Liquid Biopsy Cell-free RNA-based Machine Learning Enables Preoperative Risk-Stratification of Uterine Leiomyosarcoma

Area of Science:

• Gynecologic Oncology
• Molecular Diagnostics
• Machine Learning in Medicine

Background:

• Distinguishing uterine leiomyoma (UM) from uterine leiomyosarcoma (UMS) preoperatively is challenging, impacting surgical decisions and patient outcomes.
• Current diagnostic methods like imaging and biomarkers lack sufficient accuracy for reliable differentiation.
• Plasma circulating cell-free RNA (cfRNA) offers a potential noninvasive biomarker for tumor classification.

Purpose of the Study:

• To develop and evaluate a plasma cfRNA-based machine learning classifier for preoperative differentiation of UMS from UM.
• To assess the diagnostic performance and biological generalizability of the developed classifier.

Main Methods:

• Prospective, multicenter study involving 102 women with suspected myometrial tumors.
• Plasma cfRNA profiles analyzed to train a 100-gene signature machine learning model.
• Diagnostic performance evaluated using AUROC, sensitivity, and specificity, with cross-validation and external cohort testing.

Main Results:

• A 100-gene cfRNA signature reliably differentiated UM from UMS with an AUROC of 0.868 (sensitivity 0.732, specificity 0.813).
• Performance was consistent across different age groups and robust in an independent external tissue cohort (AUROC 0.892).
• The classifier demonstrated biological and translational robustness.

Conclusions:

• Plasma cfRNA profiles show distinct differences between UMS and UM, suggesting potential for preoperative risk stratification.
• Further prospective validation is needed to establish an accurate diagnostic test.
• The positive predictive value of the test will be influenced by population prevalence.