Liquid biopsy cell-free RNA-based machine learning enables preoperative risk-stratification of uterine leiomyosarcoma.

BACKGROUND

Accurate preoperative distinction between uterine leiomyoma and uterine leiomyosarcoma remains a major clinical challenge. Misclassification can lead to inadvertent dissemination of occult malignancy during minimally invasive procedures, while cautious management increases the use of more invasive surgery with greater morbidity. Current diagnostic approaches, including imaging and serum biomarkers, lack sufficient accuracy and standardized criteria. Circulating cell-free RNA in plasma represents a promising alternative for noninvasive tumor classification, but its clinical utility for uterine leiomyosarcoma has not been established.

OBJECTIVE

To develop a plasma-based circulating cell-free RNA machine learning classifier for the preoperative differentiation of uterine leiomyosarcoma from uterine leiomyoma, and to evaluate its diagnostic performance and biological generalizability.

STUDY

DESIGN

This prospective, multicenter study enrolled women undergoing surgery for suspected myometrial tumors at 16 hospitals in Spain. Peripheral blood was collected immediately before surgery, and postoperative histopathology served as the reference standard. The final cohort included 102 patients (78 uterine leiomyoma; 24 uterine leiomyosarcoma). Plasma cell-free RNA profiles were used to train and evaluate a model consisting of a 100-gene signature. Diagnostic performance was assessed using area under the receiver operating characteristic curve, sensitivity, specificity, positive predictive value, and negative predictive value. Age effects and cross-tissue generalizability were also evaluated.

RESULTS

A 100-gene circulating cell-free RNA signature reliably differentiates uterine leiomyoma from uterine leiomyosarcoma before surgery. This regularized logistic regression classifier achieved an area under the receiver operating characteristic curve of 0.868, with a sensitivity of 0.732, and a specificity of 0.813 in Monte Carlo cross-validation. Diagnostic performance remained consistent across age groups, with an area under the receiver operating characteristic curve of 0.879 for women below 55 years of age and 0.882 for women above 55 years. Notably, when evaluated in an independent external tissue cohort, the gene signature retained strong discriminatory performance, achieving an area under the receiver operating characteristic curve of 0.892, supporting its biological and translational robustness.

CONCLUSION

These preliminary findings suggest that plasma cell-free RNA signals differ between uterine leiomyosarcoma and uterine leiomyoma and may, after prospective validation in appropriately selected populations, contribute to preoperative risk stratification. They do not yet establish an accurate diagnostic test, and positive predictive value will depend on the prevalence of the population in which the assay is applied.