Related Experiment Video
Updated: Mar 22, 2026
An R-Based Landscape Validation of a Competing Risk Model
Published on: September 16, 2022
Prospective temporal validation of mathematical models to calculate risk of endometrial malignancy in patients with
P Sladkevicius1, G Opolskiene1,2, L Valentin1
1Department of Obstetrics and Gynecology, Skåne University Hospital Malmö, Lund University, Malmö, Sweden.
Objectives:
To validate prospectively five mathematical models published in 2007 for calculating the risk of endometrial malignancy in a defined high-risk group of patients with postmenopausal bleeding and sonographic endometrial thickness ≥ 4.5 mm.
Methods:
Of 1012 consecutive patients, 379 fulfilled our inclusion criteria, which were the same as those of the original study in which the models were created (endometrial thickness ≥ 4.5 mm, no fluid in the uterine cavity, detectable Doppler signals in the endometrium). A standardized history was taken, and clinical and transvaginal grayscale and power Doppler ultrasound examinations were performed following the study protocol. All data were collected prospectively and the five models were applied prospectively to the study patients' data to assess their risk of endometrial malignancy. Using the histological diagnosis of the endometrium as gold standard, we calculated the area under the receiver-operating characteristics curve (AUC), and sensitivity, specificity and likelihood ratios when using the same cut-offs as in the original study, for each of the five models.
Results:
Ninety-three (25%) patients had malignant endometrium. The performance of the models was similar to that in the original study, with AUCs ranging from 0.86 to 0.90. The model with the best diagnostic performance included endometrial thickness, heterogeneous endometrial echogenicity and areas of densely packed vessels on power Doppler (AUC, 0.90; sensitivity, 81%; specificity, 84% at preselected cut-off). The models were well calibrated.
Conclusions:
On temporal validation, the five models for calculating the risk of endometrial malignancy in a defined high-risk group of patients retained their good diagnostic performance and were well calibrated. The models make it possible to reclassify high-risk patients as having a low or relatively low risk, moderately high risk or very high risk of endometrial cancer, and so can be used for individualized patient management. Copyright © 2016 ISUOG. Published by John Wiley & Sons Ltd.
