Diagnostic Performance of Micro-RNAs as Biomarkers for Endometrial Cancer: A Systematic Review and Meta-Analysis

2.1 Eligibillity criteria and PICOS strategy

This systematic review and diagnostic meta-analysis were conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines.¹⁴ The research question was formulated using the PICOS framework, as summarized in Table 1. This systematic review was registered with the International Prospective Register of Systematic Reviews (PROSPERO) under protocol number CRD420261280267.

Articles were included if they met the following criteria: (1) investigated the diagnostic role of microRNAs (miRNAs) in human subjects; (2) employed an observational study design suitable for diagnostic performance evaluation, such as case–control, cross-sectional, or cohort studies; and (3) explicitly identified endometrial cancer or endometrial carcinoma as the target condition in the title or abstract. Studies were excluded if they: (1) were not published in English; (2) were reviews, editorials, conference abstracts, or case reports; or (3) provided incomplete, insufficient, or non-extractable diagnostic data.

2.2 Information sources

Using predefined selection criteria (eligibility criteria), a pair of blinded reviewers reviewed titles, abstracts, and complete articles retrieved from PubMed/MEDLINE, ScienceDirect, Embase, and Web of Science databases to detect eligible studies. To further enhance search comprehensiveness, a snowballing approach was used to manually screen the cited references of qualifying studies and associated review articles. The ultimate literature search was completed in August 2025.

2.3 Database search procedure

A comprehensive literature search was conducted for research reports available through August 2025, using a blend of standardized subject headings and keywords. The following keywords were applied across databases:

2.4 Study of screening procedure

Two investigators separately reviewed titles and summaries to assess study inclusion. Complete articles were then assessed in detail by the same reviewers. Consensus meetings resolved any disagreements during the screening phase, and when agreement was not achieved, an additional senior reviewer was consulted to provide an independent assessment.

2.5 Data extraction process and data elements

Data were independently extracted in duplicate by two reviewers employing a prestructured extraction form derived from previous diagnostic meta-analytic studies. Discrepancies were settled via discussion with a third independent reviewer. All collected data were double-checked for accuracy before analysis.

The following data items were collected: (1) First author; (2) year the article was published; (3) country where the study was conducted; (4) research design; (5) evaluated miRNA(s); (6) direction of miRNA expression; (7) number of endometrial cancer cases and controls; (8) type of specimen; and (9) diagnostic accuracy metrics, including sensitivity, specificity, and AUC as a global accuracy index.

2.6 Study potential bias for quality evaluation

A pair of reviewers independently evaluated the methodological rigor and potential bias of every selected study by means of the QUADAS-2 instrument. This instrument evaluates bias risk and relevance across four areas: selection of patients, test under evaluation, reference standard, progression, and timing of testing.¹⁵ Disagreements were resolved through mutual agreement or discussion with an independent third reviewer.

2.7 Measures of effect and analytical approach

A diagnostic meta-analysis was conducted using Stata/BE version 18.0. Pooled estimates of sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and log diagnostic odds ratio (logDOR) with 95% confidence intervals (CIs) were calculated using a bivariate random-effects model. When not directly reported, true positive, false positive, false negative, and true negative values were reconstructed from available sensitivity and specificity data. Forest plots and summary receiver operating characteristic (SROC) curves were generated to illustrate overall diagnostic performance, including the area under the curve (AUC).

Threshold effects were assessed by examining the correlation between logit-transformed sensitivity and specificity using Spearman’s rank correlation. A significant positive correlation (p < 0.05) indicated a potential threshold effect, in which case diagnostic accuracy was summarized using the SROC curve rather than pooled sensitivity and specificity estimates.

Heterogeneity was evaluated using Cochran’s Q and the I² statistic, with I² > 50% indicating substantial heterogeneity. Subgroup analyses were performed based on endometrial cancer subtype (EEC-only vs. mixed EC), specimen type (serum, plasma, tissue), biomarker type (single miRNA vs. panel), expression pattern (upregulated vs. downregulated), and control type (healthy vs. mixed controls).

Sensitivity analyses were conducted using deviance residuals, Cook’s distance, and standardized residuals to identify influential studies. These studies were sequentially excluded to assess the stability of pooled estimates. Publication bias was assessed using Deeks’ funnel plot, with p > 0.05 indicating no significant bias. Clinical applicability was further evaluated using Fagan’s nomogram to estimate post-test probabilities.¹⁶

3.1 Studies characteristics

A total of 12 eligible articles, comprising 52 studies, were included in the final analysis, encompassing a combined population of 1,098 endometrial cancer cases and 957 controls ( Figure 1, Table 2).^17–28 The articles were published between 2012 and 2023 and were conducted across multiple regions, including China, South Korea, Italy, Poland, Serbia, and several European countries. Most articles employed a case–control design, with two articles using a multiphase case–control approach and one retrospective cohort study. The majority of cases involved endometrial carcinoma, with several articles specifically focusing on endometrioid endometrial cancer (EEC) or early-stage disease. Control groups consisted of healthy women, benign gynecological conditions, or a combination of both. Regarding biological specimens, miRNA expression was evaluated using serum, plasma, peripheral blood, serum-derived exosomes, and formalin-fixed paraffin-embedded (FFPE) tissue. Several articles have assessed circulating miRNAs, while others have investigated tissue-based miRNA expression, and one study has included both plasma and tissue samples. Overall, the included articles represented diverse populations, specimen types, and study designs, providing a broad basis for the pooled diagnostic performance and subgroup analyses.

Figure 1. PRISMA diagram illustrating study screening and inclusion.

3.2 Bias risk evaluation

Overall, the potential bias was considered acceptable across articles, although variability was observed among specific domains ( Table 3). The patient selection domain was the primary source of potential bias, as several articles employed retrospective case–control designs or non-consecutive sampling, increasing the risk of bias in this area for a proportion of articles. Conversely, the index testing domain showed a consistently low risk of bias, as all articles applied standardized, validated miRNA detection methods with clearly defined thresholds. Similarly, the reference-standard domain was deemed low-risk across all included articles, as endometrial cancer was consistently confirmed by histopathological analysis. The risk of systematic error in the flow and timing domain was mostly minimal to uncertain, largely attributable to incomplete reporting of the interval between the index test and the reference standard or sparse information on excluded patients. From an applicability perspective, concerns were largely minimal across all domains, indicating that the included studies were appropriate for addressing the review question on the ability of miRNA-derived biomarkers to aid in the diagnosis of endometrial cancer. On the whole, the QUADAS-2 assessment suggests that the findings of this meta-analysis. In general, the findings are based on published articles with acceptable methodological rigor, with patient selection as the main potential source of bias.

3.3 Overall diagnostic performance

The chi-square statistic and I² measure were implemented to quantify heterogeneity across studies. The results indicated substantial heterogeneity in the pooled sensitivity (I² = 77.30%) and pooled specificity (I² = 81.99%). Therefore, the overall estimates were derived using a random-effects model diagnostic performance calculation. The compile sensitivity of miRNA-based biomarkers for the diagnosis of endometrial cancer was 86% (95% CI: 0.82–0.89), while the aggregated specificity estimate was 0.81 (95% CI: 0.75–0.85) ( Figure 2). The combined summary positive likelihood ratio was 7.97 (95% CI: 5.83–10.11), NLR was 0.41 (95% CI: 0.30–0.52) ( Figure 3), and logDOR was 3.04 (95% CI: 2.66–3.42) ( Figure 4a and 4b). The SROC curve analysis demonstrated an AUC of 0.91 (95% CI: 0.10–1.00) ( Figure 5).

Figure 2. Forest plot of pooled sensitivity and specificity.

Figure 3. Diagnostic performance of miRNAs based on negative likelihood ratio (NLR).

Figure 4. Diagnostic performance of miRNAs based on PLR (a) and logDOR (b).

Figure 5. Summary ROC (SROC) curve of the included studies.

3.4 Sensitivity analysis and publication bias

Influence analysis using Cook’s distance identified six studies (studies 4, 11, 15, 26, 33, and 43) as potentially influential, while outlier detection based on standardized residuals revealed five studies (studies 4, 11, 15, 33, and 43) outside the reference limits ( Figure 6). After removing outlier studies, the meta-analytic summary of sensitivity and specificity were 85% (95% CI: 0.81–0.89) and 78% (95% CI: 0.73–0.82), respectively, although substantial heterogeneity persisted for both estimates (I² = 74.93% and 76.43%). The pooled NLR was 0.41 (95% CI: 0.29–9.53; I² = 0.0%, p = 0.961), and the pooled PLR was 6.84 (95% CI: 5.60–8.90; I² = 99.3%, p < 0.001). The AUC was 0.79 (95% CI: 0.76–0.82), and the meta-analytic log diagnostic odds estimate was 2.88 (95% CI: 2.58–3.18), both indicating marked between-study heterogeneity. According to Deeks’ funnel ( Figure 7), asymmetry analysis did not reveal convincing evidence of publication bias (p = 0.20).

Figure 6. Diagnostic sensitivity analyses: (a) Model goodness of fit was inspected through deviance residuals, (b) the assumption of bivariate normality was tested using Mahalanobis distance, (c) influence analysis relied on Cook’s distance, and (d) standardized residuals served to flag potential outliers.

Figure 7. Deeks’ funnel plot used for the evaluation of publication bias.

3.5 Clinical utility analysis

The practical utility of miRNA-based biomarkers was examined using Fagan nomograms, which were used to derive post-test probabilities at initial risk levels of 25%, 50%, and 75% ( Figure 8). When the prespecified pre-test probability was 25%, a positive miRNA value increased the estimated probability of endometrial cancer to 60%, while a negative result decreased it to 5%. Given an initial (pre-test) probability of 50%, a positive test yielded a post-test probability of 82%, whereas a negative test reduced this probability to 15%. At the higher pre-test risk level of 75%, a positive result value raised the post-test probability to 93%, whereas a negative result reduced it to 34%.

Figure 8. Fagan nomograms corresponding to varying pre-test probabilities: (a) 25%, (b) 50%, and (c) 75%.

3.5 Subgroup analysis

Subgroup analyses showed consistent diagnostic performance of miRNA-based biomarkers across disease subtypes, specimen types, biomarker configurations, expression directions, and control definitions ( Table 4). In the subgroup limited to EEC cases, the pooled sensitivity was 0.88 (95% CI 0.84–0.91), and the reported aggregated specificity was 0.76 (95% CI 0.68–0.83), yielding an AUC of 0.79 (95% CI 0.75–0.84). Studies including mixed histologies demonstrated comparable estimates, with a pooled sensitivity of 0.85 (95% CI: 0.78–0.90), a specificity of 0.84 (95% CI: 0.76–0.89), and an AUC of 0.80 (95% CI: 0.76–0.84). For endometrial cancer subtype, studies restricted to EEC-only reported a summary sensitivity estimate of 0.88 (95% CI: 0.84–0.91) and specificity of 0.76 (95% CI: 0.68–0.83), with an AUC of 0.79 (95% CI: 0.75–0.84). In the subgroup limited to EEC cases, the pooled sensitivity was 0.88 (95% CI: 0.84–0.91) and the pooled specificity was 0.76 (95% CI: 0.68–0.83), yielding an AUC of 0.79 (95% CI: 0.75–0.84).

By specimen type, serum-based assays exihibited greater combined sensitivity and specificity (0.91 [95% CI: 0.84–0.95] and 0.92 [95% CI: 0.62–0.99]) than plasma-based assays (sensitivity 0.71 [95% CI: 0.63–0.78]; specificity 0.81 [95% CI: 0.67–0.89]), whereas tissue-based investigations showed a pooled sensitivity of 0.82 (95% CI 0.72–0.89) and a pooled specificity of 0.80 (95% CI 0.70–0.87).

When the data were stratified according to biomarker type, miRNA panels showed superior diagnostic performance compared with individual miRNA assays, with pooled sensitivity and specificity of 0.90 (95% CI 0.84–0.94) and 0.93 (95% CI 0.87–0.97), respectively, versus 0.85 (95% CI 0.80–0.89) and 0.75 (95% CI 0.70–0.81). When stratified by expression direction, downregulated miRNAs showed higher pooled sensitivity and specificity than upregulated miRNAs. Studies using healthy controls reported higher specificity than those with mixed control populations. Marked heterogeneity was observed across most subgroups, particularly for positive indices based on likelihood ratios and diagnostic odds ratios.