VOLUME 18 (Supplement)

Modelling time to β-hCG remission in gestational trophoblastic disease: Comparative evaluation of Cox and parametric survival approaches

Background: Traditional time-to-event analysis is usually compromised in the context of rare diseases like gestational trophoblastic disease (GTD), due to inherent constraints posed by limited sample sizes and non-constant hazards.

Objective: The study compared various survival models in estimating time to beta-human chorionic gonadotropin (β-hCG) remission among GTD patients and identifying consistent clinical predictors of β-hCG remission.

Methods: A retrospective cohort study of 258 post-molar evacuation patients was conducted using information from a Philippine specialty group database. Time-to-remission served as the primary outcome, with hazard ratio estimates derived across Cox proportional hazards, Firth’s penalized Cox, Negative Exponential, Weibull, and Gompertz models. Model fit and diagnostics were also compared. Sensitivity analyses excluded the upper 5 to 10% of follow-up durations, while stratified Cox models used evacuation mode and histologic type as strata.

Results: Gravidity was a consistent predictor of shorter remission time while complete mole histology and high baseline β-hCG levels were associated with delayed remission. The Gompertz model produced clinically relevant GTD remission probabilities, yet the traditional Cox model yielded the highest C-index and robustness under sensitivity analyses.

Conclusion: Advanced survival methods can yield stable insights in small GTD datasets. The findings identified consistent key predictors of remission which can be accounted for in risk stratification. Parametric models like Gompertz may augment current management through individualized risk estimates with corresponding actions. The findings support the complementary value of looking at alternative modeling strategies and suggest the role of disease registries in improving validation and translation of such statistical models.