Optimization of methylated DNA markers to rule out endometrial cancer in patients with abnormal uterine bleeding.

INTRODUCTION

While only 5-10% of postmenopausal bleeding (PMB) and perimenopausal abnormal uterine bleeding (AUB) represents an underlying endometrial cancer (EC) or precancerous lesion, the current standard of care recommends endometrial sampling to evaluate all presenting with this clinical concern. We aimed to develop a highly sensitive panel of methylated DNA markers (MDMs) to identify patients with underlying EC through less invasive testing of self-collected vaginal fluid.

METHODS

Patients with AUB/PMB ≥45 years or EC or atypical endometrial hyperplasia (AEH) ≥18 years were prospectively enrolled to self-collect vaginal fluid using a tampon. Bisulfite-converted DNA from vaginal fluid was assayed with long-probe quantitative amplified signal (LQAS) for the quantification of MDM signal. Samples were tested with the goal of building a parsimonious model to discriminate EC from benign endometrium (BE). This marker reduction study started with 19 previously identified MDMs. Down-selection was then performed on a panel of 12 MDMs in an independent marker selection study. A random forest model was developed and validated in an independent test set.

RESULTS

In marker reduction, a 19-MDM model achieved a sensitivity of 91% (83-96%) in EC with a specificity of 87% (80-92%) observed in BE in an independent set. In marker selection, a 2-MDM model generated a sensitivity for EC of 96% (84-99%), specificity of 82% (67-91%), and an AUC of 0.97 (0.94-1) in an independent test set.

CONCLUSIONS

Rigorous marker reduction from 19 MDMs to 12 allowed for marker selection of a 2-MDM panel with high sensitivity and specificity to detect EC in vaginal fluid.