Research on Personalized NIPT Detection Timing and Anomaly Judgment Based on Bayesian Optimization

With rising obesity-related NIPT failures due to insufficient fetal DNA, we introduce a pioneering clinical application of Bayesian optimization—traditionally confined to engineering and machine learning—to personalized prenatal screening. Linear mixed-effects models quantified nonlinear relationships between fetal DNA concentration and gestational age, BMI, and maternal age. Gaussian mixture clustering stratified pregnancies into five BMI-specific subgroups. Uniquely applying Bayesian optimization to determine optimal 10–25-week sampling windows for each subgroup, we minimized weighted risks of early/late testing and low DNA concentration. Results reveal distinct optimal timing across BMI categories that diverge from current uniform guidelines, with higher BMI requiring earlier collection. This innovative integration of machine learning optimization into clinical obstetrics establishes a data-driven 'one person, one time' framework, representing the first utilization of Bayesian methods for individualized NIPT scheduling decisions.