Predictors for the efficacy of methotrexate and mifepristone treatment in ectopic pregnancy

Abstract

Objective

To compare the efficacy of methotrexate (MTX) alone versus MTX plus mifepristone for stable ectopic pregnancy, identify predictors of treatment success, and explore outcome-prediction models.

Methods

We conducted a retrospective, two‑arm observational cohort of consecutive patients with stable ectopic pregnancy treated in routine care (MTX‑only n = 138; MTX+mifepristone n = 132). Treatment assignment was non‑randomized, reflecting clinician judgment and patient preference. Baseline characteristics and treatment details were recorded, and the primary outcome—resolution of the EP without surgery—was assessed. Predictive factors were analyzed via logistic regression with Hosmer–Lemeshow testing for calibration. We further evaluated advanced machine learning models to classify treatment success.

Results

Overall treatment success rates were 73.91% (MTX-only) and 78.79% (MTX plus mifepristone), with no statistically significant difference between groups (p = 0.38). Across the combined cohort, factors strongly predicting success included initial hCG <1500 mIU/mL (adjusted OR 2.70, p = 0.01), absence of fetal cardiac activity (adjusted OR 2.10, p = 0.04), lower gestational age at diagnosis (adjusted OR 0.82 per week, p = 0.02), and a ≥ 15% hCG decline by Day 4 (adjusted OR 2.40, p = 0.02). By location, success rates were similar in tubal and non‑tubal strata. The treatment × location interaction was not significant (p = 0.85), and tubal‑only results were unchanged. The multivariable logistic regression predicting treatment success vs failure showed good discrimination (AUC 0.82) and adequate calibration (Hosmer–Lemeshow p = 0.45). In exploratory benchmarking of algorithms trained to predict success vs failure on the same predictors, gradient boosting yielded the highest apparent discrimination (accuracy 84%; AUC 0.88) in this dataset.

Conclusions

MTX plus mifepristone did not significantly outperform MTX alone in resolving stable ectopic pregnancy. Key predictors of successful medical management included lower baseline hCG levels, earlier gestational age, and substantial early hCG declines. Advanced machine learning approaches may improve predictive accuracy, supporting more individualized treatment selection for ectopic pregnancy.