A comprehensive electronic literature search was conducted across Embase, MEDLINE, and the Cochrane Central Register of Controlled Trials (CENTRAL). The search date was 09 August 2024. Database interrogation, utilising the PubMed and NICE Healthcare Databases Advanced Search (HDAS) interfaces, employed a systematic search strategy incorporating Medical Subject Headings (MeSH), pertinent keywords, and Boolean operators “AND” and “OR”. Key search terms included “recurrent implantation failure”, “repeated implantation failure”, “immune profiling”, “comparison”, “cytokine”, “immune cell”, “assay”, “analyte”, and “biomarker”. Full search string combinations are detailed within the Supplementary Material File .

Following the removal of duplicate records, two independent reviewers [DB and YA] conducted a two-stage screening process. Initially, titles and abstracts were assessed against pre-defined eligibility criteria. Subsequently, full-text articles meeting the initial criteria were reviewed. DB and YA independently extracted data into an electronic spreadsheet. These data related to study characteristics, patient demographics, and outcomes. Discrepancies arising during any stage were resolved through further review, discussion and, where necessary, consultation with a third reviewer [LM].

Original studies were included if they met the following criteria: (i) reported comparative outcomes between patients with uRIF and controls; (ii) reported quantitative levels of the following immune factors (where provided) in blood and tissue samples: cytokines and chemokines, growth factors, angiogenic markers, and coagulation factors, adhesion molecules, and matrix metalloproteinases (MMPs) and inhibitors; (iii) were published in the English language; and (iv) were published on or after 1 January 2000.

Studies were excluded if they met any of the following criteria: (i) reported on patients with RIF and a history of genital tract abnormality, infectious or chronic autoimmune disease, or other conditions affecting systemic inflammation; (ii) did not provide a direct comparison between two or more groups; (iii) did not utilise controls with a history of pregnancy; (iv) utilised controls with a history of recurrent pregnancy loss (v) were not published as full-text manuscripts (encompassing abstracts, conference proceedings, and investigations with incomplete datasets); (vi) were not primary research articles (therefore excluding review articles, meta-analyses, case reports, and letters to the editor) or (vii) animal studies.

Outcome measures were pre-specified a priori. The primary outcome measure was defined as the differential concentration of immune analytes between individuals with uRIF and control participants. When extracting data, units were recorded as pg/mL (which is equivalent to pg/mL). Meta-analysis was performed if at least 3 studies provided comparative results for a specific immune analyte. Meta-analysis was performed after conversion of summary statistics to mean and standard deviation, if median and range/inter-quartile range (IQR) were provided instead. As study populations exceeded n=25 in each case, mean and median were considered interchangeable (14). Standard deviation was calculated by either range*0.25 (14) or IQR/1.35 (15).

Meta-analyses were performed using the “meta” package in RStudio (Version 2024.12.1 + 563). For each analyte, a random-effects model (DerSimonian-Laird method) was used to pool the mean differences between RIF and control groups. Heterogeneity between studies was assessed using the I² statistic and Cochran’s Q test. A p-value <0.05 was considered statistically significant.

To appraise the quality of the findings from the meta-analysis, the certainty of evidence was evaluated using principles described in the GRADE framework (16). This entailed an assessment of study quality using the AXIS appraisal system (17) with scores of 18 or higher suggesting good quality, 15–17 moderate quality, and 14 or less suggesting low quality. Heterogeneity was assessed using the I² statistic and Cochran’s Q test, and publication bias through the generation and interpretation of Funnel Plots.

The 12 included studies were published between 2003 and 2024. All studies were observational in design and cross-sectional from the description of their methods. Study characteristics are presented in Table 1 . Some 8 studies investigated the immune profile in blood samples alone, 3 in both blood and uterine tissue samples (including tissue and fluid aspirates) and 1 in uterine irrigation fluid alone. All studies compared a population of women with RIF to fertile controls.

The definition of RIF varied across studies; RIF was defined by the number of transfers in some studies (≥3) and the number of cycles in others (≥2 or ≥3). The number of embryos transferred in each cycle as part of these definitions also varied across studies from ≥2 to ≥10. The cycle stage during which samples were collected was specified as mid-luteal in most studies (8/12) and not defined in the remainder. Control groups varied greatly in detail across the studies, but they all generally included women with successful pregnancies either with or without the use of assisted reproductive technology ( Table 1 ).

There were a variety of laboratory-based methods employed to analyse samples across the 12 studies. Enzyme-linked immunosorbent assays (ELISA), and variations such as Sandwich Immunoassays were utilised for single analyte quantification, while multiplex assays, including Luminex and AimPlex, facilitated the simultaneous measurement of multiple analytes through bead-based flow cytometry. Cytometric bead arrays (CBA) offered another flow cytometry-based approach for multiplex analysis. Additionally, Western blotting and Immunohistochemistry (IHC) were utilised in those studies analysing tissue samples. Study design considerations are presented in Table 2 .

Owing to a paucity of studies reporting on immune mediators in the local uterine compartment, the analysis of this review is centred on blood analytes. A total of four studies investigated the uterus, with two examining endometrial biopsies and the other two analysing mediator concentrations in uterine cavity irrigation fluid.

Some 8 studies reported differential blood concentrations between RIF patients and controls for a variety of immune markers. The immune markers analysed in each study are summarised in Table 2 . Abdulhaleem et al. (18), in a case control study of 154 women demonstrated a significantly lower level of Angiopoietin-2 in patients with RIF compared to controls (2915pg/ml vs 3236pg/ml, p=0.007) and those with IVF success (2915pg/ml vs 3166pg/ml, p=0.009). Benkhalifa et al. (19), in a cohort study of 44 women, demonstrated lower levels of MMP-7 (119.97pg/ml vs 281.11pg/ml, p=0.03) and VEGF (30.93pg/ml vs 82.54pg/ml, p=0.022) in women with RIF as compared to controls. Comins-Boo et al. (20) studied the widest panel of immune analytes of all the studies, a total of 23 in an observational study of 55 patients; positive findings were higher levels of PDGF-BB (3498pg/mL vs 1659pg/mL, p<0.05), TGF-beta1 (25817pg/mL vs 15567pg/mL, p<0.05) and TGF-beta3 (249pg/mL vs 140pg/mL, p<0.01) in patients with RIF as compared to healthy controls. Guo et al. (21) demonstrated lower levels of IL-10 (2.18pg/mL vs 3.37 pg/mL, p=0.034), lower levels of G-CSF (5.36pg/mL vs 7.83pg/mL, p=0.033) and higher levels of IL-6 (3.61pg/mL vs 2.45pg/mL, p=0.042) in patients with RIF as compared to controls in a study of 70 women from China. They also investigated cytokine ratios, demonstrating higher ratios of IL-2/IL-10 as well as of IFN-γ/IL-10 in patients with RIF as compared to controls. Liang et al. (22) demonstrated significantly higher levels of IFN-γ, IL-6, IL-1b, IL-4, and TGF-b1 in 34 women with RIF as compared to 25 healthy controls. Furthermore, higher ratios of IFN-γ/IL-4, IFN-γ/IL-10, IFN-γ/TGF-b1, IL-6/IL-10, IL-6/TGF-b1, and IL-1b/TGF-b1 were seen in the RIF cohort. Nenonen et al. (23) demonstrated lower levels of IFN-γ (5.45pg/mL vs 5.78pg/mL, p=0.001) and IL-2 (0.26pg/mL vs 0.45 pg/mL) in RIF patients compared to controls. Kharamani et al. (24) performed the largest study included in this review: a comparative analysis of 400 RIF patients and 400 controls. They focussed on Fibroblast Growth Factors alone and demonstrated significantly lower levels in the blood of women with RIF (17pg/mL vs 23pg/mL, p=0.008). Kalem et al. (25) also focussed on a single analyte, CCL2, key for monocyte recruit to tissue, and demonstrated higher levels in RIF patients (and those with recurrent miscarriage) than controls (29.8pg/mL vs 22.7pg/mL, p<0.001). Gong et al. (26) demonstrated increased levels of IL-21 (21.8pg/mL vs 14.01pg/mL, p<0.05) and IL-6 (14.45pg/mL vs 9.87pg/mL, p<0.05) as compared to controls. Bastu et al. (27) studied the role of two glycoproteins: Mucin 1 and Glycodelin A. Both were found to be significantly lower in patients with RIF as compared to controls. Finally, Taheripanah et al. (28) demonstrated lower glycodelin A concentrations in the blood of RIF patients than controls (30.1 ng/mL vs 44.5 ng/mL, p<0.001).

Five blood analytes were reported in at least 3 studies and so enabled quantitative review with meta-analysis; they were IFN-γ, IL-4, TNF-alpha, IL-2 and IL-6. Studies included in the meta-analysis were Comins Boo et al. (20), Guo et al. (21), Nenonen et al. (23) and Gong et al. (26). One additional study (Liang et al. (22)) was identified as having cytokine concentration levels many orders of magnitude higher than those reported in other studies. The inclusion of this study would have resulted in a failure of the meta-analysis model to converge, indicating an undue influence and violation of assumptions of the meta-analysis. Consequently, this study was excluded from all meta-analyses to ensure the validity and stability of the results.

Heterogeneity was not significant in any of the 5 meta-analyses performed, as determined by the I² statistic and Cochrane Q test. Publication bias was not a concern given the appearances of Funnel Plots for the 5 meta-analyses performed ( Supplementary Material ). A statistical test for publication bias using Egger’s test was not possible given the low number of studies in each meta-analysis.

In terms of study quality, the 4 studies were assessed using the AXIS appraisal system, which was designed specifically for cross-sectional studies (17). There were concerns raised by the AXIS tool for study quality across the 4 studies. 3 studies [Nenonen (23), Guo (21) and Comins-Boo (20)] were graded as “moderate” quality (score range 15–17 inclusive) and 1 [Gong (26)] as “poor” quality (score 14 or less). The scoring sheet is provided in the Supplementary Material .

There was inconsistency between studies in terms of definitions for both RIF and the nature of the controls. Furthermore, as described earlier, 1 study [Liang et al. (22)] was also excluded from meta-analysis due to the concerns with unexplained magnitude of result compared to other studies, for absolute cytokine concentrations.

Overall, there would be low confidence in the findings from the meta-analysis when all the considerations above are considered within a GRADE framework.

This systematic review and meta-analysis represent, to our knowledge, the first attempt to quantitatively synthesise the evidence comparing specific immune mediators (encompassing cytokines, angiogenic factors, and tissue remodelling factors) between women with unexplained recurrent implantation failure (uRIF) and fertile controls. Our analysis of twelve observational studies published since 2000 revealed significant heterogeneity in study design, RIF definitions, control group selection, and laboratory methodologies. Despite pooling data for five peripheral blood cytokines (IFN-γ, IL-4, TNF-α, IL-2, IL-6) reported in three or more studies, only Interleukin-4 (IL-4) demonstrated a statistically significant difference, being lower in women with RIF compared to controls (MD -0.0298, 95% CI: -0.0436 to -0.0159, p < 0.0001). Meta-analyses for peripheral IFN-γ, TNF-α, IL-2, and IL-6 levels did not reveal statistically significant differences between the groups. Individual studies reported varied associations for a wide array of other analytes in both blood and uterine samples (including tissue and fluid), such as lower levels of Angiopoietin-2, MMP-7, VEGF, FGF1, Glycodelin A, and MUC1, and higher levels of PDGF, TGF-β isoforms, CCL2 and IL-21 in RIF cohorts. However, due to the limited number of studies investigating each specific analyte and methodological inconsistencies, further quantitative pooling was largely prevented. The overall certainty of the evidence derived from the meta-analysed studies was assessed as low using the GRADE framework (16), primarily due to concerns regarding study quality and clinical heterogeneity.

We observed significantly lower peripheral IL-4 concentrations in women with RIF. This finding is important, especially considering the low statistical heterogeneity in this specific meta-analysis (I²=12.0%). IL-4 is a key Th-2 cytokine, typically associated with promoting immune tolerance crucial for embryo implantation, in contrast to pro-inflammatory Th1 responses implicated in implantation failure (13). Reduced systemic IL-4 levels could therefore suggest a shift away from this required state of immune tolerance. However, this observation is not consistent across all individual studies (e.g., Liang et al. (22), excluded from meta-analysis due to data anomalies) and challenges the established Th1/Th2 model. The biological importance of this single systemic finding remains uncertain, particularly whether it accurately reflects the critical immune dialogue at the feto-maternal interface within the endometrium. The absolute serological level difference of IL-4 between RIF and control patients was small and therefore raises the question of whether there is any clinical significance despite there being statistical significance.

Conversely, the lack of significant pooled differences for peripheral IFN-γ, TNF-α, IL-2, and IL-6 is also informative. Although individual studies in our review and the wider literature report associations between these cytokines and RIF [e.g. Guo et al. (21), Liang et al. (22), Nenonen et al. (23)], our meta-analysis indicates no consistent, statistically robust difference in their systemic levels based on current quantitative evidence. This result might reflect several factors: (i) a true absence of a strong systemic association; (ii) inadequate statistical power due to the small number of studies; (iii) the masking effect of substantial clinical and methodological heterogeneity; or (iv) peripheral blood measurements may be poor indicators of the dynamic local immune environment within the endometrium during implantation. The moderate heterogeneity for IL-2 (I²=56.9%) and IL-6 (I²=61.5%), although not statistically significant (perhaps due to low study numbers), suggests underlying variability between studies that requires cautious interpretation.

The varied findings from individual studies across diverse analytes (e.g., VEGF, FGF, MMP-7, Glycodelin A (GdA), CCL2) highlight the complex nature of immune pathways involved in RIF. Reduced levels of factors involved in angiogenesis (e.g., Ang-2, VEGF) or endometrial receptivity (e.g., GdA, MUC1), along with changes in tissue remodelling enzymes (MMPs) and growth factors (e.g., PDGF, TGF-β), could reasonably contribute to implantation failure.

Due to the limited number of studies, our review was unable to provide a detailed interpretation of findings from analyses of endometrial tissue or uterine fluid. Only four studies investigated the uterine compartment: two examined endometrial tissue (26, 27), and two analysed uterine fluid (28, 29), with findings related to markers such as lower glycodelin-A (GdA), higher IL-21/CXCR5, and altered matrix metalloproteinases (MMPs) and cytokines. Although data on endometrial mediators in RIF are scarce, significant research has been conducted in other clinical contexts. Notably, a recent randomised controlled trial by Lédée et al. demonstrated improved live birth rates in patients undergoing their first embryo transfer following personalised immunomodulatory interventions (30). These interventions were guided by endometrial immune profiling, specifically assessing mediators related to natural killer (NK) cell function (e.g., IL-15 and IL-18). This approach highlights a promising area for future investigation in RIF patients.

The endometrial compartment offers the most direct insight into the maternal-fetal immunological interface. However, the practicality and feasibility of investigating this compartment, especially during the narrow window of implantation, remains a topic of debate. While systemic blood measurements may not perfectly mirror the immunological status of the uterine niche, they could provide valuable insights into a general inflammatory state that either contributes to or results from implantation failure, thereby perpetuating a non-receptive endometrial environment (31).

This systematic review followed a rigorous protocol, adhering to PRISMA guidelines (13) and prospective PROSPERO registration, enhancing transparency and minimising reporting bias. We used a comprehensive search strategy across multiple databases. The review focussed on studies reporting quantitative analyte levels, allowing objective comparison and meta-analysis where feasible. Inclusion and exclusion criteria were defined precisely to target unexplained RIF and appropriate fertile controls, aiming to isolate immunological factors. Data extraction and study selection were performed independently by two reviewers to reduce error. We formally assessed the risk of bias within studies using the AXIS tool (17) and evaluated the overall certainty of meta-analysis evidence using GRADE principles (16), providing a critical perspective on the findings’ reliability.

However, several important limitations affect the interpretation and generalisability of our findings. First, all included primary studies were observational, preventing the establishment of causality. Second, there was substantial heterogeneity in RIF definitions (varying numbers of failed cycles/embryos) and control group characteristics, introducing a risk of confounding. Historically, RIF definitions have varied greatly (e.g., ≥2 failed cycles or ≥3 failed cycles) (1), while more recent ESHRE guidelines propose an individualised approach based on cumulative predicted chance of implantation (9). The variability in RIF definitions across studies makes direct comparison and quantitative synthesis difficult, and we accept this as a limitation of meta-analysing such data.

Third, methodological variability was considerable, including different sample types (blood, tissue, fluid), diverse assays (ELISA, multiplex, IHC, Western Blot) with varying sensitivity and specificity, and inconsistent reporting of sample timing (though most targeted the mid-luteal phase).

Fourth, the small number of studies per meta-analysis (n=3 or 4) limited statistical power to detect subtle differences and prevented robust assessment of publication bias or meaningful subgroup analyses. Furthermore, this review was restricted to specific soluble mediators and did not capture the full complexity of immunological assessment, which often includes cellular components (e.g., NK cell counts/activity, Treg populations) or genetic factors (e.g., KIR-HLA interactions). Finally, the reliance on peripheral blood studies may not accurately reflect crucial local events within the endometrium. The few studies examining tissue or uterine fluid, although possibly more relevant, were limited in number and used varied techniques. The quality assessment revealed moderate-to-poor methodological rigour in many studies, further reducing confidence in the pooled estimates, as reflected by the overall ‘Low’ GRADE assessment.