Validation of an online tool for early prediction of the failure-risk in gestational trophoblastic neoplasia patients treated with methotrexate

Purpose In a low-risk gestational trophoblastic neoplasia (GTN) treated with methotrexate (MTX), the modeled hCG (human chorionic gonadotropin) residual concentration (hCGres), calculated with NONMEM program® (NM) during the first 50 treatment days, is a predictor of MTX-resistance risk. This model was implemented with another algorithm on https://www.biomarker-kinetics.org/hCG . The objective was to confirm the validity of the website estimations with respect to NM. Methods The consistencies of modeled hCGres estimated by NM and by the website were assessed in a dataset of 60 fictive patients with simulated hCG profiles, as well as in an independent database of 531 actual patients. Moreover, the hCGres predictive values regarding MTX failure-risk were assessed. Results The values of hCGres obtained with both methods were highly consistent in the fictive patient and in the actual patient datasets: median relative prediction errors (RPE) were − 0.059 and 9.9 × 10 –7 , respectively. The ROC AUCs for predictions of MTX failure-risk were 0.90 (95% CI 0.87,0.93) with both NM and the website. The gradual association between increasing hCGres and the 2-year MTX failure-free survival was confirmed. Conclusion There is a high consistency of hCGres estimates obtained with the two methods. The website is meant to help clinicians in the interpretation of hCG decline curves of MTX-treated GTN patients. hCGres is now validated for more than 1690 patients in four independent datasets, and its recognition as an early predictor of MTX resistance for treatment adjustment and for the future studies should be considered.