Research On an Optimal Nipt Timing Decision Model Driven by Multidimensional Factors Based on Machine Learning

Abstract

The accuracy of non-invasive prenatal testing (NIPT) is highly dependent on selecting the optimal testing time, which varies due to individual differences among pregnant women. To address the limitations of the traditional “one-size-fits-all” testing approach, this paper constructs an intelligent model framework that integrates prediction, stratification, and decision-making. First, the framework employs a CatBoost regression model to accurately predict the key indicator of fetal cell-free DNA (cffDNA)—Y chromosome concentration—based on multidimensional physiological indicators of pregnant women, achieving a coefficient of determination (R²) of 0.717 on the test set. Next, a K-means clustering algorithm is used to achieve data-driven refined stratification of the study population, and principal component analysis (PCA) is applied to process high-dimensional features. Building on this, the study innovatively introduces a Cox proportional hazards model to construct a risk function and establishes an optimization model to calculate personalized optimal NIPT timing for different stratified groups of pregnant women. For example, in the comprehensive multi-factor model, the optimal testing times determined for the low-BMI and high-BMI groups were 11.997 weeks and 10.420 weeks, respectively. In addition, to address the issue of detecting female fetal chromosomal aneuploidy, a decision tree classification model was developed, achieving an average accuracy of 0.886 with five-fold cross-validation. The core contribution of this study lies in the deep integration of machine learning prediction models with operational optimization and decision models, achieving a paradigm shift from “passive prediction” to “active decision-making,” and providing a scientific, efficient, and individualized decision support solution for clinical NIPT practice.