Machine learning-driven survival prediction in gestational trophoblastic neoplasms: a focus on PSTT and ETT prognosis
Sakhr Alshwayyat1, Zena Haddadin1, Mustafa Alshwayyat1
1Faculty of Medicine, Jordan University of Science & Technology, Irbid, Jordan.
Introduction:
The clinicopathological characteristics and prognosis of placental site trophoblastic tumor (PSTT) and epithelioid trophoblastic tumor (ETT) have not been well summarized. Consequently, we conducted the largest to date series of samples of both types and employed machine learning (ML) to assess treatment effectiveness and develop accurate prognostic models for patients with GTN. Gestational choriocarcinoma (GCC) was used as the control group to show the clinical features of PTSS and ETT.
Methods:
The Surveillance, Epidemiology, and End Results (SEER) database provided the data used for this study's analysis. To identify the prognostic variables, we conducted Cox regression analysis and constructed prognostic models using five ML algorithms to predict the 5-year survival. A validation method incorporating the area under the curve (AUC) of the receiver operating characteristic (ROC) curve was used to validate the accuracy and reliability of the ML models. We also investigated the role of multiple therapeutic options using the Kaplan-Meier survival analysis.
Results:
The study population comprised 725 patients. Among them, 139 patients had ETT, 107 had PSTT, and 479 had GCC. There were no significant differences in survival between the different tumor groups. Multivariate Cox regression analysis revealed that metastasis was a significant prognostic factor for GCC, while older age and radiotherapy were significant prognostic factors for PTSS and ETT. ML models revealed that the Gradient Boosting classifier accurately predicted the outcomes, followed by the random forest classifier, K-Nearest Neighbors, Logistic Regression, and multilayer perceptron models. The most significant contributing factors were tumor size, year of diagnosis, age, and race.
Discussion:
Our study provides a method for treatment and prognostic assessment of patients with GTN. The ML we developed can be used as a convenient individualized tool to facilitate clinical decision making.
