利用尿液细胞外囊泡中的miRNA对妇科肿瘤进行无创筛查

Background: Gynecologic malignancies constitute the second leading cause of cancer-related morbidity and mortality among women worldwide, following breast cancer. Despite the availability of effective screening programs, participation remains limited in several regions because of psychological reluctance toward pelvic examinations (e.g., approximately 40% in Japan) and constrained clinical resources. To address this issue, we explored a non-invasive screening strategy based on comprehensive microRNA (miRNA) profiling of urinary extracellular vesicles (uEVs). Methods: This study enrolled both pregnant and non-pregnant women with gynecologic diseases, along with healthy counterparts. In total, 456 urine samples were collected, from which uEVs were isolated and subjected to comprehensive miRNA profiling. A subset comprising 121 disease cases (84 malignant and 37 benign tumors) and 121 age-matched healthy non-pregnant controls was used to establish a diagnostic model. The dataset was randomly divided into training (N = 90) and holdout (N = 31) sets for performance evaluation. The resulting model was subsequently applied to pregnant women with gynecologic diseases and healthy pregnant women to assess its generalizability and diagnostic performance. Results: Differential expression analysis in the training set between disease cases and healthy controls identified 25 miRNAs with significant expression changes. A diagnostic model constructed using these differentially expressed miRNAs achieved an AUC of 0.907, with sensitivity and specificity of 0.867 and 0.856, respectively. When applied to the holdout set, the model maintained high performance (AUC = 0.937; sensitivity = 0.889; specificity = 0.944). Both malignant and benign tumors were detected with high scores in non-pregnant women, irrespective of cancer type or disease stage. Although healthy pregnant women, who were not included in model training, showed low predicted cancer probabilities, pregnant women with gynecologic diseases exhibited slightly reduced scores compared with their non-pregnant counterparts. Conclusions: Our results suggest that urinary uEV-miRNA profiling combined with machine learning enables accurate screening of gynecologic diseases. This non-invasive approach may complement general health checkups and enhance gynecologic screening rates, thereby contributing to earlier diagnosis and improved outcomes.