Author(s)

Shuocheng Li

Corresponding Author

Shuocheng Li

Abstract

The accuracy of non-invasive prenatal testing (NIPT) is susceptible to interference from factors such as gestational weeks (GW) and maternal body mass index (BMI), necessitating optimization of the testing timing and improvement of abnormal chromosome discrimination capabilities. Based on data from 515 male fetus samples and 382 female fetus samples, this study constructed a multi-stage coupled model system: Firstly, Spearman correlation and polynomial ridge regression were used to quantify the nonlinear relationship between fetal Y-chromosome concentration and GW/BMI. Further, a risk-constrained BMI dynamic grouping and timing decision framework was established by combining K-means clustering and survival analysis. Age, height, and other multi-features were introduced to extend the clustering model, evaluating error sensitivity. Finally, the Monte Carlo Dropout method was employed to enhance the robustness of female fetus chromosome abnormality discrimination. The results indicate that GW is the main explanatory variable for Y-chromosome concentration (explaining 55.1% of the variance), timing optimization based on BMI grouping significantly reduces clinical risk (timing shift <0.3 weeks under error perturbation), and the high BMI group is more sensitive to measurement errors (qualified rate decrease reached 8.25%). The discrimination model achieved a true positive rate (TPR) >80% while controlling the false positive rate (FPR) <15%. This study provides quantifiable decision support for clinical personalized NIPT timing selection and quality control.

Keywords

Non-invasive prenatal testing; Timing optimization; Clustering analysis; Risk assessment; Monte Carlo method