Validation of the predictive value of modeled HCG residual production “P” in low-risk gestational trophoblastic neoplasia (GTN) patients treated in GOG-174 phase III trial.

Abstract

5110 Background: In low-risk GTN, chemotherapy is changed according to serum hCG levels. We previously showed mathematical modeling of hCG kinetics provides a parameter production “P”, a useful early predictor of methotrexate (MTX) resistance (You et al; Ann Oncol 2010; ASCO and ISSTD 2011). We applied this approach to patient cohort of GOG-174 trial, in which weekly MTX (Arm 1) was compared to dactinomycin (Arm 2). Methods: Database (210 patients, including 78 with resistance) was split into 2 sets. A 126 patient Model Set was initially used to adjust model parameters. Patient hCG kinetics from day7 to day50 were fit with NONMEM™ program to: “[hCG(time)] = hCG0i * exp(–k*time) + P”; where P is residual hCG tumor production, hCG0i is the initial hCG level, and k is the rate constant. Three putative P-based classifiers of resistance were assessed using ROC analyses. Then an 84 patient Test Set with blinded-resistance status was used to assess the validity of predictions. The primary endpoint was treatment resistance defined as relapse and/or lack of hCG normalization. Results: Due to initial surge in 14% patients, hCG kinetic modeling was started on day7. Individual hCG decline profiles of Model Set patients were modeled. There was no impact of treatment arm on variability of kinetic parameter estimates. The best P cut-offs to discriminate resistant vs. sensitive patients were 7.7 in Arm 1 and 74.0 in Arm 2. They were combined to define 2 predictive groups with low vs. high risks of resistance (ROC AUC = 0.82; Se = 93.8%; Sp = 70.5%). The model was then applied to Test Set patient cohort. The predictive value of P-based predictive groups regarding resistance was reproducible (ROC AUC = 0.81; Se = 88.9% (95% CI: 70.8%-97.7%); Sp = 73.1% (95% CI: 60.0%-84.4%)). Both P and treatment arm were associated with resistance using multivariate logistic regression tests. Predictive value of P was less accurate in dactinomycin arm. Conclusions: The early predictive value of the modeled kinetic parameter P regarding resistance appears promising in GOG-174 study, especially in MTX arm. This is the second positive evaluation of this procedure study. Prospective validation is warranted.