Optimization of Non-Invasive Prenatal Testing Timing and Quantification of Influencing Factors Based on Gradient Boosting Trees and Survival Analysisa

Abstract

To enhance the accuracy of non-invasive prenatal testing and safeguard maternal health, this study mathematically models and investigates the relationship between fetal Y chromosome concentration and maternal characteristics, aiming to provide optimal testing timing recommendations. First, exploratory data analysis was conducted to examine the correlation between fetal Y chromosome concentration and maternal gestational age, BMI, and age. The Shapiro-Wilk normality test confirmed non-normal distribution, prompting the use of Spearman's correlation coefficient. Results revealed a weak positive correlation between Y chromosome concentration and gestational week at testing, and weak negative correlations with maternal BMI and age. Subsequently, a gradient-boosted decision tree was employed to construct a relational model. The F-test confirmed that gestational week, maternal BMI, and age all exerted statistically significant effects on Y chromosome concentration. Model evaluation revealed comparable mean squared errors across training and testing datasets, indicating no significant overfitting. To determine the optimal NIPT timing based primarily on maternal BMI, K-means clustering divided pregnant women carrying male fetuses into three BMI clusters: Cluster 0, Cluster 1 , and Cluster 2. Kaplan-Meier survival analysis and log-rank tests confirmed that gestational age significantly influenced Y-chromosome concentration Cluster 1, and Cluster 2. Kaplan-Meier survival analysis and log-rank tests confirmed significant differences in the distribution of Y chromosome detection timepoints across clusters. Ultimately, by constructing a risk-cost function and optimizing its solution, the optimal NIPT testing time points for each cluster were determined: 12.8 weeks for Cluster 0, 13.0 weeks for Cluster 1, and 16.2 weeks for Cluster 2. Sensitivity analysis demonstrated the model's robustness to detection errors, providing scientific evidence for clinical NIPT timing decisions.