Page 86 - The diagnostic work-up of women with postmenopausal bleeding

Chapter 4
Another limitation is the fact that partial verification was performed in both databases. In women with an endometrial thickness below the applied threshold no histological assessment was performed.This is in agreement with clinical policy since more than 15 years in both Sweden and the Netherlands, and for practical and ethical reasons we have not included this assessment for research purposes only. For both validation databases however, efforts were made to collect information on these women by assessing patient charts and in the Swedish database matching the patients with the regional cancer registry. Evidently, there remains some uncertainty whether indeed no endometrial cancers were missed in this group, but with our approach we minimised this risk by our follow-up efforts.
Several prediction models have been published to estimate the risk of endometrial cancer in women with postmenopausal bleeding.13 Opolskiene et al developed four prediction models including clinical and ultrasound information for women with endometrial thickness 4.5 mm.The first model is based on patient characteristics and in the development database an AUC of 0.74 was found.The second model is based on patient characteristics and endometrial thickness as measured withTVS,with an AUC of 0.82.The last two models are based on patient characteristics combined with sonographic endometrial thickness and two different Doppler characteristics, with AUC of 0.89 and 0.91 respectively. The authors concluded that the models are fairly good in excluding endometrial cancer when power Doppler is added.15 Burbos et al developed two models based on patient characteristics with (AUC 0.77) and without endometrial thickness (AUC 0.73).24 The authors concluded that the model based on patient characteristics has a reasonable discriminatory ability and the model based on patient characteristics and endometrial thickness has a fair accuracy in separating women without cancer from women with cancer.These findings are similar to the findings of internal validation of the models by Opmeer et al16 None of these models have yet been externally validated.13
Before a prediction model can be implemented in clinical practice, external validation is essential.14 In this external validation we found that the ‘patient characteristics and TVS’ model shows good discriminative performance (AUC) and a reasonable performance on calibration, however it is comparable to the use of TVS-only. All three strategies based on the ‘patient characteristics and TVS’ model could be safely implemented in daily practice, i.e. without missing any additional (pre) cancers compared to TVS-only, which is the current daily practice.To choose which strategy is used best in clinical practice, one could focus on the availability and use of different diagnostic tests. In situations were no ultrasound is available, women
84