# Serum miRNA-based diagnostic models for endometriosis: from discovery to validation

## TL;DR

Researchers developed a serum miRNA-based model to help diagnose endometriosis, showing moderate accuracy and potential for future clinical use.

## Contribution

A novel diagnostic model using 11 serum miRNAs was developed and validated for endometriosis detection with machine learning.

## Key findings

- A model using 11 miRNAs achieved 65.8% accuracy in distinguishing endometriosis from controls.
- A six-miRNA model showed 75.9% accuracy for identifying deep infiltrating endometriosis.
- The model had a lower accuracy (62.4%) for differentiating ovarian endometrioma from controls.

## Abstract

Can a serum miRNA signature serve as a potential diagnostic biomarker for endometriosis (END)?

A miRNA-based diagnostic model demonstrated an accuracy of 65.8% in distinguishing END patients from control subjects (CTR), demonstrating good sensitivity but limited specificity.

Existing research has examined the potential utility of circulating miRNAs as biomarkers for END diagnosis, revealing their differential expression between women with END and CTR. Nevertheless, the findings remain conflicting, and at present, neither a single miRNA nor a panel of them has yet been established as a reliable diagnostic test in clinical practice for the management of END.

We previously reported different miRNA expression patterns in serum samples from 67 END patients and 60 CTR by high-throughput RT-qPCR. In this multicenter study, a total of 364 patients with pathology-confirmed diagnosis of END or a benign non-END gynecological condition were retrospectively selected from a biobank or prospectively enrolled. The aims of the present study were to analyze, in the entire cohort of patients, a set of 23 potential diagnostic miRNAs via RT-qPCR and to create models capable of diagnosing END through cross-validated machine learning algorithms.

Total RNA was extracted from serum samples collected before surgical treatment and miRNAs were evaluated by RT-qPCR. Diagnostic models were developed using both the Random Forest and Logistic Regression algorithms. The performance assessment of the various models was derived from internal validation, using repeated cross-validation.

The most effective diagnostic model was constructed with 11 miRNAs: miR-140-3p, miR-181a-5p, miR-192-5p, miR-22-3p, miR-29a-3p, miR-30b-5p, miR-338-3p, miR-340-5p, miR-342-3p, miR-486-5p, and miR-652-3p. The diagnostic efficacy of the model was defined by an AUC of 70.4%, a sensitivity of 75.6%, a specificity of 53.5%, and an accuracy of 65.8%. The model that used six miRNAs (miR-192-5p, miR-30b-5p, miR-335-5p, miR-338-3p, miR-486-5p, miR-652-3p) was the best at identifying deep infiltrating endometriosis compared to the control group, with an AUC of 80.4% and an accuracy of 75.9%. A lower accuracy was achieved by the model differentiating ovarian endometrioma (OMA) from CTR (AUC = 65.8%; accuracy = 62.4%).

miRNA expression profiles have been deposited in NCBI’s Gene Expression Omnibus and are accessible through GEO Series accession numbers GSE279435.

Despite the internal cross-validation, the models still need to be tested on larger cohorts of prospectively enrolled patients across several centers to enhance their accuracy and robustness. This testing will also facilitate monitoring the model in a real-world setting, potentially integrating the miRNA-based model with other diagnostic tools, such as ultrasound.

If proven effective in larger cohorts, this model could serve as a tool for the diagnosis of END, thereby enhancing early identification and clinical care of this disease. Moreover, given its low false negative rate, the miRNA-based model may be useful as a screening tool to help identify patients who are likely to have END but warrant further evaluation to confirm END diagnosis.

This research was financed by the Italian Ministry of Health, grant number “LOMBARDIA ENDO-2021-12371946”, project title: FREEDOM TRIAL. The authors disclose no conflicts of interest.

N/A.