Development of a Machine Learning‑Based Prognostic Model for Intermediate Trophoblastic Tumors: A Single-Center Study With Web-Based Tool Implementation.
retrospective_cohort · Level III
Where this comes from
- Record sourced from PubMed, PMID 42284544.
- Also identified by DOI 10.1200/PO-25-00881.
- No licence information is recorded for this record.
- Because redistribution is not established, this page shows the abstract only. Follow the links below for the full text.
Abstract
Current prognostic systems are inadequate for intermediate trophoblastic tumors (ITTs). The aim of this study was to develop a machine learning (ML)-based model to predict progression-free survival (PFS) in patients with ITT and implement a web-based tool for individualized risk stratification. We analyzed a retrospective cohort of 236 patients with ITT treated at a national tertiary center between 2000 and 2024. A multimodal feature selection strategy-integrating Cox regression, LASSO, Gradient Boosting Machine (GBM), and Random Survival Forest (RSF)-was used to identify robust predictors from clinicopathologic and inflammatory variables. The final prognostic model was constructed using the RSF approach. Model performance was evaluated through a rigorous nested 5-fold cross-validation framework to prevent data leakage. Five key predictors were identified: International Federation of Gynecology and Obstetrics stage, interval from antecedent pregnancy, Ki-67 index, neutrophil-to-lymphocyte ratio, and systemic immune-inflammation index. The RSF model demonstrated robust discrimination with a cross-validated concordance index of 0.816 (95% CI, 0.721 to 0.895) and excellent calibration (Integrated Brier Score, 0.113). Decision-curve analysis confirmed clinical utility within the relevant 20%-60% threshold range. An interactive web tool (Shinyapps) was deployed to generate real-time individualized PFS predictions with 95% confidence intervals. To our knowledge, this study presents the first ML-based prognostic model specifically for ITT. By integrating immune-inflammatory markers with traditional clinicopathologic features, the RSF model offers superior risk stratification compared with anatomic staging or alternative models. The developed online tool serves as a proof-of-concept prototype to facilitate future external validation and research on personalized clinical decision making for this rare malignancy.