Prospective validation of two mathematical models to calculate the risk of endometrial malignancy in patients with postmenopausal bleeding and sonographic endometrial thickness ≥4.5 mm

• Two models to calculate the risk of endometrial cancer have good diagnostic performance.
• The models are well calibrated especially for the lowest and highest risks.
• The models are useful to triage timing of invasive diagnostic procedures.

AimTo prospectively validate two mathematical models for calculating the likelihood of endometrial malignancy in patients with postmenopausal bleeding (PMPB), sonographic endometrial thickness (ET) ≥4.5 mm and no fluid in the uterine cavity.MethodsThis is a prospective observational diagnostic validation study performed in a PMPB clinic in a university hospital. Of 860 consecutive patients, 350 fulfilled our inclusion criteria. A standardized history was taken, clinical and transvaginal grey scale and power Doppler ultrasound examinations were performed following a research protocol. The percentage vascularized area of the endometrium at power Doppler examination (VI) was calculated using computer software. The colour content of the endometrial scan was estimated subjectively on a visual analogue scale (VAS). Gold standard was the histological diagnosis of the endometrium. Main outcome measures were area under the receiver operating characteristic curve (AUC), sensitivity and specificity when using the cut-offs suggested in the original study, and calibration curves.ResultsEighty (23%) patients had malignant endometrium. The performance of the models was similar to that in the original study. The model including patient's age, use of hormone therapy, ET and VAS performed best (AUC 0.91; 95% confidence interval [CI] 0.87–0.95; sensitivity 70%, specificity 93%). The model including ET, VI, patient's age and hormone therapy use had AUC 0.89 (95% CI 0.84–0.93; sensitivity 79%; specificity 81%). ET had AUC 0.83 (95% CI 0.78–0.88). The models were reasonably well calibrated.ConclusionOn prospective validation both models retained their diagnostic performance. This suggests that they are robust and potentially clinically useful for individualized patient management.