Predicting live birth for poor ovarian responders: the PROsPeR concept

A number of live-birth predictive models are available, and despite clinical interest these are rarely used owing to poor performance. In addition, no predictive models specifically for poor ovarian responders (POR) are available. The aim of the current project was to develop a clinically applicable tool for predicting live birth for PORs receiving recombinant human FSH [r-hFSH].

A model was developed to predict live birth in PORs receiving r-hFSH, using data from the ESPART trial. Initially, two models were developed separately: one for patients with data from a previous assisted reproductive technology (ART) cycle and one for ART treatment-naïve patients. Subsequently, the simplified Poor Responder Outcome Prediction (PROsPeR) concept was derived.

PROsPeR considers three predictors and categorizes PORs into three scores, with predicted the live-birth rate divided by three with each worsening category. When adequately calibrated, a discrimination score up to area under the receiver operating characteristic (AUC_ROC) (95% CI) of 0.84 (0.79 to 0.88) was observed, which is superior to previously published models. Lower discriminations were observed when the PROsPeR model was used to evaluate the patients who received both r-hFSH and recombinant human LH in the ESPART study (AUC_ROC [95% CI] 0.66 [0.61 to 0.71]) and when all the patients included in the ESPART study were evaluated (AUC_ROC [95% CI] 0.68 [0.61 to 0.72]).

This model, specific to PORs receiving r-hFSH, constitutes the best compromise between precision and simplicity, and is suitable for routine practice.