Logistic regression analysis of ultrasound features for predicting borderline ovarian tumours in young women aged ≤ 40 year.

Abstract

Objectives: To determine the ability of sonographic characteristics to distinguish borderline ovarian tumours (BOT) from benign and malignant tumours in young women by using logistic regression analysis.

Material and methods: 147 patients with ovarian masses were analysed retrospectively. We recorded and compared the available preoperative serum CA125 and CA199 levels, ultrasound and pathological findings from patient records to distinguish BOT from benign and malignant tumours using single-factor and multiple stepwise logistic regression analyses.

Results: Seventy-six women aged ≤ 40 years diagnosed with BOT, 31 women with malignant tumours, and 40 women with benign cystadenomas were included. The significant features identified in the single-factor analysis were CA125 and CA199 levels, tumour size, multilocularity, presence of solid components within cysts, colour Doppler flow, presence of microcystic pattern (MCP), and proportion of the maximum solid area covering < 50% of the inner surface within the cyst (p < 0.05). The latter two ultrasound features were identified as independent predictors for differentiating BOT from benign and malignant tumours in the logistic regression analysis. The area under the receiver operating curve (AUC) was 0.893 and 0.904, respectively. The corresponding sensitivity, specificity, positive predictive value, and negative predictive value were 84.2%, 89.5%, 94.1%, and 73.9%, respectively, while the corresponding values were 93.4%, 76.3%, 88.7%, and 85.3%, respectively.

Conclusions: Combining both ultrasonic features of the microcystic pattern and the proportion of the maximum solid area covering < 50% of the inner surface within the cystic region appears to be the optimal method for characterizing BOT.