Performance of multi-parameter differential diagnostic models for ovarian cancer screening in women with adnexal masses.

Abstract

Objective: To enhance ovarian cancer diagnosis by establishing multi-parameter differential diagnostic models, combining preoperative blood tests and ultrasonic examination.

Methods: This was a retrospective cohort study. Adult women with adnexal lesions who underwent surgery between January 2013 and December 2019 were included in the study. We performed various blood tests (including routine blood work, biochemistry, thyroid hormone, reproductive hormones, and tumor markers) and ultrasonic examination on 2,929 participants. Based on these clinical features, diagnostic models were constructed and validated using the eXtreme Gradient Boosting (XGBoost) algorithm and logistic regression.

Results: Among the 2,929 participants with adnexal masses, 2,117 participants had benign conditions (574 ovarian cysts, 1,376 ovarian tumors, and 167 ovarian endometriomas), 311 had borderline ovarian tumors, and 501 had invasive ovarian cancer (413 epithelial and 88 non-epithelial types). Initially, full-feature modeling achieves area under the curve (AUC) values of 1.00 and 0.97 for the diagnosis of epithelial ovarian cancer in the training and validation sets, respectively, using a set of 20 feature parameters. Similarly, a set of 30 feature parameters yield AUCs of 1.00 (training) and 0.97 (validation) for the diagnosis of invasive ovarian cancer. XGBoost outperformed other modeling algorithms in predictive values. The differential diagnostic models constructed from multiple feature parameters exhibited better calibration than the model based on a single parameter of cancer antigen 125.

Conclusion: This study established effective multi-parameter differential diagnostic models to differentiate ovarian cancer from adnexal masses, demonstrating their potential to enhance diagnostic accuracy.