Prenatal prediction model for gestational trophoblastic neoplasia after hydatidiform mole with a coexistent normal fetus: a retrospective cohort study

A new prediction model for GTN risk in HMCF shows promise but requires further validation due to small sample size and retrospective design.

Summary and findings

This study developed and internally validated a prenatal prediction model for gestational trophoblastic neoplasia (GTN) progression after hydatidiform mole with a coexistent normal fetus (HMCF). Among 40 women with HMCF, 16 developed GTN, with high peak serum hCG levels and large molar tissue volumes identified as independent predictors. The model stratified patients into low- and high-risk groups with differing GTN rates.

Abstract

<title>Abstract</title> <p> Background Hydatidiform mole with a coexistent normal fetus (HMCF) has a relatively high risk of developing gestational trophoblastic neoplasia (GTN), but no validated prediction model is currently available to estimate GTN risk after HMCF. For these reasons, we develop and internally validate a prenatal prediction model for GTN progression after HMCF. Methods This retrospective cohort study included women with HMCF confirmed by histopathological examination. Univariate and multivariable logistic regression analyses were used to identify independent prenatal predictors of GTN, which were then incorporated into a logistic regression–based nomogram. Model performance was assessed by stratified 5‑fold cross‑validation, AUC, and calibration. A risk classification system was derived from total nomogram scores. Results Among 337,790 pregnancies and 1,785 molar pregnancies during the study period, 40 women met the inclusion criteria for HMCF; 16 (40.0%) developed GTN. Compared with women who did not develop GTN, those who developed GTN had higher peak serum hCG levels and larger maximum molar tissue volumes on ultrasound (both <italic>P</italic>  < 0.05). Multivariate analyses revealed that both high peak serum hCG levels (≥ 107,602 IU/L; OR, 12.55; 95% CI, 1.04–1830.66; <italic>P</italic>  = 0.046) and large molar tissue volume (≥ 276.3 cm³; OR, 10.27; 95% CI, 2.34–57.35; <italic>P</italic>  = 0.002) were independent predictors of GTN progression. The logistic regression, AdaBoost, and SVM models showed comparable discrimination (AUCs of 0.773, 0.783, and 0.773, respectively), with good calibration. The final logistic regression–based nomogram stratified patients into low‑ and high‑risk groups using a cutoff score of 200 points: Twenty-five women (62.5%) were classified as low‑risk (predicted GTN probability 3.9–25.0%; observed rate 16.0% [4/25]), and 15 (37.5%) were classified as high-risk (predicted GTN probability 73.0%; observed rate 80.0% [12/15]). Conclusions A simple two‑parameter nomogram based on available prenatal data showed good predictive performance and stratified women with HMCF into low‑ and high‑risk groups with markedly different GTN rates. This model may aid early prenatal risk assessment and tailoring of post-molar surveillance intensity in women with HMCF. </p>

Key findings

• 40 women with HMCF included; 16 (40.0%) developed GTN.
• High peak serum hCG levels (≥ 107,602 IU/L) had OR 12.55, 95% CI 1.04–1830.66, p=0.046.
• Large molar tissue volume (≥ 276.3 cm³) had OR 10.27, 95% CI 2.34–57.35, p=0.002.
• AUCs for logistic regression, AdaBoost, and SVM models were 0.773, 0.783, and 0.773, respectively.
• Low-risk group had observed GTN rate of 16.0% (4/25); high-risk group had 80.0% (12/15).

Limitations

• small n=40
• retrospective design
• single-site data collection
• wide confidence intervals
• potential selection bias

Elsewhere in the HCG (Human Chorionic Gonadotropin) corpus