A multivariate logistic regression analysis in predicting malignancy for patients with ovarian tumors

Our objective was to improve the preoperative diagnosis of ovarian malignancy using a multivariate logistic regression analysis on the basis of demographic, serologic, gray-scale morphological, and Doppler variables.

One hundred seventy-one patients with ovarian tumors (120 benign, 51 malignant including 9 tumors of low malignant potential) were studied with transvaginal B-mode, color, and pulsed Doppler ultrasonography before surgery. Based on the gray-scale ultrasound imaging, each tumor was classified as a unilocular cyst, multilocular cyst, unilocular cyst with solid parts, multilocular cyst with solid parts, or solid tumor. Intratumoral blood flow velocity waveforms were recorded on all tumors except unilocular cyst and were evaluated for resistance index (RI) and peak systolic velocity (PSV). Serum CA 125 levels were also measured.

Twenty tumors were unilocular cysts and were all benign. Seventy tumors including all unilocular cysts which showed no flows were all benign. The remaining 101 tumors (50 benign, 51 malignant including 9 tumors of low malignant potential) presented intratumoral blood flows. Univariate and multivariate logistic regression analyses were conducted to identify variables predictive of ovarian malignancy in these 101 tumors. The variables included age, menstrual state, serum CA 125 levels, B-mode classification, RI, and PSV. In univariate analysis, menopause, the positivity of CA 125 (> or = 35 U/ml), and PSV larger than or equal to 10.4 cm/s were found to be significantly associated with malignant tumors. The PSV value of 10.4 cm/s was the median in benign tumors. Multivariate analysis showed that serum CA 125 levels (> or = 35 U/ml) (P = 0.002) and PSV (> or = 10.4 cm/s) (P < 0.001) were to be independent predictors of malignancy.

These results suggest that intratumoral PSV is the strongest means of differentiating benign from malignant ovarian tumors with suspicious gray-scale ultrasonographic findings.