A Hybrid Modeling Framework for NIPT: Integrating Discrete-Time Survival Analysis and XGBoost for Enhanced Prenatal Screening

Abstract

This study combines discrete-time survival analysis regression trees with machine learning algorithms, specifically logistic regression and XGBoost, to develop an integrated predictive framework aimed at optimizing the clinical application of Non-Invasive Prenatal Testing (NIPT). The research demonstrates that this hybrid approach effectively models the time-to-threshold progression of Y-chromosome concentration in male fetuses and improves the classification accuracy of fetal abnormalities in female fetuses by leveraging key biomarkers such as chromosome Z-scores and maternal physiological parameters. The incorporation of SHAP values enhances the interpretability of the model, providing transparent explanations for predictions. Sensitivity analysis confirms the robustness of the model under varying parameter settings, supporting its potential for personalized prenatal screening protocols and improved clinical decision-making.