We wish to welcome the study by Upson et al on prognostic biomarkers of endometrial hyperplasia resistance when treated with oral progestins. We agree with the authors that the prognosis of women with endometrial hyperplasia when treated with oral progestins is underresearched and clinicians cannot reliably identify the women at risk. Oral progestins are a popular therapeutic choice, but we have shown that levongorgestrel intrauterine system (LNG-IUS) achieves higher regression rates than oral progestins. The intrauterine progestin release is also associated with higher patient satisfaction resulting in higher compliance, which may also explain its better efficacy in treating endometrial hyperplasia compared with oral progestins. However, when it comes to identifying prognostic markers for response to progestin treatment of endometrial hyperplasia, patient compliance represents an important confounding factor. Accounting for this confounding is difficult, and it may dilute the prognostic accuracy of biomarkers used to predict response to treatment. Hence, we have preferred to assess a similar panel of biomarkers using the LNG-IUS in which the patient compliance is not an issue.

We also wish to add that the study design may affect the estimate of prognostic accuracy. The largest effect on the estimation of accuracy is generated by studies using cases and controls, as in the study by Upson et al, and this represents spectrum bias. It is described that this design can overestimate accuracy up to 3-fold. We have generated a similar hypothesis in which a panel of biomarkers was found to be promising, but now that we are testing this hypothesis in a large prospective cohort study, we believe our previous findings were overestimated because of the study design. We would suggest for the authors to test their hypothesis as well in a prospective cohort study to confirm more reliably the accuracy of their biomarkers.