Endometrial scratching in women undergoing IVF/ICSI: an individual participant data meta-analysis

Abstract

Background: In IVF/ICSI treatment, the process of embryo implantation is the success rate-limiting step. Endometrial scratching has been suggested to improve this process, but it is unclear if this procedure increases the chance of implantation and live birth (LB) and, if so, for whom, and how the scratch should be performed.

Objective and rationale: This individual participant data meta-analysis (IPD-MA) aims to answer the question of whether endometrial scratching in women undergoing IVF/ICSI influences the chance of a LB, and whether this effect is different in specific subgroups of women. After its incidental discovery in 2000, endometrial scratching has been suggested to improve embryo implantation. Numerous randomized controlled trials (RCTs) have been conducted, showing contradicting results. Conventional meta-analyses were limited by high within- and between-study heterogeneity, small study samples, and a high risk of bias for many of the trials. Also, the data integrity of several trials have been questioned. Thus, despite numerous RCTs and a multitude of conventional meta-analyses, no conclusion on the clinical effectiveness of endometrial scratching could be drawn. An IPD-MA approach is able to overcome many of these problems because it allows for increased uniformity of outcome definitions, can filter out studies with data integrity concerns, enables a more precise estimation of the true treatment effect thanks to adjustment for participant characteristics and not having to make the assumptions necessary in conventional meta-analyses, and because it allows for subgroup analysis.

Search methods: A systematic literature search identified RCTs on endometrial scratching in women undergoing IVF/ICSI. Authors of eligible studies were invited to share original data for this IPD-MA. Studies were assessed for risk of bias (RoB) and integrity checks were performed. The primary outcome was LB, with a one-stage intention to treat (ITT) as the primary analysis. Secondary analyses included as treated (AT), and the subset of women that underwent an embryo transfer (AT+ET). Treatment-covariate interaction for specific participant characteristics was analyzed in AT+ET.

Outcomes: Out of 37 published and 15 unpublished RCTs (7690 participants), 15 RCTs (14 published, one unpublished) shared data. After data integrity checks, we included 13 RCTs (12 published, one unpublished) representing 4112 participants. RoB was evaluated as 'low' for 10/13 RCTs. The one-stage ITT analysis for scratch versus no scratch/sham showed an improvement of LB rates (odds ratio (OR) 1.29 [95% CI 1.02-1.64]). AT, AT+ET, and low-RoB-sensitivity analyses yielded similar results (OR 1.22 [95% CI 0.96-1.54]; OR 1.25 [95% CI 0.99-1.57]; OR 1.26 [95% CI 1.03-1.55], respectively). Treatment-covariate interaction analysis showed no evidence of interaction with age, number of previous failed embryo transfers, treatment type, or infertility cause.

Wider implications: This is the first meta-analysis based on IPD of more than 4000 participants, and it demonstrates that endometrial scratching may improve LB rates in women undergoing IVF/ICSI. Subgroup analysis for age, number of previous failed embryo transfers, treatment type, and infertility cause could not identify subgroups in which endometrial scratching performed better or worse. The timing of endometrial scratching may play a role in its effectiveness. The use of endometrial scratching in clinical practice should be considered with caution, meaning that patients should be properly counseled on the level of evidence and the uncertainties.

Keywords: ART; ICSI; IPD; IVF; endometrial injury; endometrial sampling; endometrial scratch; endometrial scratching; frozen embryo transfer; individual participant data meta-analysis.

Individual participant data-analysis shows that endometrial scratching in IVF/ICSI is likely to increase live birth rates; the effect did not differ between subgroups of women. FET, frozen embryo transfer; RCT, randomized controlled trial. Created with BioRender.com.

Figure 1.

Flowchart of the systematic literature search, screening, and study selection. Flow chart describing the systematic search for both published (full articles and conference abstracts) and unpublished trials, and subsequent study selection. ^aPublished as full paper or as a conference abstract. ^bNumber of participants unknown because it is unclear if/how many participants have been included. ^cLimited available data due to various reasons, a.o. closure of one of the centers or withdrawal directly after randomization. ^d13 studies, of which 11 published as full text, 1 published as conference abstract, 1 registered but unpublished. n, number; IPD, individual participant data; AD, aggregate data; ITT, intention to treat; AT, as treated; AT+ET, as treated with embryo transfer; LB, live birth; OP, ongoing pregnancy; CP, clinical pregnancy; BP, biochemical pregnancy.

Figure 2.

Risk of bias assessment of the included trials. Risk of bias assessment according to the Cochrane Risk of Bias Tool 2. All studies that were included in the IPD analysis and had a published manuscript or conference abstract were assessed for risk of bias. Thus, the registered but not published trial by Lee et al. was not assessed for risk of bias.

Figure 3.

Forest plot of two-stage Intention to Treat analysis, outcome: live birth. Analysis of studies that did share IPD combined with a recent, yet unpublished, study that did not (yet) share IPD. OR, odds ratio. 14 studies (5155 participants). Shown event numbers are observed values only. This forest plot shows the two-stage analysis of the 13 studies that were included in the IPD analysis combined with a single recent RCT that has been published while drafting this manuscript but that could not yet share IPD (Metwally et al., 2020). For the 13 studies that shared IPD, the per-study and pooled odds ratios are based on imputed data in our IPD dataset. The outcome ‘live birth’ was imputed for 100 participants (fraction of missing outcome data 2.4%). For the single study that did not share IPD (Metwally et al., 2020), the data were withdrawn from their manuscript. Numbers of missing data are unknown.

Figure 4.

Participant-level interaction of age and previous failed transfers with the scratch-effect on live birth. (A) Plot of the interaction between participant age and the scratch-effect on live birth at participant level (one-stage within-trial analysis). The log-OR of scratch versus control for live birth is represented as a function of age (blue line), with pointwise 95% CI (red lines). The slope of the blue line is −0.010 (standard error (SE) 0.02). The black striped line indicates the log-OR of the reference participant with a mean age. P-value 0.60. (B) Plot of the interaction between the number of previous failed embryo transfers (participant-level) and the scratch-effect on live birth (one-stage within-trial analysis). The log-OR of scratch versus control for live birth is represented as a function of previous failed embryo transfers (blue line), with pointwise 95% CI (red lines). The slope of the blue line is 0.04 (SE 0.06). The black striped line indicates the log-OR of the reference participant with 0 previous failed embryo transfers. P-value 0.23.

Figure 5.

Interaction of age with scratch-effect on live birth, combined representation of IPD participant-level analysis, IPD study-level analysis, and analysis within each trial. Plot of the interaction between age and the scratch-effect on live birth, at participant- and study-level. The red line represents the log-OR of scratch versus control for live birth as a function of mean participant age (one-stage participant-level analysis), and the green line represents the meta-regression of per-trial treatment effect estimates versus mean participant age (two-stage study-level analysis). The dashed black lines represent the interaction between age and treatment effect on live birth within each trial (two-stage analysis).

Figure 6.

Funnel plot of studies that did and did not share IPD and reported live birth. Two-stage ITT analysis, outcome live birth. Blue triangles : studies that did share IPD. Red dots : studies that did not share IPD. Funnel plot representing all published trials and their reported endometrial scratching effect on live birth. The funnel (black dotted lines) indicates the lower and upper limits of the 95% confidence region. The black dotted mid-line is the pooled effect (OR) of endometrial scratching on live birth. The slight asymmetry to the left corner could indicate either publication bias or small study effect. IPD, individual participant data; Y-axis, standard error; X-axis, scratching effect on live birth, expressed as OR.

Figure 7.

Risk of bias of studies that did and did not share IPD. The Cochrane Risk of Bias Tool 2 was used. Overview of the Risk of Bias assessment of all published studies. Studies that were registered but unpublished could not be assessed. Studies above the dashed line were included in the IPD analysis. Studies below the dashed line did not share IPD, or were excluded from IPD analysis (Mahran et al., 2016; Hebeisha et al., 2018). IPD, individual participant data available; AD, only aggregate data available.

Figure 8.

Forest plots of studies that did and did not share IPD. Two-stage intention to treat analysis, outcome live birth. 23 studies representing 6374 participants: 10 with aggregate data representing 2262 participants and 13 with IPD representing 4112 participants. Forest plots of the two-stage intention to treat analysis of the outcome live birth, split into three analyses: studies that shared IPD and were included in the IPD analysis, studies that were not yet able to share data, and studies that declined participation in the IPD or could not be contacted despite multiple attempts. Aggregate data events, total numbers, and calculations are based on the numbers as reported in the manuscript, where we have tried to adhere to the intention to treat principle as much as possible. Individual participant data events and total numbers are based on observed values only, representing the individual participants that were included in the final analysis of this IPD-MA. The calculations for OR and 95% CI are based on the imputed data included in this IPD-MA. LB, live birth; OR, odds ratio. Aggregate data: studies that did not share individual participant data (IPD).