as opposed to per clinical pregnancy) and therefor were not included in this analysis. 22  Subgroup analyses were not conducted, as the number of studies per each comparison was low and not appropriate for a meta-analysis.

| Data synthesis
Most analyses resulted in low to moderate heterogenicity, with only two subgroup analyses (CPR and LBR of studies including patients with two or more previous failed cycles) presenting an I-squared statistic of 60% and 61%.
Sensitivity analyses were conducted by omitting studies oneby-one from the analyses. In each comparison, this action did not change the significance of results, apart from one LBR subgroup of analysis, as mentioned above and in comparisons that resulted in borderline significance.
Comparisons of CPR, LBR, and miscarriage rates were assessed for publication bias by funnel plots (Supplementary Figures 1-3). No asymmetry was detected; however, due to absent or retrospective clinical trial registration in nine studies (as mentioned), the risk for publication bias was defined as moderate.
Using the GRADE criteria, overall quality of existing evidence was initially described as "high" in light of RCTs regarding data acquisition. Nevertheless, the final grading was defined as "moderate," mainly due to moderate risk of bias in most included studies and moderate inconsistency.

| Main findings
Treatment of RIF patients is often frustrating, as the optimal management is not certain. The hope that an endometrial biopsy may help these couples has encouraged many physicians to examine it. 37 The effect of EI on reproductive outcomes has been repeatedly studied; even since our recent meta-analysis, 8  however, it has been shown that the age-related decline in female fertility is mostly related to oocyte quality rather than endometrial receptivity. 38,39 Hysteroscopy has been studied to have an independent EI effect, thus subgroup analyses omitting studies that included hysteroscopy as part of the treatment or protocol were conducted. These subgroup analyses showed that the CPR and LBR were no longer improved. Reaching a conclusion from these results is difficult as the studies varied in hysteroscopy use. This information emphasizes that hysteroscopy is indeed a confounding factor needs to be further addressed in future studies.
Our most interesting and surprising result refers to the optimal number of EI needed to be performed to achieve the best reproductive outcome. Studies included in our meta-analysis performed EI once or twice, mostly in the luteal phase but not exclusively (

| Strengths and limitations
The present updated meta-analysis presents the analysis of all published data from RCTs examining the effects of EI in women with previous failed IVF cycles. Also, we approached the authors of all studies for additional data to conduct more accurate comparisons.
We present novel aspects of EI, regarding the optimal procedure characteristics and the possible need of more than one procedure for most favorable outcomes.
F I G U R E 7 Multiple pregnancy rate-Forest plot In view of varying inclusion criteria and EI application in the included RCTs, we were not able to eliminate all confounding factors (eg, stage and quality of embryos transferred). The type of EI may also have clinical impact as a Pipelle catheter, metal scratching, and aspiration may yield different results. In addition, of the 17 included studies, nine provided the reproductive outcomes in women with at least two previous failed cycles, more suitable for the definition of RIF. Methodological issues, also noted by Li et al. 41 stress the need for future high-quality RCTs, which in turn will translate into high-quality evidence in reviews and meta-analyses.
In our opinion, the optimal study that will prove whether an EI effect truly exists with minimal confounding factors is an RCT of EI in ovum donation cycles in RIF patients. Such study has not yet been published. "/>

| Conclusion
To conclude, the optimal population and procedure characteristics that may yield the greatest benefit from EI are still unknown and a matter of clinical discussion. 42 Our data suggest that the relative contribution of endometrial receptivity to the chances of implantation may decrease with increased age and when performed in women with many failed cycles.
The effect possibly increases when performed two or more times.
Even though, we should embrace these results with caution, as sources of bias were detected in the analyzed studies.
In summary, EI should be offered restrictively, trying to identify which patient could truly benefit from the procedure. According to the present meta-analysis, these may be the younger patients, with at least one IVF failure, and with EI performed twice in the cycle preceding the current treatment.
To confirm the observed beneficial effect of performing more than one endometrial biopsy, an RCT comparing EI in the follicular phase, luteal phase, and/or both should be conducted.

ACK N OWLED G EM ENT
We are grateful to Mrs. Leora Mauda, librarian at the Alfred Goldschmidt Medical Sciences Library of the Technion Institute of Technology, who provided useful assistance in the systematic literature search.

CO N FLI C T O F I NTE R E S T
Chen Nahshon, Lena Sagi-Dain and Martha Dirnfeld declare that they have no conflict of interest.

H U M A N/A N I M A L R I G HTS
This article does not contain any studies with human and animal subjects performed by the any of the authors.

A PPROVA L BY E TH I C S CO M M IT TE E
Not applicable (systematic review and meta-analysis).