Endometrial microbiota-targeted therapies for chronic endometritis-associated recurrent implantation failure and their impact on IVF outcomes: A systematic review and methodological quality assessment.

B.D.: Methodology, Conceptualization, Formal analysis, Software, Writing—review and editing, Visualization, Validation. O.‐D.I.: Methodology, Conceptualization, Data curation, Formal analysis, Software, Writing—original draft, Writing—review and editing. A.‐M.D.(C): Data curation, Formal analysis, Writing—original draft. I.S.: Data curation, Formal analysis, Writing—original draft. G.L.: Data curation, Formal analysis, Writing—review and editing, Validation. M.D.: Data curation, Writing—original draft. C.I.: Methodology, Writing—review and editing, Visualization, Validation. All authors have read and agreed to the published version of the manuscript.

This work did not receive any specific grant from any funding agency in the public, commercial, or not‐for‐profit sector.

From an initial total of 1850 records identified, 1306 were retrieved through the primary database, and an additional 544 were found via citation tracking and related article searches. These latter sources included duplicates and entries deemed irrelevant upon closer inspection ( File S4 ). Following a rigorous, multiphase screening and eligibility evaluation conducted according to the PICOS framework, only four studies 61 , 62 , 63 , 64 met the predefined inclusion criteria. Full details of the screening process are detailed in Files S2 and S4 , and summarized visually in the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses flowchart (Figure  1 , Tables  1 and 2 ). Among the studies included in the review, they were published in 2022, 61 , 63 2023 62 and 2024, 64 and were conducted by research groups based in East Asia, specifically Japan 61 , 62 , 63 and China. 64 Notably, three out of four studies were prospectively registered in The University Hospital Medical Information Network Center ( https://www.umin.ac.jp/ ) (accessed on March 15, 2025) and are accessible via the following registry numbers: UMIN‐CTR 000036990, 61 UMIN‐CTR 000036050, 62 and UMIN‐CTR 000029449/000031909 63 respectively. The cumulative number of women enrolled across all studies was 1515, though this population was distributed unequally in two experimental arms, particularly 372 within the experimental group, whereas 1143 as the reference group. PRISMA flowchart. Methodological data of the studies included for review. 36.4 ± 3.2 vs. 38.2 ± 3.8 a 22.7 ± 3.8 vs. 23.0 ± 4.1 b Male factor 31 (26.5%) vs. 19 (34.5%) PCOS 17 (14.5%) vs. 1 (1.8%) EMS 13 (11.1%) vs. 11 (20.0%) TFI 32 (27.4%) vs. 4 (7.3%) Unexplained etiology 38 (32.5%) vs. 7 (12.7%) DOR 7 (6.0%) vs. 31 (56.4%) Gravidity 0 (0–3) vs. 0 (0–2) Parity 0 (0–1) vs. 0 (0–2) No. of cycles 7.9 ± 1.8 vs. 2.1 ± 0.8 No. of ET 9.7 ± 2.1 vs. 2.8 ± 1.2 No. of morphologically good ET 4.2 ± 0.5 vs. 1.1 ± 0.4 No. of AT use 2.6 ± 0.7 vs. 1.0 ± 0.3 No. of HRMU 1.7 ± 0.3 vs. 0.2 ± 0.1 Cigarette smoking 1 (0.9%) vs. 1 (1.8%) Alcohol drinking 2 (1.7%) vs. 1 (1.8%) No. of cycles 7.9 ± 1.8 vs. 2.1 ± 0.8 No. of ET 9.7 ± 2.1 vs. 2.8 ± 1.2 No. of morphologically good ET 4.2 ± 0.5 vs. 1.1 ± 0.4 Number of AT use 2.6 ± 0.7 vs. 1.0 ± 0.3 No. of HRMU 1.7 ± 0.3 vs. 0.2 ± 0.1 36.17 ± 3.45 vs. 35.50 ± 3.56 a 21.92 ± 2.79 vs. 21.95 ± 2.84 b 3.55 ± 3.12 vs. 3.43 ± 3.00 c 8.08 ± 3.84 vs. 8.31 ± 3.35 d Male factor 39 (29.8%) vs. 19 (29.7%) EMS 20 (15.3%) vs. 12 (18.8%) TFI 20 (15.3%) vs. 15 (23.4%) Mean no. of previous ET cycles 4.31 ± 2.40 vs. 3.97 ± 1.94 Infertility duration 40.31 ± 30.80 vs. 40.51 ± 27.60 Delivery 29 (22.1%) vs. 17 (26.6%) Miscarriage 62 (47.3%) vs. 23 (35.9%) Duration of infertility (min) 40.31 ± 30.80 vs. 40.51 ± 27.60 No. of ET cycles 4.31 ± 2.40 vs. 3.97 ± 1.94 36.1 ± 5.1 vs. 37.2 ± 4.3 a 22.1 ± 4.2 vs. 21.6 ± 4.0 b Male factor 27 (32.1%) vs. 288 (29.3%) PCOS 5 (6.0%) vs. 69 (7.0%) EMS 13 (15.5%) vs. 104 (10.6%) TFI 19 (22.6%) vs. 202 (20.5%) Unexplained etiology 41 (48.8%) vs. 435 (44.2%) DOR 10 (11.9%) vs. 151 (15.3%) Gravidity 0 (0–2) vs. 0 (0–2) Parity 0 (0–1) vs. 0 (0–2) No. of cycles 3.9 ± 0.5 vs. 4.1 ± 0.7 No. of ET 4.1 ± 0.6 vs. 4.4 ± 1.2 No. of morphologically good ET 3.2 ± 0.5 vs. 3.4 ± 1.0 No. of AT use 1.8 ± 0.4 vs. 1.9 ± 0.7 No. of HRMU 1.0 ± 0.4 vs. 1.1 ± 0.6 Cigarette smoking 1 (1.2%) vs. 5 (0.5%) Alcohol drinking 2 (2.4%) vs. 38 (3.9%) No. of cycles 3.9 ± 0.5 vs. 4.1 ± 0.7 No. of ET 4.1 ± 0.6 vs. 4.4 ± 1.2 No. of morphologically good ET 3.2 ± 0.5 vs. 3.4 ± 1.0 No. of AT use 1.8 ± 0.4 vs. 1.9 ± 0.7 No. of HRMU 1.0 ± 0.4 vs. 1.1 ± 0.6 32.88 ± 2.03 vs. 30.57 ± 2.64 a 21.11 ± 1.75 vs. 19.96 ± 1.61 b 3.01 (2.08, 6.83) vs. 3.99 (2.10, 6.03) c 6.36 (4.36, 7.11) vs. 6.50 (5.83, 7.12) d 5.01 (1.41, 0.64) vs. 5.43 (1.96, 5.00) e 38.62 (35.00, 49.99) vs. 35.72 (27.37, 43.28) f 0.34 ± 0.23 vs. 0.24 ± 0.15 g 20.58 ± 7.74 vs. 18.30 ± 4.74 h Infertility duration 2 (1, 2) vs. 2 (1, 2) Previous conception 0 (0, 1) vs. 0 (0, 1) Duration of infertility (years) 2 (1, 2) vs. 2 (1, 2) Abbreviations: AT, assisted hatching; DOR, diminished ovarian reserve; EMS, endometriosis; ET, embryo transfer; HRMU, hyaluronan‐rich medium use; NS, not specified; PCOS, polycystic ovarian syndrome; TFI, tubal factor infertility. Age. Body mass index. Anti‐Müllerian hormone. Follicle‐stimulating hormone. Luteinizing hormone. Estrogen. Progesterone. Prolactin. Summary of key methodological and clinical features of included studies. CE‐Histopathology, HSC Microbiome—NGS 117—RIF 55—non‐RIF 24 excluded 19 RIF vs. 5 non‐RIF—underwent antibiotic treatment Infertile RIF vs. non‐RIF women CE vs. non‐CE; microbiota responders vs. nonresponders Clinical pregnancy rate Miscarriage rate Live birth rate CE—HSC Microbiome—NGS Amoxicillin and clavulanic acid for 8 days Metronidazole/Clindamycin for 7 days Lactoflora® for 7–10 days/10–17 days 131—RIF 64—non‐RIF 5 excluded (2—lack of data; 3—dropped out) IVF patients with RIF Sequenced vs. nonsequenced; treated vs. untreated Clinical pregnancy rate Ongoing pregnancy rate Multiple pregnancy rate Biochemical pregnancy rate Miscarriage rate Stillbirth rate Live birth rate CE—Histopathology Microbiome—NGS 3473—RIF 84—MDR‐CE 984—Antibiotics‐sensitive CE 38 excluded (24—lack of data, 14—dropped out) CE subtypes MDR vs. sensitive CE Clinical pregnancy rate Miscarriage rate Ongoing pregnancy rate CE—HSC, IHC Microbiome—NGS 83—RIF 40—CE 40—non‐CE 3 excluded (persistent CE) RIF ± CE CE vs. non‐CE Clinical pregnancy rate Miscarriage rate The four studies included in this analysis 61 , 62 , 63 , 64 exhibited substantial methodological heterogeneity, particularly in the definitions of RIF and CE, patient selection criteria, diagnostic methodologies, and statistical modeling approaches. These discrepancies significantly complicate interpretation and comparability of findings, especially when evaluating the role of CE, endometrial microbiota, and therapeutic interventions in the pathophysiology of RIF. While all studies enrolled infertile women with a history of failed ETs, the operational definitions of RIF were inconsistently applied. Kitaya et al. 61 , 63 and Zhang et al. 64 broadly defined RIF as failure to conceive after two or more ETs of high‐quality embryos, but did not consistently specify embryo grading criteria, morphological standards, or control for transfer stage (blastocyst vs. cleavage). Iwami et al., 62 though focusing on microbiota‐guided interventions, did not provide a clear definition of RIF; patients with any history of implantation failure were included, without information on the number or quality of embryos transferred. These inconsistently applied criteria introduce substantial patient heterogeneity and raise concerns about the validity of RIF classification across studies. When compared against the 2022 Lugano Workshop consensus statement on RIF, none of the studies meet the recommended diagnostic rigor. The Lugano criteria propose that RIF should not be a fixed diagnosis but a descriptive label for patients who fail to achieve pregnancy after three or more ETs with morphologically high‐quality embryos, or after cumulative transfer of four to six embryos depending on age and embryo quality. In contrast, the included studies adopted more permissive thresholds, potentially including patients without a confirmed diagnosis of RIF. This definitional leniency introduces the risk of misclassification bias, hinders comparability across cohorts, and may dilute or distort treatment effect estimates or conclusions regarding microbial causality. In addition to diagnostic inconsistency, the statistical methodologies used across the studies reflect limited analytical rigor. Only Iwami et al. 62 applied multivariate regression to adjust for clinical confounders such as age, endometrial thickness, and embryo quality. The remaining studies used either descriptive statistics or univariate comparisons, without accounting for critical variables like body mass index, hormonal profile, or infection history. Moreover, none of the studies conducted power calculations or sample size justifications, increasing the risk of type II errors and reducing confidence in the robustness of reported associations. Importantly, no study adjusted for multiple comparisons, despite analyzing several outcomes or subgroups. Confidence intervals were inconsistently reported or omitted altogether, and no effect sizes were supported by sensitivity analyses. Additionally, none of the studies employed modern or complex statistical techniques such as interaction modeling, composite indices, or machine learning, which could help elucidate the multifactorial nature of microbial dysbiosis and implantation failure. Substantial heterogeneity also exists in microbiome characterization and CE diagnostic modalities. Iwami et al. 62 utilized 16S rRNA sequencing of intrauterine samples, providing high‐resolution microbial profiling. In contrast, Kitaya et al. 61 , 63 and Zhang et al. 64 used histopathological assessment (e.g., CD138+ plasma cells) and, in some cases, culture‐based identification of pathogens. These methodological disparities in microbial detection, particularly in sensitivity and specificity, may contribute to inconsistent findings regarding the prevalence of dysbiosis and its relation to reproductive outcomes. Another critical methodological aspect concerns the diagnostic definition of CE, which varied considerably across the included studies and constitutes a major source of heterogeneity. Kitaya and colleagues 61 , 63 defined CE histologically, using CD138 (B‐A38) antibody immunohistochemistry (IHC) on 4‐μm endometrial sections collected during the proliferative phase. Stromal CD138‐positive plasma cells with a characteristic nuclear heterochromatin pattern were counted across ≥20 high‐power fields (HPFs, ×400), and an endometrial stromal plasmacyte density index was calculated as the total number of CD138+ cells divided by the number of HPFs examined. CE was diagnosed when endometrial stromal plasmacyte density index ≥0.25, as proposed in previous work by the same group. 17 , 65 To support histopathologic interpretation, office fluid hysteroscopy was also performed between days 6–12 of the menstrual cycle, and images were evaluated following the International Working Group for Standardization of Chronic Endometritis Diagnosis consensus. 66 , 67 In contrast, Iwami et al. 62 based CE diagnosis primarily on hysteroscopic findings, including stromal edema, focal or diffuse hyperemia, and micropolyps <1 mm, confirmed by subsequent biopsy, but without providing a numeric plasma‐cell threshold or specifying the number of HPFs examined. Similarly, Zhang et al. 64 adopted a combined diagnostic approach integrating hysteroscopy and CD138 IHC, consistent with previously described protocols. 16 , 68 , 69 Hysteroscopy was first performed to evaluate uterine morphology and exclude structural abnormalities, followed by endometrial biopsy during the follicular phase. The biopsy sections were cut at 6 μm thickness and stained for CD138‐positive stromal plasma cells to confirm histologic evidence of CE. CE was diagnosed when the number of CD138‐positive stromal plasma cells exceeded four per HPF (≥4 CD138 + cells/HPF), as defined in earlier diagnostic studies. 16 Taken together, these divergent diagnostic strategies underscore the absence of a standardized, universally accepted definition of CE across the available literature. While Kitaya et al. 61 , 63 applied a quantitative endometrial stromal plasmacyte density index (≥0.25 across ≥20 HPFs), Zhang et al. 64 employed a per‐HPF threshold (≥4 CD138 + cells), and Iwami et al. 62 relied solely on hysteroscopic descriptors without numeric quantification. Such variability in diagnostic modalities, plasma‐cell counting criteria, and the use of adjunctive hysteroscopy introduces substantial methodological heterogeneity and potential misclassification bias, particularly when correlating CE status with microbial profiles or reproductive outcomes. This inconsistency in diagnostic thresholds not only limits cross‐study comparability but also emphasizes the urgent need for harmonized histologic and IHC standards to improve reproducibility and strengthen evidence regarding microbiota‐targeted interventions in CE‐associated RIF. Last, the pilot nature and small sample sizes of Kitaya et al. 61 , 63 and Zhang et al. 64 heighten the risk of publication bias, especially given the lack of preregistration, funnel plot analyses, or other bias detection tools. Although statistical tests like Egger's are not feasible with such small samples, the absence of any bias mitigation strategies reinforces the need for caution. In summary, the considerable methodological variability in diagnostic definitions, microbiological assessments, and statistical strategies across these four studies imposes significant limitations on interpretability and generalizability. These issues highlight the urgent need for standardized diagnostic criteria, harmonized microbiota testing protocols, and statistically robust study designs in future research exploring the endometrial environment in RIF populations. As shown in Table  3 and Figure  2 , the overall methodological quality of the four studies ranged from intermediate to high, with scores varying from 5 to 9 out of a possible 9 stars. Quality assessment score of the eligible cohort studies based on NOS. Graphical representation of the overall score following NOS for the eligible cohort studies. Zhang et al. 64 demonstrated the highest methodological rigor, achieving a score of 9. This prospective study clearly defined a representative exposed cohort and an appropriate comparison group. It utilized robust diagnostic tools, including 16S rRNA sequencing and CD138 IHC for CE. Notably, it was the only study that adequately controlled for potential confounding variables through both study design and statistical analysis. The researchers also explicitly reported a 6‐month follow‐up period with minimal attrition. The outcome assessments were based on clinically validated endpoints, which contributed to the high internal validity of the findings. The study by Iwami et al. 62 received a score of 8, indicating strong selection and outcome domains, as well as partial adjustment for confounding variables. Although the study utilized a control group and based treatment on microbial sequencing results, the extent of statistical adjustment was limited. Additionally, while the follow‐up included outcomes from two ET cycles, the specific duration and instances of loss to follow‐up were not fully documented. Nevertheless, the study was prospective in design and provided credible evidence supporting targeted microbiome‐based interventions in cases of RIF. In contrast, the studies conducted by Kitaya et al. 61 , 63 received scores of 5 and 6, respectively, indicating an intermediate quality. Both studies were limited by inadequate control of confounding factors in the comparability domain and lacked robust follow‐up protocols. Although they employed appropriate diagnostic methods for intervention, such as microbiota analysis and histology, their designs were exploratory or pilot in nature. For instance, Kitaya and Ishikawa 61 did not report the duration of follow‐up or consider potential losses to follow‐up, which diminished confidence in the reliability of their outcomes. Their subsequent study 63 showed modest improvements by tracking outcomes postantibiotic therapy over a 12‐month period; however, it still failed to adequately address potential biases or losses to follow‐up. The findings collectively highlight a significant variability in the methodological quality of the studies reviewed. Although all studies utilized valid intervention assessments and outcome measures, the strength of their conclusions depends largely on the duration of follow‐up and the control of confounding factors. The studies conducted by Zhang and Iwami 62 , 64 offer higher quality evidence and should be prioritized in guiding clinical practice or future research. Conversely, the studies by Kitaya et al. 61 , 63 are valuable for generating hypotheses but need to be validated through more rigorous and controlled designs. The included cohort studies were evaluated for internal validity using the ROBINS‐I tool, which systematically assesses seven domains of potential bias in nonrandomized studies of interventions. Across all studies, the per‐protocol effect was estimated, reflecting outcomes among participants who adhered to the intervention, which is consistent with the observational nature of the designs. The most consistent methodological vulnerability arose from bias due to confounding (Domain 1), where studies variably failed to control for key prognostic factors such as patient age, embryo quality, ER, or prior treatment history. Kitaya et al. 61 , 63 were judged to have a “Serious” and “Critical” risk of bias, primarily due to lack of confounder adjustment, unclear participant flow, and insufficient reporting on loss to follow‐up. Although Iwami et al. 62 performed better by including a comparison group and providing prospective outcome tracking, residual confounding and limitations in analytical adjustment warranted a judgment of “Moderate” risk of bias in several domains. Following established guidance from the Cochrane Handbook, where multiple “Moderate” judgments across domains may indicate an overall “Serious” risk of bias, Iwami's study could arguably fall into this category depending on interpretation consistency. 57 , 58 By contrast, Zhang et al. 64 stood out as methodologically sound, with “Low” risk of bias across all domains, having implemented adequate statistical control for confounders, complete outcome data, and clear, prospective reporting. This gradient in risk underscores the varying evidentiary strength of the included studies and supports the prioritization of findings from Zhang and, to a lesser extent, Iwami, when considering implications for clinical practice (Figure  3 ). Graphical representation of the risk of bias following ROBINS‐I in the eligible cohort studies. (A) Risk of bias domains. (B) Overall risk of bias. Clinical pregnancy was consistently defined across all four studies as the presence of an intrauterine gestational sac with fetal heartbeat detected by transvaginal ultrasound at 7 weeks of gestation. 61 , 62 , 63 , 64 A total of 187 clinical pregnancies were confirmed among 370 women, yielding an overall rate of 50.5%. 61 , 62 , 63 , 64 However, considerable heterogeneity exists in terms of patient selection, therapeutic approaches, and microbiota classifications, limiting the feasibility of direct comparison or meta‐analytic pooling. Kitaya et al. 61 , 63 evaluated subgroups including 74 patients with non‐ Lactobacillus ‐dominant microbiota without CE, and patients with multidrug‐resistant (MDR) or antibiotic‐sensitive CE. After oral enteric‐coating lactoferrin treatment aimed at improving microbiota, clinical pregnancy rates improved significantly from 22.2% (2/9) in the dysbiotic group to 71.4% (10/14) posttreatment ( p  = 0.02). 61 However, antibiotic regimens (moxifloxacin or azithromycin) in MDR‐CE groups did not yield significant differences in pregnancy rates: 47.4% (9/19) vs. 44.4% (8/18), p  = 0.85. 61 , 63 Zhang et al. 64 found a significant difference favoring non‐CE patients, with pregnancy rates of 62.5% (25/40) vs. 37.5% (15/40) in CE patients treated with doxycycline ( p  = 0.04). Similarly, Iwami et al. 62 observed improved pregnancy rates after gene‐sequencing‐guided intervention; 42.0% (55/131) vs. 12.5% (8/64), p  < 0.001. These findings highlight variability in study designs and therapeutic approaches, underscoring the need for cautious interpretation and narrative synthesis. Ongoing pregnancy (presence of fetal heartbeat at 12 weeks) was reported by only two studies, 62 , 63 with a total of 107 ongoing pregnancies recorded among 267 women (40.07%). Kitaya et al. 63 reported no significant difference in ongoing pregnancy rates between antibiotic groups: 28.9% (11/38) vs. 32.4% (11/34) ( p  = 0.75). In contrast, Iwami et al. 62 demonstrated a marked increase in ongoing pregnancies in the intervention groups compared to controls: 33.6% (44/131) vs. 9.4% (6/64) ( p  < 0.001), supporting potential benefits of personalized microbiome‐based therapy. Live birth data were reported in three studies, 61 , 62 , 63 accounting for 105 live births among 255 women (41.17%). Kitaya and Ishikawa 61 noted a significant increase in live birth rate postlactoferrin treatment, reaching 57.1% (8/14) compared to 11.1% (1/9) ( p  = 0.04), whereas third‐line oral antibiotic therapy showed no significant effect: 31.6% (6/19) vs. 33.3% (6/18) ( p  = 0.91). 63 Iwami et al. 62 observed an increase in live birth rates after two frozen embryo transfer cycles from 33.6% (44/131) to 48.9% (64/131) ( p  < 0.001) compared with controls, reflecting the influence of cumulative ET. Multiple pregnancy was assessed only by Iwami et al., 62 who reported no significant differences between intervention and control group: 9.4% (6/64) vs. 5.0% (1/20), p  = 0.68. Overall, the rate of multiple pregnancy was low (8.33%), totaling seven cases, and the study was not powered to detect differences in this outcome. Miscarriage, variably defined as pregnancy loss before 12 62 , 64 or 22 61 , 63 weeks, was reported in all studies. Across 195 women, 39 miscarriages (20.00%) were reported. Zhang, Iwami, and their collaborators 62 , 64 found no significant difference in miscarriage rates between CE and non‐CE or intervention and control groups; 13.8% (4/29) vs. 21.1% (4/19) ( p  = 0.79), 64 and 13.0% (15/79) vs. 11.8% (4/25) ( p  = 0.49). 62 Kitaya et al. 61 , 63 also reported similar miscarriage rates irrespective of treatment or microbiota status; 20.0% (2/10) vs. 50.0% (1/2) ( p  = 0.39), 61 and rates of 31.3% (5/16) vs. 26.7% (4/15) ( p  = 0.78), suggesting miscarriage risk may be independent of microbiota‐targeted interventions. 63 However, variability in definitions limits comparability across studies. Biochemical pregnancy, defined as a positive β‐hCG without ultrasound confirmation, was reported exclusively by Iwami et al. 62 Rates were 21.4% (28/131) vs. 10.9% (7/64) ( p  = 0.111) resulting in an overall biochemical pregnancy rate of 17.94%, particularly 35 cases. Stillbirth was rarely reported and occurred in only one case across the studies. Iwami et al. 62 recorded 0% (0/64) in the intervention compared to 4.8% (1/21) in the control group, representing just one case and an overall rate of 1.17%. Collectively, these outcome data should be interpreted with caution, as the aggregated reproductive results likely reflect the combined effects of heterogeneous microbiota‐targeted interventions. Antibiotic, probiotic, and nutraceutical therapies act through distinct mechanisms, ranging from pathogen eradication to microbiome restoration and immune modulation, and their individual efficacy may differ substantially. This therapeutic variability may smooth the apparent overall success rates across studies and contribute to inter‐study heterogeneity.

This systematic review evaluated the effectiveness of therapeutic interventions targeting endometrial microbiota in women with CE‐associated RIF, synthesizing evidence across clinical outcomes, study quality, and risk of bias. Notably, we examined the impact of antibiotics, probiotics, and nutraceuticals on reproductive success, with particular attention to methodological robustness and risk of bias. Across the four included studies, clinical pregnancy outcomes were reported in 370 women, with a rate of 50.5%, while ongoing pregnancy was confirmed in 40.1% of 267 women. Live birth outcomes totaled 105 cases among 255 women (41.2%). Other reproductive endpoints included miscarriage (20.0%), biochemical pregnancy (17.9%), and multiple pregnancy (8.3%), with a stillbirth rate of 1.17%. However, considerable heterogeneity in RIF and CE definitions, diagnostic thresholds for CE, microbiota analysis methods, and intervention types hindered direct comparison and meta‐analytic synthesis. The methodological quality and risk of bias were evaluated using NOS and the ROBINS‐I tool, respectively. According to the NOS, study quality ranged from intermediate to high, with Zhang et al. 64 achieving a score of 9 due to comprehensive cohort selection, robust diagnostic methods, and complete follow‐up. Iwami et al. 62 followed with a score of 8, reflecting strong selection criteria and partial adjustment for confounders. Kitaya et al., 61 , 63 with scores of 5 and 6, respectively, showed methodological limitations in confounding control and outcome reporting. These NOS findings were consistent with ROBINS‐I assessments, which identified bias in D1 as the most common and critical issue. Kitaya et al. 61 , 63 were judged to have a “Critical” 61 and “Serious” 63 overall risk of bias, while Iwami et al. 62 were classified as “Moderate,” and Zhang et al. 64 demonstrated “Low” risk across all domains. A key finding across the included studies is the consistent alteration of specific microbial taxa associated with RIF and CE. As summarized in Table  4 , Lactobacillus species, particularly Lactobacillus crispatus , were generally decreased in dysbiotic endometrial environments, while opportunistic or pathogenic bacteria such as Gardnerella vaginalis , Atopobium vaginae , Prevotella spp., and Sneathia spp. were commonly increased. These shifts in microbial communities appear to contribute to inflammation and impaired ER. Notably, studies involving microbiota‐targeted interventions, such as antibiotics combined with probiotics or lactoferrin supplementation, reported restoration of Lactobacillus dominance alongside improved reproductive outcomes. This underscores the potential therapeutic value of modulating the endometrial microbiota to enhance implantation success. However, variations in detection methods, study populations, and treatment protocols highlight the need for standardized approaches in future research to better elucidate the causal relationships and mechanistic pathways underlying microbiota‐associated infertility. Key microbial taxa alterations in endometrial microbiota across included studies on RIF and CE. Lactobacillus crispatus (↑) Gardnerella vaginalis (↓) Atopobium vaginae (↓) Prevotella spp. (↓) Gardnerella vaginalis (↑) Streptococcus spp. (↑) Bacteroides spp. (↑) Escherichia coli (↑) Lactobacillus spp. (↓) Gardnerella vaginalis (↑) Atopobium vaginae (↑) Prevotella spp. (↑) multidrug‐resistant Escherichia coli (↑) Sneathia spp. (↑) Gardnerella vaginalis (↑) Atopobium vaginae (↑) Lactobacillus crispatus (↓) Note : In Kitaya and Ishikawa (1), only patients with a non‐ Lactobacillus ‐dominant microbiota received oral enteric‐coated lactoferrin supplementation (700 mg/day for ≥28 days), whereas those with a Lactobacillus crispatus ‐dominant profile did not undergo microbiota‐targeted therapy. In contrast, Kitaya et al. (3) evaluated antibiotic‐based treatment for MDR‐CE, administering moxifloxacin 400 mg/day for 10 days or azithromycin 500 mg/day for 3 days, without the use of probiotics or nutraceutical agents. These distinctions clarify that Lactobacillus crispatus dominance in the earlier cohort reflected a baseline eubiotic microbiota, while microbiota‐targeted therapy was reserved for dysbiotic cases. Another important methodological consideration concerns the techniques used for microbiota profiling across studies. While 16S rRNA gene sequencing, conventional culture, and polymerase‐chain reaction‐based methods were employed to characterize microbial communities, these approaches offer limited taxonomic resolution and fail to capture strain‐level diversity or functional capacity. Although 16S rRNA gene sequencing has been instrumental in identifying global compositional patterns of the endometrial microbiota and remains widely applied in reproductive tract microbiome research, its capacity for species‐ or strain‐level resolution and functional or mechanistic inference is inherently limited, as it relies on a single conserved marker gene and indirect functional prediction. 70 , 71 In contrast, shotgun metagenomics and metatranscriptomic approaches enable comprehensive genomic and functional characterization, including pathway analysis, antimicrobial resistance profiling, and multi‐kingdom detection, and are therefore increasingly considered essential for detailed mechanistic investigations of host–microbe interactions. 70 , 72 Nevertheless, in current clinical and translational settings, where standardized disease definitions, microbial thresholds, and interpretative frameworks are still evolving, 16S rRNA sequencing continues to represent a practical, cost‐effective, and informative tool for community‐level profiling and exploratory diagnostics, particularly in large‐scale or routine clinical settings. 37 , 72 Furthermore, the adoption of standardized sampling, sequencing, and bioinformatic pipelines will be essential to ensure reproducibility and cross‐study comparability in CE‐related dysbiosis research and to optimize the evaluation of microbiota‐targeted interventions. 36 , 72 , 73 , 74 Another critical but often overlooked methodological limitation concerns the timing of endometrial sampling and the temporal dynamics of microbiota restoration following intervention. None of the included studies performed paired pre‐ and posttreatment endometrial sampling or analyzed microbiota composition immediately before ET as a true internal control. Consequently, it remains uncertain whether the observed clinical improvements reflect stable microbial restoration or merely transient shifts in bacterial composition induced by short‐term treatment. Current evidence does not define how long the endometrial microbiome requires to re‐establish a eubiotic state after antibiotic, probiotic, or nutraceutical interventions. This represents a major knowledge gap that hampers the interpretation of causal relationships between microbial modulation and implantation outcomes. Future prospective studies should therefore incorporate longitudinal sampling designs, with microbial and histologic assessments both before and after intervention, and ideally re‐evaluation prior to ET, to confirm the persistence and functional stability of microbiota normalization over time. 20 , 32 , 72 Recent research has increasingly highlighted the role of the uterine microbiota in assisted reproductive technology outcomes. In the current diagnostic setting, where standardized disease definitions, microbial thresholds, and clinically validated cutoffs are still evolving, 16S rRNA gene sequencing via next‐generation sequencing represents a practical and informative approach that has been widely adopted in both clinical and translational research. Emerging evidence suggests that endometrial microbiome profiling based on 16S rRNA sequencing can identify distinct patient subgroups and may be associated with differential pregnancy outcomes following targeted interventions, supporting its current clinical utility, even as more comprehensive metagenomics approaches are being progressively developed and implemented. 75 Studies of vaginal and endometrial microbiota alterations linked to CE 21 , 40 , 73 , 76 , 77 indicate that 16S rRNA sequencing provides a more comprehensive and nuanced microbial profile than classical diagnostic like the Nugent score. 62 However, findings remain contradictory; some studies report increased microbial diversity and evenness in CE patients, 64 whereas others observe comparable Lactobacillus ratios in MDR versus antibiotic‐sensitive cases. 63 This suggests that the abundance of Lactobacillus 61 , 63 and Burkholderia 35 remains relatively stable in infertile women, regardless of RIF diagnosis. Multiple investigations have demonstrated a predominance of Firmicutes and Lactobacillus in the cervix, contrasting with a predominance of Actinobacteriota , Proteobacteria , and Bacteroidota in the endometrium. 36 , 39 This reinforces the lack of Lactobacillus dominance in the endometrial niche. 36 , 38 , 78 , 79 Findings by Zhang et al. 64 align with those of Bednarska‐Czerwińska et al., 80 who described a dynamic microbial interplay between these niches during implantation, influenced by Escherichia coli and Gardnerella vaginalis . Studies focused on IVF patients 81 revealed physiological and pathological microbiota shifts consistent with prior reports. 39 , 73 Cumulative evidence supports Zhang et al.'s 64 identification of bacterial taxa associated with miscarriage, such as Phyllobacterium , Gardnerella , Enterococcus , and Pseudomonas , and the enrichment of genera like Prevotella , Streptococcus , and Romboutsia in CE cases with poor pregnancy prognosis. 23 , 40 , 82 Notably, Enterococcus and Streptococcus have been implicated in CE pathogenesis. 83 Despite advances, the relationship between a non‐ Lactobacillus ‐dominant microbiota and CE as a precursor to RIF remains complex. 20 , 25 Nonetheless, eubiosis in the vaginal microbiota is characterized by Lactobacillus dominance, 36 , 84 , 85 , 86 which correlates with improved live birth rates in RIF patients undergoing ET. 32 , 33 , 34 , 74 , 87 , 88 , 89 , 90 Conversely, the persistence of non‐ Lactobacillus ‐dominant microbiota‐associated CE reflects a shift in bacterial taxa that may trigger localized inflammation, immune dysregulation, and metabolic disturbances, 23 , 39 all of which adversely affect implantation success. This is significant because inflammatory mediators must be tightly regulated during blastocyst adhesion to the endometrial epithelial wall. 91 These findings underscore the mechanistic role of microbiota in mucosal cell morphology changes relevant to decidualization, 92 and implantation, 93 while commensal microbes restrict pathogenic proliferation. 33 Reported concordance between vaginal and endometrial cultures ranges from 32.6% to 50.2%, indicating variability that challenges the role of Lactobacillus in the endometrium. 32 , 36 , 94 These discrepancies may result from contamination risk along the reproductive tract, where bacterial loads are generally low, 36 and from differences in sampling devices and methodologies. 76 , 95 This highlights the need for optimized, standardized approaches. 72 The lack of consistency in the diagnostic definition of CE across studies remains a fundamental methodological challenge with direct implications for both clinical management and research outcomes. Although most authors rely on CD138‐based IHC to identify stromal plasma cells, no universal threshold or consensus standard currently exists: quantitative cutoffs range from categorical presence to ≥4 CD138 + cells per HPF, or composite indices such as endometrial stromal plasmacyte density index, and are often supplemented or substituted by hysteroscopic descriptors (stromal edema, hyperemia, micropolyps). Recent efforts toward standardization, including unified hysteroscopic criteria, 67 and histologic proposals linked to outcomes, 68 alongside conceptual updates on CE diagnosis and management, 20 support an integrated diagnostic framework that combines histology/IHC with hysteroscopy; however, implementation remains inconsistent. Standardizing CE criteria is essential before microbiota‐targeted interventions can be reliably compared across trials, as differences in plasma‐cell thresholds, sampling frames, and adjunctive assessments directly affect case classification and therapeutic inference. 20 , 67 , 68 Moreover, molecular studies show that some CE‐diagnosed patients exhibit a Lactobacillus ‐dominant endometrial microbiota, raising uncertainty as to whether such cases represent true inflammatory CE or a low‐grade, respectively, resolving state within an otherwise balanced ecosystem. These findings highlight the need to complement histopathology with molecular and immune biomarkers to refine disease boundaries, distinguish infection from dysbiosis, and identify CE phenotypes most likely to benefit from microbiota‐targeted therapies. 32 , 36 , 72 Non‐ Lactobacillus ‐dominant microbiota is not exclusive to RIF but is prevalent among infertile women generally. To mitigate the risk of MDR bacteria associated with CE, prolonged broad‐spectrum antibiotic use should be avoided in favor of antibiogram‐guided therapies. 96 CE prevalence has remained stable over the past decade, between 31.4% 63 and 33.7%. 17 , 97 However, doxycycline efficacy has declined due to rising MDR rates, from 7.7% 17 to 7.8% and 21.2%. 63 Cicinelli et al. 15 demonstrated improved pregnancy outcomes following a 14‐day antibiotic course of doxycycline and ciprofloxacin targeting hysteroscopic CE findings. 14 , 17 , 98 Bouet et al. 16 reported synergistic effects of metronidazole and moxifloxacin as a second‐line treatment for doxycycline‐resistant CE, although reproductive outcomes were not detailed. Conversely, probiotic regimes lasting 7–8 days markedly increased Lactobacillus spp. abundance without reliance on broad‐spectrum antibiotics. 62 It is also important to acknowledge that the observed improvements in pregnancy and live birth outcomes across studies may be partially influenced by the heterogeneity of microbiota‐targeted therapies. Antibiotics, probiotics, and nutraceuticals do not share identical biological targets or therapeutic mechanisms. Antibiotic regimens primarily address infectious or dysbiotic inflammation, whereas probiotics and nutraceuticals such as lactoferrin are aimed at restoring Lactobacillus dominance and modulating local immunity. Consequently, the pooled reproductive outcomes should be viewed as a composite reflection of multiple, nonequivalent therapeutic modalities. Future trials should therefore stratify analyses by intervention type to delineate specific efficacy profiles and mechanistic contributions. Emerging evidence highlights the potential interplay between gastrointestinal health and endometrial function, particularly via systemic inflammation. Increased intestinal permeability, commonly referred to as “leaky gut” has been associated with chronic low‐grade inflammation, which may impair ER and contribute to implantation failure and miscarriage. 99 This inflammatory cascade may exacerbate CE or disrupt local immune tolerance, suggesting that gut‐endometrial cross‐talk warrants further study. Additionally, recent 16S rRNA gene‐sequencing studies have reported significant differences in microbial composition between RIF patients with and without CE, with increased microbial diversity and specific bacterial taxa associated with miscarriage and nonpregnancy outcomes. 64 Another study confirmed distinct endometrial microbiota patterns in CE‐positive versus CE‐negative women, further reinforcing the association between dysbiosis and reproductive failure. 100 These findings underscore the need for integrated diagnostic frameworks and mechanistic studies to better understand how microbial imbalances contribute to infertility. This systematic review has several notable strengths. It is the first to focus specifically on prospective cohort studies investigating microbiota‐targeted therapies in women with CE and RIF, a population with complex and often refractory reproductive challenges. The review applied validated methodological tools, particularly NOS and ROBINS‐I, to evaluate study quality and risk of bias in a structured and transparent manner. By systematically reporting on a wide range of pregnancy‐related outcomes, including clinical pregnancy, ongoing pregnancy, live birth, miscarriage, biochemical pregnancy, multiple pregnancy, and stillbirth, this review offers a comprehensive overview of the therapeutic landscape. Furthermore, the narrative synthesis approach was particularly well suited to the heterogeneity of the included studies, allowing for nuanced interpretation across varying diagnostic criteria, treatment strategies, and microbiota profiling methods. However, several limitations must be acknowledged. A quantitative meta‐analysis could not be performed due to substantial methodological and clinical heterogeneity across the four included studies. This small number of studies, some of which were exploratory or pilot in nature, limited the statistical power and robustness of conclusions. Three of the four studies were rated as having moderate to serious or even critical risk of bias, largely due to inadequate adjustment for confounding factors and incomplete follow‐up data. Additionally, the generalizability of findings is restricted by the fact that most studies were conducted by closely affiliated research groups in East Asia, with limited ethnic and geographical diversity. Finally, the inconsistent application of diagnostic criteria for both CE and RIF across studies may have introduced misclassification bias, further complicating comparisons and limiting the strength of the evidence base. Therefore, these results should be interpreted cautiously, and future well‐designed trials are needed to compare individual intervention types under standardized diagnostic and outcome criteria.

This systematic review underscores the growing clinical relevance of personalized therapeutic strategies targeting the endometrial microbiota to improve IVF‐ET outcomes in women with CE‐associated RIF. Despite the limited number and substantial heterogeneity of available studies, preliminary evidence suggests that microbiota modulation through antibiotics, probiotics, or nutraceutical supplementation may improve clinical pregnancy and, in some cases, live birth rates in this challenging population. However, marked methodological variability across studies, including inconsistent definitions of CE and RIF, heterogeneous diagnostic thresholds, diverse treatment regimens, and nonuniform microbiota assessment techniques, precludes quantitative meta‐analytic synthesis, and limits the certainty of current conclusions. This is further reflected in the risk of bias evaluations, with ROBINS‐I indicating predominantly “Moderate” to “Critical” concerns, particularly related to confounding and outcome reporting, and NOS assessments revealing variable methodological quality. Although randomized controlled trials are traditionally regarded as the highest level of evidence, their implementation in infertility research, especially when involving ET poses substantial ethical and practical challenges, as withholding or delaying potentially beneficial interventions may not be feasible. Accordingly, future research in this field should prioritize methodologically robust and ethically feasible alternatives, including well‐designed prospective cohort studies, pragmatic or stepped‐wedge trial frameworks embedded within routine clinical care, and standardized diagnostic‐therapeutic algorithms that allow systematic evaluation of microbiota‐targeted interventions without denying treatment. Future investigations should focus on harmonizing diagnostic criteria for CE and RIF, standardizing the timing and methodology of endometrial assessment, and defining clinically meaningful reproductive endpoints. Longitudinal designs incorporating follow‐up sampling may further help clarify causal relationships and treatment responsiveness. In parallel, mechanistic studies integrating microbiological, immunological, and clinical data are needed to elucidate the pathways linking endometrial dysbiosis to impaired implantation and to identify reliable biomarkers predictive of therapeutic response. Ultimately, by critically synthesizing the available evidence and outlining realistic pathways for future research, this review highlights both the clinical potential and the current limitations of microbiota‐focused interventions and provides a structured framework to support the translation of emerging endometrial microbiome science into personalized fertility care.

Infertility, as defined by the World Health Organization (WHO), is the inability to achieve pregnancy after 12 months of regular, unprotected sexual intercourse. 1 , 2 It affects millions worldwide and is recognized as a major public health concern, with estimated prevalence ranging from 8% to 12% in the general population and up to 17.5% in specific subgroups. 3 , 4 , 5 The advent of assisted reproductive technologies has transformed infertility treatment; yet, despite technological advancements, a significant proportion of patients experience suboptimal outcomes. 6 Among these challenges is RIF, which affects approximately 8–10% of women undergoing IVF with intracytoplasmic sperm injection (ICSI). RIF is a diagnosis applied only after comprehensive evaluation and exclusion of uterine abnormalities and other contributing factors. It is characterized by the failure to achieve sustained implantation following at least three consecutive transfers of morphologically high‐quality, euploid blastocysts or an equivalent number of unscreened embryos, adjusted for maternal age and expected aneuploidy rates. 7 , 8 , 9 The etiology of RIF is multifactorial, encompassing not only embryonic competence but also the intricate synchronization between a competent blastocyst and a receptive endometrium. 10 Other contributing elements include individualized patient management and avoidance of non‐evidence‐based interventions. 4 , 11 , 12 , 13 Emerging diagnostic insights have identified CE, a persistent, often asymptomatic low‐grade inflammatory condition frequently linked to microbial dysbiosis, as a potentially underrecognized factor in RIF pathophysiology. 14 , 15 , 16 , 17 , 18 , 19 , 20 CE is predominantly diagnosed histologically through the presence of plasma cells in the endometrial stroma, identified by CD138 immunostaining. 20 It is increasingly reported in women with unexplained reproductive failure and has been associated with a range of adverse outcomes, including implantation failure and miscarriage. 15 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 Antibiotic treatment of CE has shown promise in restoring endometrial receptivity (ER) and improving fertility outcomes. 17 , 28 However, indiscriminate or prolonged antibiotic use risks disrupting the symbiotic balance of the endometrial microbiota. Despite the growing body of literature exploring the role of the uterine microbiome in embryo implantation and pregnancy success, significant uncertainty remains regarding the influence of non‐ Lactobacillus ‐dominant microbial communities and specific pathogens. 23 , 29 , 30 , 31 Current evidence suggests that a Lactobacillus ‐dominated endometrial environment correlates positively with higher implantation, pregnancy, and live birth rates, 30 , 32 , 33 , 34 , 35 yet a comprehensive understanding of the native endometrial microbiota is still evolving. 36 , 37 , 38 In pathological conditions such as CE, microbial imbalance, characterized by reduced Lactobacillus abundance and proliferation of potentially pathogenic species, has been documented and is postulated to impair ER. 23 , 39 , 40 Accordingly, this systematic review aims to determine whether targeted interventions intended to restore a Lactobacillus ‐dominant endometrial microbiota improve reproductive outcomes in women diagnosed with CE‐associated RIF. Specifically, we assessed clinical, ongoing, and biochemical pregnancy rates, live birth, miscarriage, multiple pregnancy, and stillbirth following various interventions including antibiotics, probiotics, and nutraceuticals.

The authors confirm that there is no conflict of interest that could be perceived as prejudicing the impartiality of the research reported as it was carried out without any commercial or financial ties.

This study was designed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) 2020 guidelines Checklist Protocol, 41 thereby ensuring methodological transparency. In line with these standards, the protocol was prospectively registered in the Open Science Framework (OSF) 42 ( https://doi.org/10.17605/OSF.IO/63FHC ). This systematic review did not require prior approval from an Institutional Review Board (IRB), nor expert evaluation or third‐party endorsement, as the research data extracted for the established endpoints were previously published in peer‐reviewed scientific journals. An extensive electronic search was conducted across several major biomedical bibliographic databases, including PubMed‐MEDLINE–United States National Library of Medicine (NLM, 1996), Web of Science™ (WOS) (Clarivate Analytics, 1997), Scopus (Elsevier, 2004), Excerpta Medica dataBASE (EMBASE) (Elsevier, 1947), and the Cochrane Central Register of Controlled Trials (CENTRAL) (Cochrane Library, 1993). These repositories were selected based on scientometric assessments, attributive to the subsequent bibliometric and altmetric analyses, 43 , 44 , 45 , 46 , 47 , 48 , 49 , 50 , 51 revealing quantitative and qualitative output differences between platforms. 52 Customized search strings were employed to identify, rank, and evaluate relevant studies, spanning the period from January 1, 2015 to January 31, 2025. Queries were formulated using MeSH (Medical Subject Headings) and Emtree terms. “Microbiota” (2014) and other Entry Terms were listed as Major Topic [Majr] with the following credentials: Tree Number(s): G06.591, G16.500.275.157.049.100.500, N06.230.124.049.100.500, MeSH Unique ID: D064307 in parallel with “Virome” (2021)—Tree Number(s): G06.591.968, G16.500.275.157.049.100.500.968, N06.230.124.049.100.500.937, MeSH Unique ID: D000083422 and “Mycobiome” (2017)—Tree Number(s): G06.591.875, G16.500.275.157.049.100.500.875, N06.230.124.049.100.500.750, MeSH Unique ID: D000072761. In parallel, “Endometrium”—Tree Number(s): A05.360.319.679.490, MeSH Unique ID: D004717, along with non‐indexed but scientifically recognized terms such as “Repeated Implantation Failure” or “Recurrent Implantation Failure,” and “Chronic Endometritis” were integrated as primary free‐text keywords. To further optimize retrieval breadth, thesaurus dictionaries were used to support synonym expansion and enhance term variability (accessed on March 15, 2025). Boolean operators such as “AND” or “OR” were applied to formulate a comprehensive primary search string (Search #1), enabling broad coverage of the targeted concepts. Additionally, thematic clusters were created for a secondary, more focused query approach (Search #2), allowing tailored refinement across specific subtopics. The complete set of search strategies, including database‐specific syntax variation, is provided in File S1 . Potentially eligible references identified during the screening phase were imported into Mendeley—Reference Management Software (v.1.19.8) (Elsevier, 2013) due to its relatively high accuracy in duplicate detection (93%). Mendeley's sensitivity and specificity rates are comparable to those of Ovid/Rayyan (97%) and Covidence (96%), and it outperforms EndNote desktop X9 (76%) and Zotero (80%) in terms of management efficiency. 53 Although newer tools such as Deduklick and ASySD have reported precision levels approaching 99%–100%, Rayyan remains notable for its sensitivity that spans from 99% to 100% and a lower false‐negative duplicate count: 36 for Mendeley compared to EndNote's 258, and by reference to those false positives ranging from 0 (OVID) to 43 (EBSCO). 53 , 54 , 55 To ensure accuracy, the “Check for Duplicates” function was used to de‐duplicate the citations, followed by an additional manual verification of records for consistency and eliminate residual redundancies. The titles ± abstracts were independently screened by each reviewer for relevance, while the full‐texts of articles meeting the inclusion criteria were assessed by BD, O‐DI, and CI. Any discrepancies or conflicting assessments were resolved through consensus‐based discussion among all authors. A tabular overview of all retrieved entries is provided in File S2 . The main objective of this study was to evaluate whether microbiota‐targeted therapeutic interventions improve pregnancy outcomes in women with CE‐associated RIF undergoing IVF‐ET. This investigation was guided by a Population (P), Intervention (I), Comparison (C), Outcome (O), and Study Design (PICOS) framework, as detailed in File S3 . Standard methodological information from the included studies was organized in a structured tabular form using Microsoft Excel 2010 (Microsoft Corporation, Redmond, WA, USA) to facilitate retrieval and storage, otherwise used for sorting and coding raw data. This approach carried out by BD, O‐DI, A‐MD(C), IS, GL, and MD included the following parameters detailed in the Data Extraction section ( File S3 ). When a study had multiple publications, the authors combined these reports to focus on the study as a whole instead of addressing each report individually. Therefore, each study was assigned a single study ID that encompassed multiple references. 56 Methodological characteristics were primarily reported as means ± standard deviations for continuous variables or medians with interquartile ranges where appropriate, and as absolute numbers and percentages (%) for categorical variables. Key outcome parameters were also summarized using these formats. The quality of each included study was independently evaluated by BD and O‐DI using the ROBINS‐I (v.2—November 22, 2024), 57 which can be accessed via https://sites.google.com/site/riskofbiastool/ (accessed on March 15, 2025). This evaluation involved providing, making judgments, and utilizing different packages based on the type of study. The tool offers a framework that categorizes biases into seven distinct domains: (1) due to confounding, (2) in selection of participants into the study (or into the analysis), (3) in classification of interventions, (4) due to deviations from intended interventions, (5) due to missing data, (6) in measurement of outcomes, and (7) in selection of the reported result. Each category relies on prompt question(s) whose responses vary from “yes,” “probably yes,” “probably no,” “no,” or “no information,” and calculating the overall risk may be categorized as “low risk of bias,” “moderate risk of bias,” “serious risk of bias,” and “critical risk of bias.” Notably, a judgment of “moderate risk of bias” across multiple domains was regarded as indicating a “serious risk” of overall bias. 57 , 58 For transparency of the ROBINS‐I evaluation, we employed the Risk‐of‐bias VISualization 59 in the manuscript. Additionally, BD and O‐DI independently assessed the quality of the studies using the NOS 60 due to the nonrandomized design of the included manuscripts. This tool evaluates three methodological categories: Selection (score range 0–4), Comparability (score range 0–2), and Outcome (score range 0–3). Based on the number of asterisks assigned to each category, the quality of the studies was classified as high (score of 7–9), intermediate (score of 4–6), or low (score of 0–3). To be deemed eligible, manuscripts had to contain original data and be written exclusively in English, regardless of adherence to the IMRaD format. Additionally, they must be published in peer‐reviewed journals and report experiments conducted solely on human subjects. Any other type of manuscript or analysis was automatically excluded.

File S1. Detailed searching strategies. File S2. Overall tabular number of entries per year of publication and database. File S3. PICOS approach. File S4. Eligibility assessment phases.