# Multi-Steroid Profiling and Machine Learning Reveal Androgens as Candidate Biomarkers for Endometrial Cancer Diagnosis: A Case-Control Study

## TL;DR

This study finds that combining steroid hormone levels with other biomarkers improves endometrial cancer diagnosis accuracy.

## Contribution

The novel use of multi-steroid profiling combined with machine learning to enhance EC diagnosis accuracy.

## Key findings

- EC patients had significantly higher levels of classic androgens and glucocorticoids compared to controls.
- A multivariate model combining steroids, CA-125, HE4, BMI, and parity achieved 87% AUC in distinguishing EC from benign conditions.
- HE4 was the strongest marker for lymphovascular space invasion and deep myometrial invasion.

## Abstract

Endometrial cancer (EC) is the second most common gynecologic malignancy, with its incidence rising due to demographic changes. Current diagnostic methods are invasive and insufficiently specific, highlighting the need for accurate, non-invasive biomarkers. In this study, we applied mass spectrometry-based multi-steroid profiling and machine learning to analyze systemic steroid levels—focusing on androgens, 11-oxyandrogens, glucocorticoids and mineralocorticoids—as potential diagnostic and prognostic biomarkers. Our cohort included 62 patients with EC and 70 controls with benign uterine conditions. We identified distinct steroid level alterations between cases and controls. While steroids alone had limited diagnostic and prognostic value, a multivariate model combining classic androgens, CA-125, HE4, BMI and parity achieved an AUC of 0.87, 79.1% sensitivity and 74.7% specificity in distinguishing EC from benign conditions. This model outperformed those based on CA-125, HE4 or their combination with BMI. These findings underscore the potential of integrating steroid profiling with established biomarkers to enhance EC detection.

Objective: To evaluate the diagnostic and prognostic potential of preoperative serum steroid levels in endometrial cancer (EC) alone and in combination with clinical parameters and biomarkers CA-125 and HE4. Methods: This single-center observational study included 62 patients with EC and 70 controls with benign uterine conditions who underwent surgery between June 2012 and February 2020. Preoperative serum levels of classic androgens, 11-oxyandrogens, glucocorticoids and mineralocorticoids were measured using liquid chromatography–tandem mass spectrometry (LC-MS/MS). Machine learning was used to assess their diagnostic and prognostic value alone and combined with clinical parameters and tumor biomarkers. Results: Patients with EC had significantly higher serum levels of classic androgens (androstenedione, testosterone), 11-oxyandrogens (11β-hydroxy-androstenedione, 11β-hydroxy-testosterone) and glucocorticoids (17α-hydroxy-progesterone, 11-deoxycortisol) compared to controls. While individual steroids had limited diagnostic value, a multivariate model including classic androgens, CA-125, HE4, BMI and parity achieved an AUC 0.87, 79.1% sensitivity and 74.7% specificity in distinguishing EC from benign uterine condition. This model outperformed our previously published model based on CA-125, HE4 and BMI (AUC: 0.81, p < 0.0001). Prognostically, HE4 was the strongest marker for lymphovascular space invasion (LVSI) (AUC: 0.79) and deep myometrial invasion (MI) (AUC: 0.71). Among steroids, androstenedione was the most predictive of LVSI (AUC: 0.67), while 11β-hydroxy-testosterone was the strongest predictor of deep MI (AUC: 0.64). Conclusions: Patients with EC exhibit distinct steroid hormone profiles. While steroids alone offer modest diagnostic and prognostic value, integrating them into multivariate models improves diagnostic accuracy.