Logistic Regression Model to Distinguish Between the Benign and Malignant Adnexal Mass Before Surgery: A Multicenter Study by the International Ovarian Tumor Analysis Group

TLDR

The study aims to develop and prospectively validate a logistic regression model that distinguishes malignant from benign adnexal tumors before surgery. Data from 1,066 patients across nine European centers were used, with more than 50 clinical and sonographic endpoints recorded and histologic classification as the outcome. The 12‑variable model achieved AUCs of 0.95 in the development set and 0.94 in the test set, with a 0.10 cutoff yielding 93% sensitivity and 76% specificity.

Abstract

To collect data for the development of a more universally useful logistic regression model to distinguish between a malignant and benign adnexal tumor before surgery.Patients had at least one persistent mass. More than 50 clinical and sonographic end points were defined and recorded for analysis. The outcome measure was the histologic classification of excised tissues as malignant or benign.Data from 1,066 patients recruited from nine European centers were included in the analysis; 800 patients (75%) had benign tumors and 266 (25%) had malignant tumors. The most useful independent prognostic variables for the logistic regression model were as follows: (1) personal history of ovarian cancer, (2) hormonal therapy, (3) age, (4) maximum diameter of lesion, (5) pain, (6) ascites, (7) blood flow within a solid papillary projection, (8) presence of an entirely solid tumor, (9) maximal diameter of solid component, (10) irregular internal cyst walls, (11) acoustic shadows, and (12) a color score of intratumoral blood flow. The model containing all 12 variables (M1) gave an area under the receiver operating characteristic curve of 0.95 for the development data set (n = 754 patients). The corresponding value for the test data set (n = 312 patients) was 0.94; and a probability cutoff value of .10 gave a sensitivity of 93% and a specificity of 76%.Because the model was constructed from multicenter data, it is more likely to be generally applicable. The effectiveness of the model will be tested prospectively at different centers.