Author(s)

Yibo Peng, Wei Li, Linhua Wang

Corresponding Author

Yibo Peng

Abstract

Non-invasive prenatal testing (NIPT), a genetic screening technology based on cell-free fetal DNA (cfDNA) in maternal peripheral blood, has become a crucial approach for the early detection of fetal chromosomal abnormalities. However, its detection accuracy and optimal testing window are influenced by multiple factors, including gestational age, maternal body mass index (BMI), and sequencing quality. To address the mismatch between unified testing schedules and individual physiological differences in existing clinical workflows, this study proposes an adaptive decision-making model for NIPT timing based on BMI stratification and joint risk optimisation. A total of 605 clinical samples were processed through data standardisation, missing-value imputation, and categorical feature encoding. Spearman correlation and random forest regression were employed to analyse the multidimensional relationships between Y-chromosome concentration, gestational age, and BMI. Results showed that Y-chromosome concentration was significantly positively correlated with gestational age (ρ = 0.923) and negatively correlated with BMI (ρ = –0.855). Based on these findings, K-means clustering was applied to stratify maternal BMI into three groups (20–31, 31–36, and 36–46), followed by the construction of a time–performance bi-objective risk model to determine the optimal testing window. The model indicated that the optimal gestational weeks for testing were 13.9, 14.2, and 13.6 weeks for the respective BMI groups, with a stable interval concentrated around 13–14 weeks, validating the robustness and physiological consistency of the proposed framework. Furthermore, for female-fetus samples, an XGBoost-based anomaly discrimination model achieved Precision = 0.96, Recall = 0.87, and F1-score = 0.91 on the test set. SHAP based interpretability analysis identified Chr13_GC_Content, Chr21_GC_Content, and Maternal_BMI as the major contributing features. The results demonstrate that the proposed stratified optimisation model effectively enhances individualised timing recommendations for NIPT and provides interpretable bioinformatics insights for abnormal sample identification.

Keywords

Non-invasive prenatal testing (NIPT); Risk optimization; K-means clustering; XGBoost; SHAP interpretability model