Prediction of endometrial malignancy in women with postmenopausal bleeding and sonographic endometrial thickness ≥ 4.5 mm

Abstract

To build mathematical models for evaluating the individual risk of endometrial malignancy in women with postmenopausal bleeding and sonographic endometrial thickness ≥ 4.5 mm using clinical data, sonographic endometrial thickness and power Doppler ultrasound findings. Of 729 consecutive patients with postmenopausal bleeding, 261 with sonographic endometrial thickness ≥ 4.5 mm and no fluid in the uterine cavity were included. They underwent transvaginal two-dimensional gray-scale and power Doppler ultrasound examination of the endometrium. The ultrasound image showing the most vascularized section through the endometrium as assessed by power Doppler was frozen, the endometrium was outlined and the percentage vascularized area (vascularity index) was calculated using computer software. The ultrasound examiner also estimated the color content of the endometrial scan on a visual analog scale (VAS) graded from 0 to 100 (VAS score). A structured history was taken to collect clinical information. Multivariate logistic regression analysis was used to create mathematical models to predict endometrial malignancy. There were 63 (24%) malignant and 198 (76%) benign endometria. Women with a malignant endometrium were older (median age 74 vs. 65 years; P = 0.0005) and fewer used hormone replacement therapy and warfarin. Women with a malignant endometrium had a thicker endometrium (median thickness 20.8 vs. 10.2 mm; P = 0.0005) and higher values for vascularity index and VAS score. When using only clinical data to build a model for estimating the risk of endometrial malignancy, a model including the variables age, use of warfarin and use of hormone replacement therapy had the largest area under the receiver-operating characteristics curve (AUC), with a value of 0.74 (95% confidence interval (CI), 0.67-0.81). A model including age, use of warfarin and endometrial thickness had an AUC of 0.82 (95% CI, 0.76-0.87), and one including age, use of hormone replacement therapy, endometrial thickness and vascularity index had an AUC of 0.91 (95% CI, 0.87-0.95). Using a risk cut-off of 11%, the latter model had sensitivity 90%, specificity 71%, positive likelihood ratio 3.14 and negative likelihood ratio 0.13. The diagnostic performance of models predicting endometrial cancer increases substantially when sonographic endometrial thickness and power Doppler information are added to clinical variables. The models are likely to be clinically useful but need to be prospectively validated.