Author(s)

Mengyuan Wang, Jiajing Wang, Beibei Dong

Corresponding Author

Mengyuan Wang

Abstract

As a key early screening technique for fetal chromosomal abnormalities, the accuracy and timing of Non-Invasive Prenatal Testing (NIPT) heavily depend on fetal cell-free DNA concentration. This study, based on NIPT clinical data from pregnant women with high BMI in a specific region, developed an integrated solution following a "data preprocessing - statistical modeling - algorithm optimization - result validation" framework to address three core objectives: modeling male fetal Y-chromosome concentration, optimizing testing timing, and determining female fetal aneuploidy. In the first step, after data cleaning (removing samples with abnormal GC content, low read alignment rates, or missing key indicators) and standardization, statistical tests confirmed non-normal distribution of variables. Spearman rank correlation analysis revealed that Y-chromosome concentration was significantly positively correlated with gestational age and negatively correlated with maternal BMI, independent of age, height, and weight. After grouping by BMI, cubic polynomial regression models were constructed to quantify this relationship. With coefficients of determination (R²) ≥ 0.82 and Root Mean Square Error (RMSE) ≤ 0.03, validated by t-tests and F-tests, these models reliably described the variation of Y-chromosome concentration with gestational age across different BMI levels. The second step established an optimization model for BMI grouping and ideal test timing. Using Particle Swarm Optimization to minimize the earliest time to reach a target probability under specific constraints, five optimal BMI intervals were identified. The recommended testing windows for these groups ranged from 11.2 to 14.8 weeks, demonstrating more personalized timing than the conventional uniform 12-week approach. Sensitivity analysis showed an average decrease of 12% in the target achievement probability when simulated test error increased by 50%, confirming model stability. The third step expanded the model to a comprehensive five-variable optimization, incorporating maternal age, height, weight, BMI, and gestational age. A five-variable quadratic polynomial model improved prediction accuracy by approximately 15% compared to the single-BMI model. The constrained optimization based on this model further refined the recommended testing times by 0.5–1.2 weeks, highlighting the advantage of multifactorial modeling for personalized timing optimization.

Keywords

Polynomial regression; Particle swarm optimization algorithm; Greedy algorithm; Spearman correlation analysis